9 replies [Last post]
agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff

Hi Foldit players! Here are the Office Hours transcripts: They begin with the first one but you can find the latest one HERE

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office Hours 5/15- led by Beta_Helix

[10:00 AM] beta_helix: Hi everyone. I'll be here for the next hour, with the first (hopefully of many) Foldit Office Hours!
[10:01 AM] beta_helix: For anyone who doesn't know me. I'm here: http://www.cis.umassd.edu/~fkhatib/cv.html(edited)
[10:01 AM] beta_helix: and I'll be happy to chat about anything you like!
[10:02 AM] skippysk8s: hey beta helix....I think the theme of the day is hot cross buns,,,maddening to try to cover them
[10:03 AM] beta_helix: Oh yeah... those are a pain in the BUNS
[10:04 AM] beta_helix: But unfortunately, they are extremely important in design
[10:05 AM] skippysk8s: we assume so.. it is nice to see that all that work on monomer design being applied. What has surprised you about what is turning out to be useful in the last couple months
[10:07 AM] beta_helix: Honestly... it hasn't been a surprise to me. After the amazing results all of you showed the world last year with a Nature paper full of novel designs that folded up just like you predicted in the game... this is the logical next step!
[10:08 AM] beta_helix: What makes this new task harder, of course, is that we are no longer just asking you to make sure your novel design can fold on its own...
[10:10 AM] beta_helix: That is the next difficult challenge: making sure it folds up properly AND binds to the COVID-19 spike protein.
[10:10 AM] Susume2: seems like in theory we could take soem of those stable folds and fit them into the target groove
[10:11 AM] skippysk8s: is there any way we could load some of the prior proven folds into the current puzzles and try to fit them? I'm the want my Lego folder
[10:11 AM] beta_helix: That's a great point, Susume... the only main difference would be ensuring that the side facing covid can bind to the spike (as opposed to being exposed to water).
[10:12 AM] Susume2: yes, definitely need to mutate and adjust (and repeat ad nauseam)
[10:12 AM] skippysk8s: agreed-- I'm looking for less lysine and arg on at least one edge so it will fit snugly
[10:13 AM] beta_helix: That might be a series of fun puzzles, though! Everyone starts from the same Foldit Nature design, and tries to fix the binding sides
[10:13 AM] beta_helix: (I'm writing this down!)
[10:13 AM] skippysk8s: sort of like our contest... fit the block into the space
[10:14 AM] beta_helix: We always say: key in the lock!
[10:14 AM] Susume2: is it possible to evaluate shape complementarity based on just the backbone, without knowing which sidechains will be selected (well we know the ones on the virus)?
[10:15 AM] skippysk8s: does the tumbler have to be perfect? our team got some but never all of the potential bonds to meet along the edges. And can we get a filter to make the score improve the better the fit (2 questions)
[10:16 AM] beta_helix: Susume, you can get a very basic idea... but you would still want to represent the sidechains somehow, you wouldn't want to just strip them off.
[10:16 AM] Susume2: maybe make them all generic medium sized centroids
[10:16 AM] beta_helix: Many programs use centroids for example (and I know that word gives you the chills if you were around when we tried centroid mode in Foldit!)
[10:17 AM] beta_helix: exactly
[10:17 AM] Susume2: I'm just wondering if there is a way you guys could preselect a foldit design from the nature paper that will likely fit if we pick good sidechains
[10:18 AM] beta_helix: Yes, I think that would be the way to do, because some of the topologies will just fit better than others.
[10:18 AM] Susume2: auntdeen and I had lots of fun breaking centroids as a team project ;)
[10:18 AM] beta_helix: hahahahhahahahhahahaha
[10:19 AM] beta_helix: Skippysk8 for question 2, that could be possible... depending on how fast those calculations would end up being. It might just be easier to screen the folds initially on our end. But indeed, I can see how that would be a useful filter!
[10:20 AM] Susume2: as long as we know the filter takes x minutes to run, we can plan how often to apply it
[10:20 AM] skippysk8s: we were frustrated as a team with Bruno's fold... we felt like we could go for score but didn't believe the result would be useful for science
[10:21 AM] skippysk8s: many of us started trying to link those few unmet bonds, but ended up just going for score
[10:22 AM] beta_helix: Yeah, I think it would be very useful when you are trying to figure out if your topology is going to fit or if it's a waste of time. (I'll ask the team)
[10:22 AM] beta_helix: Indeed Skippysk8, I know how frustrating that is...
[10:22 AM] beta_helix: I remember the giant pi-helices from back in the day
[10:23 AM] skippysk8s: well, it can be fun lol...
[10:23 AM] beta_helix: I kept looking at those solutions and going "why are these scoring so high!?!!"
[10:23 AM] beta_helix: But that really highlights the amazing work bkoep and the rest of the Foldit Team has done with Foldit scoring in design
[10:24 AM] beta_helix: It's such an amazing journey to see the difference in high scoring monomer designs in the early days of Foldit, compared to the Nature paper.
[10:24 AM] skippysk8s: we love to game. we want to try to keep the game fun and at the same time useful
[10:25 AM] beta_helix: That is the reason for BUNS, h-bond filters, and the new filters for these coronavirus puzzles.
[10:25 AM] beta_helix: Exactly, Skippysk8... it is a fine balance, right?
[10:25 AM] skippysk8s: agreed
[10:25 AM] RockOn: It seems my PC memory usage for FoldIt is at 1,000 MB's!
[10:26 AM] beta_helix: We could make the game painful accurate (like an MD simulation with waters) but I promise you it would not be much fun!
[10:26 AM] RockOn: What has changed Windows or FoldIt
[10:26 AM] beta_helix: Rock On!
[10:26 AM] beta_helix: oh, that's a bad thing... right. My bad
[10:26 AM] RockOn: How bad?
[10:26 AM] beta_helix: It's always easy to blame Windows (and fun too!) but indeed Foldit is constantly changing.
[10:27 AM] beta_helix: Have you noticed a difference between Foldit updates?
[10:27 AM] RockOn: 33% of the CPU
[10:27 AM] Susume2: RockOn do you mean RAM footprint? or something else?
[10:27 AM] RockOn: Yes, running taskmgr
[10:28 AM] RockOn: Just 1 client and nothing else (app related)
[10:28 AM] beta_helix: Did you have this problem with previous versions of Foldit?
[10:28 AM] RockOn: No, not unless 2 clients wee running for hours
[10:29 AM] RockOn: One crash today running helix twister
[10:30 AM] skippysk8s: my old computer was at 100% with one client -- ran alternating clients with 2 open... it was about 12 years old. the biggest problem is that all the "ligand" cells count towards puzzle size, so it is getting harder with small computers
[10:30 AM] beta_helix: Hmm... I quickly scanned our feedback and don't see any other Windows reports. https://fold.it/portal/feedback Was there a recent Windows update that was installed?
[10:30 AM] skippysk8s: it couldn't do today's puzzles at all
[10:30 AM] Susume2: devprev got pushed to main about a week ago
[10:31 AM] beta_helix: Skippysk8, you mean 1838?
[10:31 AM] Susume2: I suspect one thing that pushes RAM per client up is using all the quicksave slots - some scripts use very many of them
[10:32 AM] skippysk8s: I couldn't do any of the COVID puzzle on it
[10:32 AM] skippysk8s: puzzles
[10:32 AM] beta_helix: Undo graph as well, right Susume?
[10:32 AM] Susume2: yes probably
[10:34 AM] Susume2: changing tracks and changing back should free the undo graph memory, idk that frees the quicksave slots as well (have not tried it)
[10:34 AM] RockOn: I see the graph at 100 for length AND Memory Usage is at 100%
[10:34 AM] beta_helix: I'll bring these issues up to the rest of the team...
[10:35 AM] skippysk8s: I live at 25 undos RockOn.. not ideal. I just save before I start a new recipe each time
[10:36 AM] RockOn: I set Memory Usage to 25% and Taskmgr shows mem down to 700 or so
[10:36 AM] skippysk8s: anyhow, it might help our smaller computer owners if the ligand shows everything, but only "counts" the cells near the binding area
[10:36 AM] beta_helix: Yes, that is a good point.
[10:37 AM] RockOn: CPU still bouncing around from 29% to 36%
[10:37 AM] Susume2: I keep undo graph at 85 and undo mem 75%, and 800MB is a common RAM footprint for me
[10:37 AM] beta_helix: I've often wanted Foldit to only visually show the very far off frozen sections of spike (for example) and have Rosetta just ignore those atoms entirely.
[10:38 AM] RockOn: I'm still getting No Response from time to time
[10:38 AM] skippysk8s: it does help to try and fit the key into the lock. For example, JoannaH found a great 3 helix log with a spike that got an extra hook. it was pretty cool
[10:38 AM] beta_helix: It turns out that this is non-trivial to do.
[10:39 AM] beta_helix: Indeed, Skippysk8, I just meant the regions that are so far away they would have no interaction at all.
[10:39 AM] skippysk8s: we appreciate that...
[10:39 AM] skippysk8s: what can we do better?
[10:40 AM] skippysk8s: I'd be fine if one of my bad folds was presented as "what not to do"
[10:40 AM] beta_helix: That is a tough question, because you are all doing great! Any deficiencies are on Rosetta and our end!
[10:41 AM] beta_helix: Skippysk8, I think bkoep does that very well in his Lab Reports: showing what works and what doesn't.
[10:41 AM] Susume2: I loved the TIM barrel challenge - if you guys ever see a cool player fold and want us to try more of that shape, you could play it that way - a suggested but completely optional shape
[10:41 AM] skippysk8s: I know you are all working flat out, but feedback on the good and bad is highly appreciated. Lexie was great -- maybe she has some things she'd like us to try as a challenge
[10:41 AM] beta_helix: It really is a question of us finding the balance between filters that are scientifically accurate... but don't kill your game. (h memory/speed/fun-wise)
[10:42 AM] beta_helix: Those are h great ideas! I'm adding them to the novel of excellent suggestions today
[10:42 AM] RockOn: I'm doing a reboot to see if things change...will report in a few mins.
[10:43 AM] beta_helix: Thanks RockOn!
[10:45 AM] Susume2: in fact a slightly perturbed foldit player design could be an alternative to revisits for beginner-friendly puzzles - I know you need some ongoing data from revisits, but maybe alternate with player designs - pull them apart a little like you do the revisits
[10:47 AM] RockOn: I tried Susume2's idea track change and CPU went down and Ram use down 50%
[10:47 AM] Susume2: it would let us all play hands on with some designs from other people/teams, but in a category where you're not super hungry for diversity
[10:47 AM] skippysk8s: beta-helix do thank the team for getting a beginner puzzle up on covid... maybe we can give them a few things to dock and perhaps one to try and work on shape conformance.... It is good to have the change to expand our folding community
[10:47 AM] Susume2: great!
[10:48 AM] beta_helix: Good call, Susume. We'd just have to mess them up enough so that they wouldn't just go back down the their global mins
[10:48 AM] RockOn: So no reboot right now
[10:48 AM] skippysk8s: yes -- please bring up all hands.... or at least let all hands be a round 2 if we can't share across groups. I worked a couple puzzles with players not on my team. It would be good to share
[10:48 AM] beta_helix: Will do, Skippysk8!
[10:49 AM] beta_helix: All Hands in the past didn't really work well, but that was before design. I think it would be worth just trying it as a new puzzle (rather than the All Hands mechanic we had)
[10:49 AM] alcor29: Is there anything that can be done to speed up the feedback cycle. It is frustrating to be on round 9 of the C19 not knowing what seemed to work better than others.
[10:50 AM] skippysk8s: I understand that first round in design should get as many ideas as possible.... but then working together would help
[10:50 AM] beta_helix: alcor29.. we feel you on that one! Every Foldit meeting we are begging for feedback on previous puzzles.
[10:50 AM] beta_helix: But science is slow, and covid-19 has slowed things down a lot more
[10:51 AM] beta_helix: Luckily, things seem to have finally been sorted out... so fingers crossed that we'll get results soon!
[10:52 AM] beta_helix: Skippysk8, I completely agree with you.
[10:52 AM] beta_helix: The key is to make sure that any starting puzzles aren't so trapped in their local minimum that there is no way to improve the score... that has often been the issue with starting from high-scoring structures.
[10:53 AM] skippysk8s: maybe pull it away and make us redock lol.... we are getting to be good at that
[10:53 AM] beta_helix:hahahahahhahahaa
[10:56 AM] RockOn: Sure wish it were a 2-way chat that is going on!
[10:57 AM] alcor29: Has foldit ever tried the cooperative model instead of the competitive gamesmanship model?
[10:57 AM] beta_helix: alcor29, w.r.t what works better than others... that has been one of the main elements of the Lab Reports
[10:57 AM] beta_helix: We participated in WeFold
[10:58 AM] beta_helix: We did, though, alcor29 https://fold.it/portal/node/986517
We messed up the All-Hands puzzle yet again!
[10:58 AM] beta_helix: but it did not go particularly well.
[10:59 AM] beta_helix: We have somewhat resorted to using the "Share with Scientists" feature, in order to get models that might not be the top scoring ones.
[11:00 AM] beta_helix: RockOn, do you not see my replies to you?
[11:01 AM] alcor29: There are many smart and capable people here but I've always wondered if they only concentrated of the results instead of the score and the group competition what the results would be. Not as a partial experiment but as a total citizen science project.
[11:02 AM] Dhalion: I have never share a low scoring design
[11:02 AM] skippysk8s: ah, defintely do... I forgot to max out segments about a year ago and got a beautiful pringles fold. It scored 5th in the contest
[11:03 AM] beta_helix: It's a valid concern, and a tough balance. I honestly feel that if we had started Foldit as just a Citizen Science project, where you just participate without rankings, leaderboards, etc... we would have still gotten a lot of promising results. But I think the human competitive element adds a lot to the motivation and to the results. There is a figure in our first Nature paper's Supp Section that shows this... let me try to find it
[11:04 AM] alcor29: When only a partial experiment, game players still retain their 'secrets' and certain 'questionable' behaviors. But I get you.
[11:04 AM] beta_helix: Nature 2009
[11:05 AM] beta_helix: So this ugly figure shows the lowest Rosetta energy PER TEAM over time (in days)
[11:05 AM] alcor29: Interesting.
[11:05 AM] beta_helix: You can see that many teams flattened out (hit their high score) around days 2-3
[11:05 AM] Susume2: (lower is better, right?)
[11:05 AM] beta_helix: ...except for the cyan team! (Yes, lowest energy = high Foldit score. Thanks!)
[11:06 AM] beta_helix: Once the Cyan team suddenly was at the top of the leaderboard, bam! Everyone started working overtime
[11:06 AM] beta_helix: (shown as the yellow rectangle)
[11:06 AM] alcor29: So maybe only 2-3 days should be allowed to avoid overrefining?
[11:06 AM] beta_helix: We have indeed discussed this very often...
[11:06 AM] beta_helix: but not everyone can play Foldit 24/7
[11:07 AM] eaglgenes101: At 2-3 days, I'd stop polishing existing solutions and start trying other possible avenues
[11:07 AM] eaglgenes101 Which may or may not be effective
[11:07 AM] beta_helix: The 7 day window (or longer in certain cases, like the Monkey Virus one) is to ensure that everyone has the opportunity to play at least a few days.
[11:07 AM] beta_helix: eaglgenes101, that is exactly what you want to do!
[11:07 AM] skippysk8s: actually, our team enjoys the end game...we cross from the Ural mountains through the US with an occasional Australian. So we take the last few days for fun
[11:08 AM] alcor29: Good discussion. Many thanks beta_helix.
[11:08 AM] beta_helix: That was actually the motivation for our short-lived Exploration puzzles, and the Move Limit.
[11:08 AM] alcor29: Hope you do it again sometime.
[11:08 AM] skippysk8s: thanks beta helix
[11:08 AM] beta_helix: To force you to try a completely different fold, because it's very likely that your top score is stuck in it's local minimum.
[11:08 AM] beta_helix: Me too! Thank you all!
[11:09 AM] skippysk8s: well, just leave us some play time lol
[11:09 AM] Susume2: thanks beta!
[11:10 AM] beta_helix: Thank YOU are for your great suggestions, I'll bring them to Monday's Foldit Team meeting! Take care everyone, and keep up the great folding...

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office Hours 5/26- led by Horowsah

[4:00 PM] horowsah: Hey all- just wanted to say that my "office hour" today is officially starting. This means that I can try to answer questions for the next hour. A couple disclaimers- I didn't put together the puzzles that are currently up so I'm not super familiar with their details, and I'm not a COVID-19 expert. My expertise is more biochemistry in general.
[4:01 PM] Formula350 Hello :)
[4:01 PM] Jeff101: hi
nspc hi !
[4:02 PM] ipatrol: awwww
[4:02 PM] Jeff101: I've got a question about b-strands. Are the ideal zigzag structures stable if there are no hydrogen bonds?
[4:02 PM] Jeff101: like what do b-strands look like before the hydrogen bonds form?
[4:03 PM] horowsah: let's see if this answers the question- b-strands are the most extended form possible of a chain of residues
[4:03 PM] horowsah: so when it forms h-bonds to form a sheet, it's basically the same shape for each strand before and after
[4:03 PM] horowsah: just now with h-bonding
[4:04 PM] horowsah: in real life, that can happen, but it will be very dynamic and moveable
[4:04 PM] horowsah: there's a class of proteins called Intrinsically Disordered Proteins that are kind of like that
[4:04 PM] horowsah: is that answering the question?
[4:05 PM] Jeff101: when the b-strands form h-bonds, do they start near the b-hairpin bends or farther from the bends?
[4:05 PM] ipatrol: Is there any way to ensure that puzzle solutions will be specific to the target and not just bind to other common protein motifs?
[4:06 PM] horowsah: @Jeff101 hmm- that's a question that I don't know a firm answer to, but I'm guessing it's not known for most proteins. But my thought would be that it would in most cases likely start from the loop and move outward
[4:08 PM] Jeff101: I've read some about denaturation studies. Seems there are many different ways to denature sheets (chemicals, temperature, pulling on the ends).
[4:09 PM] horowsah: @ipatrol: this is the big challenge of foldit in general. To have lots of different puzzles with different facets, it becomes hard to make it as specific for each case as we want. This is definitely an active area of development for the types of COVID puzzles we've been running. My thought is that the work Foldit players are putting in is great, but that there's definitely room for improvement to make this aspect work better from the game side.
[4:09 PM] ipatrol: A strong oxidizer will denature darn near anything
[4:09 PM] horowsah: Yeah, denaturation is a big deal in protein folding studies
[4:09 PM] horowsah: I've used at least four methods of doing it in my lab work, and there's tons more
[4:10 PM] Jeff101: do they all show the same intermediates?
[4:10 PM] ipatrol: I think urea or a gunaidinium salt is usually the least destructive way of denatuging proteins at room temperature
[4:10 PM] ipatrol: *denaturing
[4:11 PM] wboler: Would it be too difficult to add filters for protein motifs we know won't work out?
[4:11 PM] horowsah:that's a good question- in general, this is a point of discussion as to whether all forms of denaturation are created equal, or even physiologically relevant
[4:11 PM] Jeff101: which way agrees best with molecular dynamics calculations?
[4:11 PM] ipatrol: Well I mean, if it'll kill you, it's probably not very physiologically relevant :-P
[4:12 PM] horowsah: that depends who you ask For example, oxidation is a common protein unfolder in cells. So that one's important. Does it show the same intermediates as temperature denaturation? Probably yes in a lot of cases, but not all.
[4:13 PM] horowsah: @wboler: i know there's been a bit of discussion on that idea, but it's practically really difficult because there's so many different ways to do something that isn't great
[4:14 PM] ipatrol: Hmm, I wonder, would it be possible to modify the game to allow us to play with some non-standard amino acid as sidechains?
[4:15 PM] Jeff101: like phosphorylated ones
[4:15 PM] horowsah: to the urea or guanadinium question, these are the most common ways to denature proteins in a lab environment without changing temp. The reason is because it's the easiest to reverse and it usually won't chemically modify the protein outside of unfolding it (and maybe causing it to aggregate)
[4:15 PM] Formula350 --- Poster child for "How to Do Stuff in a Not Great Way" :P
[4:15 PM] wboler: Would it be possible to train a neural network somehow for things that don't work? My guess is one of the limitations would be available data. Maybe some other AI mechanism.
[4:16 PM] ipatrol I'm afraid to ask, Formula350
[4:16 PM] Jeff101: what experimental technique is best for monitoring hydrogen-bonding between b-strands?
[4:17 PM] horowsah: non-canonical amino acids! In theory, they do work in Foldit, but we don't give puzzles on them hardly ever. I'm not sure if there's a specific reason why, just the science hasn't trended that way yet. The one thing is that if we're trying to design something, it's harder to put non-standard amino acids in proteins than the regular ones in the lab, so it usually leads to things that will be harder to implement outside a computer
[4:17 PM] nspc this is high level questions yes oO
[4:18 PM] Formula350 I don't follow what's going on either, don't worry :P
[4:18 PM] horowsah: @wboler on the AI- there have been a few attempts at similar things, but they've not gotten too far. It seems to me like the way Foldit players do things is especially hard for a computer to learn
[4:19 PM] jmbrownlee333 If I could ask a completely unrelated question, are there plans for small molecule design and coronavirus? I am thinking about a viral protease target, most specifically.
[4:19 PM] horowsah: @Jeff101: best technique for monitoring h-bonds: if you want a static picture of it folded, then usually crystallography, but NMR and cryo-EM do a good job. If you want to see it wiggle and jiggle in solution, then NMR is usually the way to go.
[4:20 PM] horowsah: small molecules! I'm sure that's on the minds of the people working on that aspect of Foldit long-term (because it's on everyone's minds right now), but I really don't know if it's something on the horizon there or not.
[4:20 PM] jmbrownlee333 it has been ages since we had a small molecule puzzle.
[4:20 PM] serca How do I find a protein with unknown tertiary structure (not listed in pdb) that has known environment (so I wouldn't try to fold a transmembrane protein with Foldit)? Is it really hard to find one?
[4:21 PM] horowsah: @Serca: maybe uniprot.org?
[4:22 PM] Jeff101: @horowsah what projects do you work on? that might help us ask better questions
[4:22 PM] horowsah: @jmbrownlee333: small molecules present a whole different scientific set of problems than regular foldit puzzles, so it's hard to get it just right h from a scientific game development side, and from the perspective of letting players know what is the best way to tackle those problems
[4:23 PM] horowsah: so it's something that is still under heavy development
[4:23 PM] serca never heard about that site, thank you
[4:24 PM] horowsah: what projects do i work on? Lots of things...
[4:24 PM] horowsah: within foldit, i tend to work mostly on educational tools and integrating crystallography and cryo-EM tools
[4:24 PM] horowsah: but I have a wet lab that studies how nucleic acids impact protein folding and aggregation, which I'd love to do in Foldit someday
[4:26 PM] horowsah: @sarca: uniprot is my go-to whenever i hear about a protein that i need to learn more about and know nothing about it
[4:26 PM] Formula350: How well do 310 (3-10?) Helices fold in the lab (say, compared to the traditional alpha helix we use in the game), and does the Alpha Helix do the best or is there another helix type that does similarly well?
[4:27 PM] serca: Do you know how the Foldit Remix database is generated? When I try Remix tool I often get glycine and proline forming a shape that is detected as helix by Foldit autostructures. I read in a paper that glycines and prolines are pretty common in transmembrane helices because the environment is not that agressive to break the secondary structure. Is that the reason why Remix tool tends to make helices from h these AAs?
[4:27 PM] ipatrol So yeah, I understand the difficulty with synthesizing some non-canonical AAs. I think it's pretty easy to pin arbitrary AAs to the end of an existing protein though, so maybe they can be allowed for just the ends?
[4:28 PM] horowsah: @Sarca: sorry, I don't know much about the details behind remix. I can't be too much help on that one
[4:28 PM] Josh: Dang y'all ask a LOT of good questions! For the record, this all goes over my head too =P
[4:29 PM] horowsah: @Formula350: I don't know specifically how well 310 helices do in designed proteins, but they're pretty rare in natural proteins, presumably for good reason. One thing to know is that Rosetta (which is the scientific engine Foldit is built upon) tends to have a love of alpha helices beyond what nature does. They are the most stable regular secondary structure, but it's not always the best for every scenario
[4:30 PM] horowsah: @ipatrol: yeah, ends of proteins are easier to modify than in the middle, but that really restricts what you can do with them. It's way more fun to put them all over the protein, just harder. Still, that does sound like a fun idea to explore in future puzzles.
[4:31 PM] Formula350 I accidentally made one a couple weeks back, which piqued my interest. Then my fold on NSP6 Prediction we're working on ended up creating a 310 on its own, which certainly gave me pause. Thanks :)
[4:33 PM] Jeff101: when you get a raw set of cryo-EM data for a multi-protein complex, how do you pick out which proteins go where?
[4:33 PM] Jeff101: how do you decide what to trim from the EM cloud?
[4:33 PM] horowsah: So there are tools that are used called segmentation tools, and they're job is to help you find gaps in the density where it looks like two proteins come together
[4:34 PM] horowsah: That's typically the first thing one does with a cryo-em map
[4:34 PM] Jeff101: are you working on any tools to help us solve ED/EM puzzles in Foldit?
[4:34 PM] horowsah: Then once it's been split up, it becomes a rather tough problem of figuring out which protein could even go where
[4:35 PM] serca And returning to the jeff's question about beta-sheets. Is that correct that even without H-bonds formed, beta-sheets tend to form extended straight lines? How likely loop-biased AAs tend to form that kind of straight lines?
[4:35 PM] horowsah: to my understanding, this is mostly trial and error for most people. Start building something and then figure out which amino acids it is later. it used to be that the maps weren't high enough res to do that, but now it's possible. Before, you'd have to do things like attach antibodies to different parts of your structure to see what went where.
[4:36 PM] ZombieRaccoon: Is there any possibility of being able to color the cloud with a paint program or something ? I think it would be very helpful to deal with it in small chunks and try to note areas by coloring bits of the cloud
[4:36 PM] Jeff101: we had a puzzle #1667 with too much density. I think its ED cloud was for perhaps 10 proteins, and we were supposed to find where a single protein fit into it. We didn't know the sequence for the 9 other proteins there.
[4:36 PM] Jeff101: https://fold.it/portal/node/2007726
[4:36 PM] horowsah: @Jeff101: have you guys gotten a puzzle with the new radius tool for ED/EM yet? I actually can't remember.
[4:37 PM] Formula350: I loaded a Beginner ED puzzle a week or two ago and there was a Radius slider as I recall. Not sure if that's been there or brand new, though.
[4:37 PM] horowsah: @Serca: so prolines for example tend to not be in beta sheets because they will cause a kink. Glycines are also usually too flexible. The other amino acids that don't show up in sheets often actually have more to do with the side chains and how they fit, usually.
[4:38 PM] serca: Formula350 i guess you got 3-10 on nsp6 because hydrophobis deformed the helix pretty much, trying to get buried
[4:38 PM] horowsah: Ah yes, 1667. We really weren't sure what was going to happen with that puzzle, but we thought players might like a taste of what the whole problem was like before it gets broken up.
[4:38 PM] Jeff101: I think 1797 was the last ED puzzle we got. It was in February.
[4:39 PM] horowsah: So the radius tool is one to definitely use if you haven't before in ED puzzles. We have some more tools for ED we're starting on, but there's no real timetable for them yet.
[4:39 PM] ZombieRaccoon: I like the ED puzzles, just have a hard time keeping track of area
[4:39 PM] horowsah: On ED puzzles, I tend to use Q a lot
[4:40 PM] horowsah: That's the easiest way for me to keep things in the right parts of my screen with the density
[4:41 PM] Formula350 Might be, Serca, but there's not a whole lot of them on the one that turned 310 lol (mind you, it's less 310 looking currently, than it had been 2 days ago)
[4:41 PM] Formula350 IMAGE: http://fold.it/portal/files/chatimg/irc_817348_1590532907.png
[4:42 PM] Jeff101: below is a feedback I wrote after puzzle 1667 with ideas for ED Puzzles.
[4:42 PM] Jeff101: https://fold.it/portal/node/2007742
[4:44 PM] serca: Formula350 i got some 3-10 like helix too near the phenylalanine parts of the helix
[4:44 PM] horowsah: I remember that feedback. it's interesting, the way foldit has in the past done ED puzzles is really different from how crystallographers/em people do it. For example, I never use labels, but that's because I always thread the backbone first, and figure out sidechains later. However, in ED puzzles so far, we don't let players do that
[4:44 PM] horowsah: If you're curious as to the reason, it's hard to enforce the "right" sequence if we give you a design puzzle
[4:45 PM] Formula350: That one in the screenshot is #49 (I deviated after running ""FormPRED"" on it lol)
[4:47 PM] ZombieRaccoon: Have you considered using a generic em cloud for design puzzles to define where the designed protein should be, and penalize parts outside the cloud
[4:48 PM] ipatrol: Isn't that already a puzzle type?
[4:48 PM] horowsah: I've actually done that with classes before. I've gotten good feedback and the fun level, but it tends not to produce the best science.
[4:49 PM] ZombieRaccoon: Bummer
[4:49 PM] horowsah: I'm still working on a way to make it better, though!
[4:50 PM] Jeff101: why doesn't Foldit let players thread the backbone in ED puzzles?
[4:51 PM] horowsah: So threading the backbone would be like having a design ED puzzle in which you first thread it, then change the amino acids. In theory, this should work, but it requires a ton of detailed mutation work after the fact not to what would be most stable, but whatever the protein has originally. That means things like auto-mutate don't work right, and the score function will penalize you for getting the correct sequence.
[4:52 PM] horowsah: So we'd have to put in a sequence filter on top of it, but not that would outweigh the other important things in an ED puzzle... not easy
[4:52 PM] serca: horowsah is there any characteristic of the helices/sheets like having different degrees of freedom comparing to the loops? I am trying to make the SS prediction in Foldit with getting some statistics about how they Remix in Foldit. I've noticed that the are some AAs sequences that have much more solutions from Remix database than others. That is correct for helix-propensity AAs, so I wonder if that is correct for sheets. What do you think about ponte
[4:53 PM] horowsah: When you say different degrees of freedom, could you expand on that?
[4:54 PM] ipatrol I would guess he means in parameter space terms
[4:55 PM] Jeff101: maybe remix gives more possible solutions for loops than for sheets than for helices
[4:55 PM] horowsah: so i think it was designed to give more for loops
[4:56 PM] horowsah: as to sheets vs helices, i'm not sure. Building good loops is notoriously tough.
[4:56 PM] Formula350 (Serca, for the record, your message cut off at "What do you think about ponte...")
[4:57 PM] Jeff101: for threading ED, could we start with a poly-alanine protein, fit its backbone in, and then mutate to match the desired sequence?
[4:57 PM] Jeff101: or poly-gly?
[4:57 PM] horowsah: yep, that's the way it would work
[4:57 PM] horowsah: i'll work on seeing if we can give that a try
[4:58 PM] Jeff101: the mutation wouldn't be that much computation, because the AA's would always follow a certain sequence
[4:58 PM] Jeff101: if one sequence doesn't match, just shift it along the polypeptide until it matches
[4:59 PM] horowsah: I have to head out in a minute y'all, so maybe just one more question if anyone has one?
[4:59 PM] Formula350 Sorta followup to "good loops are notoriously tough" (partial ramble here, I apologize) If ideal loops play an important role in our proteins folding in a lab, would it perhaps work better if we could instead work on an "Asymm Trimer"? Being able to make three individual helices that all want to strongly bond together, and effectively work as a Monomer, but being Asymm would then still let us determine one side as the docking side for the target (same as how we design what we do already). Or would this be too easy get destabilized and fall apart?
[4:59 PM] Jeff101: phosphates on sidechains
[4:59 PM] :mib_hgy8xr: @Jeff101- there was a puzzle like that 5 years ago: https://fold.it/portal/node/2000180
[5:00 PM] Jeff101: (thanks mib)
[5:00 PM] horowsah: Asymm trimer, like as three separate polypeptide chains?
[5:01 PM] Formula350 Yea, basically. Not connected by anything in the puzzle since they would be effectively individuals, and created individually, but later introduced next to each other to bond.
[5:01 PM] Marsfan--Formula350 I just had a crash trying to run all filters on 1840. What files do I need to send and how should I send them?
[5:02 PM] Formula350 Mars: Did you copy out the log.txt and/or debug.txt before starting Foldit up again?
[5:02 PM] horowsah: Practically, that would be pretty hard, simply because you'd have to co-express h proteins simultaneously in E. coli to make them, which increases the number of things that can go wrong by some large number. But there are applications for things like this out there. Is it easier than finding a good loop? I'd say probably not in most cases.
[5:02 PM] marsfan: Copied it out
[5:02 PM] horowsah: Ok, I have to head out- thanks all!

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office Hours on 6/11/20 led by Bkoep

5:02 PM] bkoep:: Hi everyone! I'm looking for office hours.
[5:02 PM] serca: i typed "11pm friday pst zimbabwe" according to formula350 message about friday and got no right time except google links to converters
[5:02 PM] bkoep:: I'm late and I can't find the office
[5:02 PM] formula350: For example, when I'm at the cabin (in Alaska) on my hotspot, sometimes I'm located in Seattle.
[5:02 PM] wboler: I'm guessing you forgot to bring donuts as well?
[5:02 PM] bkoep:: Is #veteran right? Can we set up an office here?
[5:02 PM] susume:: this sounds like 90% of my bad dreams
[5:02 PM] formula350: Yep the office is here.
[5:03 PM] formula350: You can't see us becauase we're all in our cubicles.
[5:04 PM] bkoep:: Okay, great!
[5:04 PM] bkoep: I'm bkoep:
[5:04 PM] bkoep: I'm a scientist on the Foldit team
[5:04 PM] serca: typed in google: "2:00pm GMT here" with no proxy and got 17 hours mistake. pretty sure my timezone in windows is +-1 hour from my ip timezone
[5:04 PM] bkoep: I'm here to answer questions and chat about proteins
[5:04 PM] serca: hey bkoep
[5:05 PM] serca: bkoep- why do we have so few prediction puzzles of the proteins with unknown and unpublished structures? can we have more of them? prediction is very interesting, what is the problem with getting that kind of puzzles?
[5:05 PM] wboler: Do you have any plans on making a "continuous" version of the BUNS filter? The discrete stops seems to be overly harsh.
[5:06 PM] wboler: steps* even
[5:06 PM] bkoep: Good question @serca
[5:07 PM] bkoep: The short answer is, we think Foldit is most useful for protein design problems
[5:07 PM] alcor:29 Does Buns count buns in the core also, not just the interface?
[5:07 PM] serca: bkoep: so, de-novo only? no hope for the real prediction?
[5:07 PM] formula350: It counts them wherever they are, alcor:
[5:08 PM] alcor:29 Tx bkoep:
[5:08 PM] bkoep: @serca: how do you mean de novo vs. real prediction?
[5:08 PM] bkoep: @wboler: that's an interesting question about "continuous BUNS"
[5:09 PM] wboler: My line of thinking is attributing some sort of "distance" function to how close we are becoming to creating a BUN.
[5:09 PM] nspc: IMAGE: http://fold.it/portal/files/chatimg/irc_959223_1591830575.png
[5:09 PM] bkoep: I see
[5:10 PM] formula350: lol alcor. I know but... here's an example (to the right of my Pin)
[5:10 PM] formula350: IMAGE: http://fold.it/portal/files/chatimg/irc_817348_1591830613.png

[5:10 PM] bkoep: Well, what it really comes down to, with BUNS, is whether or not the polar atom can make a hydrogen bond
[5:10 PM] serca: prediction puzzles is something with no pdb records. de-novo proteins is something very different that has no direct connection/links to its structure
[5:11 PM] bkoep: And it's true that H-bonds come in all different strengths and energies
[5:11 PM] bkoep: But the gradient between "can make an H-bond" and "cannot make an H-bond" is pretty steep
[5:12 PM] serca: my pc is not that good to new proteins design and i like to make something like prediction job on the puzzles with inlown secondary tertiary structure
[5:14 PM] bkoep: What I mean is, very small changes in atom position can yield a huge difference in H-bond energy
[5:14 PM] serca: *with unknown secondary/tertiary strcuture
[5:14 PM] wboler: True, I just wasn't sure if there was a better way. I did something similar for my master's thesis.
[5:14 PM] spvincent: IMAGE: http://fold.it/portal/files/chatimg/irc_3010_1591830885.png
[5:14 PM] wboler: (Not related to protein folding)
[5:15 PM] bkoep: @wboler, well now I'm curious! Can you say more?
[5:16 PM] wboler: I did work on optimizing the best placement of radio antenna with respect to "zones". If we gave the algorithm discrete zones, it would constantly get stuck in local optima. To guide it, I created a mix between continuous and discrete: small "energy" that grew the closer a zone edge was reached, until constant fitness for reaching within the zone.
[5:17 PM] wboler: It made me wonder if something similar could be done for BUNS, but I wasn't sure how hard that would be.
[5:18 PM] bkoep:: @serca: I see. Unfortunately, I think the prediction puzzles are just not as fruitful, scientifically (compared to design puzzles).
[5:19 PM] bkoep: So if we have to focus our efforts, we'd like to focus it where we can make the greatest contributions.
[5:20 PM] formula350: If BUNS aren't specifically your area, that's ok, but my own BUNS related question is: Would it be possible to pre-process the target by loading the ENTIRE protein with the BUNS filter, then remove them from the target in our puzzle? That way, the ones that show up by default (ie the 15 on 1846) would be factored out, thereby eliminating a bunch of visual clutter.
[5:22 PM] susume: some of those are reachable by the foldit design tho, would not want them factored out
[5:23 PM] jmbrownlee333: I am hearing that you no longer think that crowd sourcing is a way to solve the "protein folding problem". Is that true?
[5:23 PM] formula350: Granted, but if they aren't BUNS when the entire target protein is present, then they shouldn't technically be anything WE have to worry about.
[5:23 PM] bkoep: @wboler: cool! In that sense, I guess a continuous BUNS score would be really helpful for gradient descent (wiggle)
[5:23 PM] bkoep: The problem there is that BUNS calculations are really slow
[5:24 PM] alcor: 29: I had asked because I thought the interface buns interfere with docking. The others might prevent proper folding but some of the others on the perimeter might bond with water.
[5:24 PM] wboler: That's what I gathered
[5:25 PM] formula350: Yea, my question was quite directly related to what you had asked alcor :) I know that the Target's default BUNS cause a lot of confusion for players. So I figured if some of them didn't technically exist in the entire Target Protein, then having them omitted would help us understand.
[5:25 PM] bkoep: @formula350: Yes, that's a feature we'd like to add to the BUNS objective. It did not make it into the 1st generation BUNS prototype.
[5:26 PM] formula350: Ah that's great news! Thank you :)
[5:26 PM] serca: @serca: I see. Unfortunately, I think the prediction puzzles are just not as fruitful, scientifically (compared to design puzzles).
[5:26 PM] serca: Not that fruitfull to what? We now have 2 de novo puzzles and 1 revisting puzzle. Is predicting a protein with unknown scructure is worse than all that three puzzles? So predicting puzzle is even worse than pretty senseless work on a puzzle with known structure?
[5:27 PM] bkoep: Like @susume: says, we'd want to be careful about BUNS that can be addressed in the target, so we wouldn't get too fancy—just a setting to let us exclude specific residues from the BUNS calculation
[5:27 PM] susume: is the coordinate system for the BUNS voxels anchored on the locked protein? or can it move when our protein moves? some players observe BUNS changes far from where they just made a minor change, I wondered if it is due to the voxels moving
[5:28 PM] serca: i am ok with the de-novo puzzles, just cannot undeerstand how revisting puzzles are better than precition ones
[5:29 PM] wboler: Oh yeah, can we have an API to count bonds and buns programmatically in recipes?
[5:29 PM] serca: is that really the problem to find protein with the unkown structure? what is t you algorithm to find one?
[5:29 PM] wboler: (Unless something like that already exists and I'm missing out)
[5:29 PM] formula350: That may be a Josh and Neil question, I'm not sure.
[5:30 PM] jmbrownlee333: Here is a question from a player who can’t be here just now. Skippysk8 wants to know how to think about shape complementarity, vs overall square angstroms of the interface, vis specific electrostatic interactions at the interface.What’s more important to the design.? How close to atoms need to be at the molecules interface?
[5:31 PM] bkoep:: @jmbrownleee333, I'm not saying that crowd-sourcing is useless for protein structure prediction. But I'm convinced that Foldit players can make a greater impact with protein design.
[5:32 PM] jeff101: how do you plan to use our results from puzzle 1847? it seems like you already know what the structure should be.
[5:32 PM] jeff101: 1847 is the ED puzzle for the player-designed protein
[5:33 PM] serca: bkoep: that sound like that you won't have any prediction puzzles here anymore.
[5:34 PM] susume:: I had asked bkoep for 1847 just for fun, since it's a foldit player design we haven't played before
[5:34 PM] serca: does it mean there is no need in any predictionion puzzle solution?
[5:35 PM] formula350: Maybe it may be relevant and clear up some of serca:'s confusion (which also touches on your answer to JMB): On the submitted predictions on say ORF8/6/3, what percentage of our results have proven useful?
[5:35 PM] formula350: (as I think what we may need to hear is: we basically suck at prediction and our results show it, which is why we don't get many prediction puzzles)
[5:36 PM] bkoep: @serca I see your point. A big issue with the blind prediction puzzles is that we don't have a good way to evaluate the accuracy of Foldit predictions.
[5:36 PM] serca: susume 1847 looks like a solution for you question in the last office hour about trying to use some unknown puzzles vs revisted ones
[5:36 PM] susume: serca there are many unsolved proteins, but only a few that a researcher is about to solve, or has solved and not published - if we worked on predicting the ones no researcher is working on, we will never know how well we did
[5:37 PM] Migi-irc: well formula350 my suspicion is that AlphaFold is just better than the Foldit community at prediction puzzles
[5:37 PM] susume: the coronavirus ones are good because people are working to solve them right now, we might get results soon
[5:37 PM] bkoep: In rare cases (like the MPMV protease from 2011), we have experimental data that we can use to confirm predictions.
[5:37 PM] jeff101: why not have a puzzle like 1847 before the Foldit Team solved the structure? Why not let us help you?
[5:37 PM] serca: the only question i don't understand why do we have no new puzzles with unknown sturcture, and have so much revisiting puzzles, that just seem to be scientificaly senseless
[5:38 PM] jmbrownlee333: To follow on Jeff's question.do you plan to look at the player models in terms of R and/or Free? Do player structures tend to move either of these more or less.
[5:38 PM] jmbrownlee333 ignore my question, if too wonky.
[5:39 PM] serca: susume that sounds frnakly enough. that is why i am intersting why don't we have some puzzles that scientists are really working on
[5:39 PM] susume: I think it is hard to get non-foldit scientists to share their data
[5:40 PM] wboler: My guess is the revisits have to do more with validating changes made to the program, which are useful.
[5:41 PM] formula350: I can appreciate serca:'s viewpoint. Assuming that there's no limitation on the server-end for having more than 3 puzzles running (as I know there are periodic automated submissions), it might be nice to exclude the Revisits as one of the "3", and include a Prediction puzzle, regardless.
[5:41 PM] serca: susume i even tried to find a puzzle with the known environment an unknown 3d structure and length less than 200 to work on understood that it is not hat easy to find one
[5:42 PM] serca: that is the reason of my question and that is wju i try to clear that in me next questions: i get no the clear answer from bkoep: (sorry dude, take my respect)
[5:43 PM] bkoep: @serca: It's true the scientific value of the revisited puzzles is thin. They are definitely useful for long-term, continuous benchmarking (i.e. is Foldit performance consistent over time as the software and community evolves).
[5:44 PM] bkoep:We also like the Revisited puzzles as a gateway for novice players, but that doesn't really have anything to do with scientific results
[5:46 PM] serca: and i also like the puzzles with unkown structure vs revisting ones to work as the only reason th present myself in foldit.
[5:46 PM] alcor:29 Any progress in the labs this week on Covid puzzles?
[5:47 PM] serca: but i am ok if you discriminate my 10y.o. cpu tand my hand folding skills to run just revisitng puizzles
[5:47 PM] serca: *tand=and
[5:47 PM] pc: I like design puzzles
[5:49 PM] serca: guess i like them too if my cpu is ok with all the filters
[5:49 PM] jmbrownlee333: can I re-ask skippysk8's question. Skippysk8 wants to know how to think about shape complementarity, vs overall square angstroms of the interface, vs. specific electrostatic interactions at the interface.What’s more important to the design.? How close to atoms need to be at the molecules interface?
[5:50 PM] jmbrownlee333: to=do
[5:50 PM] bkoep:: @jmbrownlee333 @jeff101 Those are good questions about the ED puzzle! When x-ray diffraction is this good, it is less work to solve the structure automatically than to set up the Foldit puzzle.
[5:50 PM] serca: or they just switch their filters to gpu, to unload my cpu
[5:51 PM] susume: (thx jmb, you understood skippy's question better than I did)
[5:51 PM] bkoep: The more interesting challenge comes from poorer quality or lower resolution data, where the automated methods fail. We hope to bring more of those puzzles to Foldit.
[5:53 PM] jeff101: if you took our 1847 solutions and removed the ED part of the scores, would the results be useful for understanding the energy landscape of the designed protein?
[5:54 PM] bkoep:: @alcor 9 Yes, the lab experiments are progressing for h the CoV spike and IL-6R binders, but I still can't give a time estimate on when we'll have data
[5:54 PM] :alcor: 29 Tx
[5:54 PM] bkoep: Frankly, I probably can't offer a time estimate until we actually have the data
[5:54 PM] alcor:29 k
[5:55 PM] jeff101 if you think of answers to some of our questions later, please make a blog post about them
[5:55 PM] serca: bkoep can you please advice any algorithm to find a puzzle with the unknown structure?
[5:55 PM] serca: I want to have some fun on the sandbox puzzle with that.
[5:56 PM] bkoep: @jmbrownlee333 @Skippysk8s That's a great question about which binder metrics to focus on.
[5:56 PM] bkoep: We don't really know
[5:56 PM] bkoep: We will try to set their relative weights to something reasonable
[5:57 PM] bkoep: But the fact is that shape complementarity, and SASA, and BUNS, and DDG, and electrostatics all show some correlation with binding success rates
[5:58 PM] jeff101 there was another binding design project about a year ago ... non-COVID ... what is the status of that project?
[5:58 PM] bkoep: For some of these, we have some target thresholds where maybe the correlation drops off
[5:59 PM] bkoep: But I'm pretty hopeful that Foldit players can optimize all of them
[6:01 PM] bkoep: The traditional way we design binders is pretty inefficient: we spend a bunch a computational time generating 1M designs without optimizing these objectives, and then we check which of those 1M designs meet the objectives and throw out the 99% that don't.
[6:02 PM] serca: @bkoep so, no algorithm to find a puzzle with the unknown structure and known environment? no answer looks like a positive answer for that. thank you.
[6:02 PM] formula350: Do you guys/gals in the lab ever look at our designs and try to "Evo" the best candidates yourselves, or do you typically just feed them to an algorithm?
[6:02 PM] bkoep: It seems like Foldit players might be able to consciously tailor designs meet all of these criteria, instead of blindly shooting birdshot at a tiny target
[6:03 PM] bkoep: @serca Sorry, I'm not sure what you mean about "known environment"?
[6:03 PM] jeff101: cytosol vs membrane?
[6:04 PM] jmbrownlee333: or excreted.
[6:04 PM] jeff101: favoring disulfides or not?
[6:05 PM] serca: i am just talking about any structure that is not listed in the pdb is interesting. just toooooooo tired of the revisiting puzzles. but "known environment" means just hydrophobcity score.
[6:05 PM] serca: anyway i guess most of us will be happy with just any random AAs that are not listed in the pdb
[6:08 PM] serca: but the best AA sequences are made by nature. that is the reason of the feeling to talk with god while folding them.
[6:08 PM] bkoep: Oh I see. There are lots of protein sequences out there without structures. Like, the NCBI BLAST database of non-redundant protein sequences.
[6:09 PM] bkoep: And people are discovering tons of new ORFs now from shotgun sequencing experiments
[6:10 PM] bkoep: Basically, take a bucket of seawater and sequence all of the DNA in it)
[6:11 PM] serca: so what stops you do add that to our puzzles and have us to fold something really known then?
[6:11 PM] susume:: Serca you could find an unsolved sequence from ncbi and challenge players to solve it in sandbox, we won't know if anyone is right but we could compare results
[6:11 PM] bkoep: @jeff101, the IL-7R binder experiments got held up by the COVID-19 shutdown, but I think we may have some of that data soon
[6:12 PM] serca: *really unkown then
[6:13 PM] jmbrownlee333 On 1847, s there any area where the x-ray structure varied from the design more? Were loops or sheets or helices farther off from the design model?
[6:13 PM] formula350: Wonder if you can make a "Contest" using the Sandbox puzzle as a base.
[6:13 PM] serca: susume i don't like your solutions, sorry, but what is your the algorithm to find the AA sequence with unknown structure?
[6:14 PM] serca: i see foldit guys don't have any algorithm to find the AA sequence no listed in pdb and that is their problem (sorry bkoep)
[6:15 PM] bkoep: @serca You can use NCBI BLAST to check if a sequence is in the PDB
[6:15 PM] bkoep: https://blast.ncbi.nlm.nih.gov/Blast.cgi
BLAST: Basic Local Alignment Search Tool. The Basic Local Alignment Search Tool (BLAST) finds regions of local similarity between sequences. The program compares nucleotide or protein sequences to sequence databases and calculates the statistical significance of matches. BLAST can be used to infer functional and evolu...
[6:16 PM] jmbrownlee333: No like bkoep said, no way to know if you are on the right track, for who knows how long.
[6:18 PM] serca: bkoep so I want to fold a nice 150 length reisdue protein at the sandbox: what should I do to find one that is not listed in rcsb pdb?
[6:18 PM] serca: could you please tell me the buttons i have to press?
[6:18 PM] susume: Serca this is not something someone can teach you in 3 lines of chat, go read about blast and educate yourself
[6:19 PM] serca: very draft how you understand that
[6:19 PM] jmbrownlee333: @serca i have some ideas, can chat in group or offline.
[6:19 PM] jmbrownlee333: Ie PM
[6:19 PM] serca: jmb that is interesting in prespective
[6:20 PM] serca: susume: you know... i guess the reason we have no any proteins with the unknown structure is that they just don't know how to find that
[6:20 PM] bkoep: @serca I'm sorry, I cannot tell you which buttons to press. But the NCBI tools are pretty nice, I think you will find them easy to use
[6:20 PM] bkoep: Okay all, I'm sorry, I have to run

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office hours 6/26- led by beta_helix and SethCooper

Office hours 6/26- led by beta_helix: and SethCooper:

[6:00 PM] beta_helix:: Hello everyone!
[6:00 PM] donuts554: hello how are you?
[6:00 PM] SethCooper: hello!
[6:00 PM] beta_helix:: Seth Cooper and i will start off today's Office Hour
[6:00 PM] jeff101: hi
[6:00 PM] formula350: Wait it's just Beta Helix? I thought it was Gamma Helix who'd be here. :( Welp, I'm leaving then......... lol
[6:00 PM] beta_helix: Doing good, donuts554! Hi jeff101!
[6:01 PM] beta_helix: hahahhahahahhaa
[6:01 PM] donuts554: how are you seth?
[6:01 PM] formula350: :P
[6:01 PM] SethCooper: doing pretty well
[6:01 PM] formula350: First thing's first.... Chocolate or Vanilla?
[6:02 PM] formula350: And then: Who would win in a fight: Beta or Seth? :}
[6:02 PM] beta_helix: Gamma Helix would beat h of us!
[6:02 PM] SethCooper: probably a tie
[6:02 PM] formula350: Alright I've gotten all my shenanigans out of the way ;)
[6:03 PM] donuts554: Chocolate and idk
[6:03 PM] alcor29:: Question: Currently Foldit Rosetta?) does Wiggle, Wiggle sidechains, and Shake sidechains as discrete operations. Has anyone ever looked at coding that would do all three as a single fluid motion, as one might imagine happens in a real environment? Would that be desirable? If so, even doable?
[6:03 PM] formula350: Donuts we're asking THEM questions, not each other lol
[6:03 PM] donuts554: oh ok
[6:04 PM] formula350: Wiggle does "Wiggle Backbone" and "Wiggle Sidechains" at the same time. Just to point that out.
[6:04 PM] alcor29: k
[6:04 PM] beta_helix: It's all good
[6:04 PM] SethCooper: yeah, I think early on we had considered having an operation that does h shake and wiggle in one operation
[6:04 PM] beta_helix: That is a very good question, alcor29
[6:05 PM] SethCooper: but I think even at that level, it would have to alternate wiggle and then shake somehow
[6:05 PM] beta_helix: One thing that the Rosetta community noticed, and I know you have as well, is that there is often a big difference between Wiggle-Shake or Shake-Wiggle
[6:06 PM] beta_helix: That is why Blue Fuse (and Relax in Rosetta) do these alternations with different thresholds (of clash importance, for example)
[6:07 PM] beta_helix: I agree that it would be a lot more realistic to have one fluid action... like in Nature. That is what molecular dynamics simulations try to do.
[6:08 PM] alcor29: Thx.
[6:08 PM] donuts554: My Question is part of something else, but is it necessary to render aromatic compounds without the see-through holes?
[6:08 PM] donuts554: In Foldit
[6:08 PM] donuts554: I mean amino acids not compounds
[6:09 PM] beta_helix: So, here is a TRP in Foldit, right?
[6:10 PM] donuts554: Yes
[6:10 PM] donuts554: I can't see the image though
[6:10 PM] beta_helix: https://foldit.fandom.com/wiki/Amino_Acids
Foldit Wiki
Amino Acids
Amino acids (also called segments or residues in Foldit) are the building blocks of proteins. Each amino acid has a unique sidechain, except for glycine. Each protein is identified by a unique...
[6:11 PM] formula350: I looked at it, it's simply the Tryptophan as you know and love (or I guess in your case, hate lol)
[6:11 PM] beta_helix: (sorry, I'm in Discord
[6:12 PM] donuts554: ok then yes it is a TRP in FOldit
[6:12 PM] beta_helix: You can render it without the hole with the Sphere view, right?
[6:13 PM] donuts554: Yes I can
[6:13 PM] beta_helix: Are you asking if we should do the same in stick mode?
[6:13 PM] donuts554: But it its more puffy and takes more space and makes the surface covered
[6:13 PM] donuts554: Yes I am asking that
[6:14 PM] beta_helix: If you look at all these amino acids from a cartoon view, such as TRP:
[6:14 PM] beta_helix: https://en.wikipedia.org/wiki/Tryptophan
Tryptophan
Tryptophan (symbol Trp or W) is an α-amino acid that is used in the biosynthesis of proteins. Tryptophan contains an α-amino group, an α-carboxylic acid group, and a side chain indole, making it a non-polar aromatic amino acid. It is essential in humans, meaning the body can...
[6:14 PM] beta_helix: Then we are essentially just drawing lines for each bond... which in this case creates these 2 rings.
[6:15 PM] beta_helix: As you mentioned, if we didn't draw it that way it would make the surface covered and takes more space.
[6:15 PM] beta_helix: Our Cartoon and Cartoon Thin are designed to give you enough information about the sidechain, without so much detail that you can't see what is going on.
[6:17 PM] donuts554: Oh ok, thanks for the answer! I have another question but i'll let others ask their own
[6:17 PM] formula350: If your followup is directly related, it might be good to ask it.
[6:17 PM] donuts554: Oh ok then I'll ask it
[6:17 PM] beta_helix: There are 2 of us, so we can handle multiple questions
[6:19 PM] jeff101: one question folks were asking the other day is if Foldit accounts for pi stacking, and if so, which energy subscores reflect it best?
[6:22 PM] donuts554: I think thats kinda related to my question somewhat
[6:22 PM] beta_helix: I know Foldit does for sure... but I don't know which subscores show that.
[6:23 PM] jeff101: sorry donuts
[6:24 PM] beta_helix: I will find that out and post it in the Forum for you (unless that question was already posted somewhere)
[6:25 PM] donuts554: Its ok, I meant that I think it was related to my previous question about aromatic compounds without the holes
[6:26 PM] donuts554: The next question im going to ask is going to be a bit long because I have to describe things
[6:26 PM] formula350: We need "Arcade Mode"... With big text popups and an announcer. Displaying like: GREAT PI STACKING!! or an annoucer proclaiming "You just flipped Covid the Zinc Finger!"
[6:26 PM] beta_helix: Type away, we'll answer any other questions
[6:26 PM] beta_helix: It's funny that you mention that, formula350:, because the early versions of Foldit had that!
[6:27 PM] formula350: Donuts, just emember that Foldit can only send messages that are so-long before it gets cut off.
[6:27 PM] beta_helix: Seth, am I right in remembering that most players turned that off?
[6:27 PM] SethCooper: i think the majority of options are left on the default
[6:27 PM] beta_helix: I mean, it didn't say "Finish him!" (this Protein!)
[6:28 PM] jeff101: what things are each of you working on these days?
[6:28 PM] formula350: haha I mean, I have sounds disabled, but I wouldn't mind visual stuff, as I think it being pointed out when you create a.... "motif" (?? I guess that's what it'd be considered as), would actually be a learning experience for some of us.
[6:29 PM] beta_helix: donuts, feel free to break your posts up so they don't get cutoff!
[6:29 PM] beta_helix: Seth, you go first (your stuff is cooler ;-))
[6:29 PM] Susume: I would love it if foldit could recognize a motif - it mostly crows about smaller things like H bonds
[6:30 PM] beta_helix: ohhhhhh, so you mean USEFUL popups?
[6:31 PM] formula350: :P Yea, like Rank Up. However, adding some comedic flair would probably make it more appealing to the younger (or more childish, in my case) crowd since it'd provide some "action".
[6:32 PM] alcor29: A little Wagner leitmotif?
[6:32 PM] Susume: dystopian foldit popups
[6:32 PM] beta_helix:: @jeff101: Scott Horowitz and I just submitted an NIH grant yesterday to make Foldit Electron Density state of the art (ie incorporate the tools used in the best ED programs out there) fingers crossed it gets funded
[6:32 PM] donuts554: Is it necessary to have a new "View Protein" option with aromatic compounds without the holes, the proline without its hole, the thick helices as cylinders, the sheets without the zigzag contours --
[6:33 PM] Susume:drools for upgraded ED tools
[6:33 PM] beta_helix: Right?!?!?!
[6:33 PM] jeff101: I've asked questions like this in previous Office Hours. Not sure if I asked you guys.
[6:34 PM] jeff101: the papers I've seen don't generally show where the h-bonds go
[6:34 PM] beta_helix: I just want to say that I've been trying to get ED funding since I left Seattle in 2013... but it is hard!
[6:35 PM] jeff101: I hope you get the funding.
[6:35 PM] SethCooper: one thing we have been working on is something like the tutorials but more dynamic
[6:35 PM] beta_helix: Me too! I'm submitting another grant to the NSF next month as well. (edited)
[6:35 PM] Susume: do you feel like there is a "market" for foldit to work on ED? do scientists maybe feel like they're doing just fine without us?
[6:36 PM] spvincent Get the word Covid in the funding application somewhere and hopefully people will throw money at you.
[6:36 PM] beta_helix: @spvincent don't worry... it's in there!
[6:36 PM] SethCooper: (in this case "we" means mostly Josh)
[6:36 PM] spvincent good!
[6:37 PM] beta_helix: @Susume it's a very good question... but I honestly think that the issue has been how unorthodox Foldit is (especially when asking for grant money)
[6:37 PM] formula350: Call us a Think Tank and it might go over a bit better?
[6:38 PM] donuts554: --to eliminate all convex angles in sidechains, so a lysine pointing straightout is not a zigzag, instead its a striaght line, a valine looking like a orange triangle, and a arginine looking like a blue triangle on top of a blue pole with blue ends --
[6:38 PM] beta_helix: You can imagine that when a panel has 100 applications and only enough money to fund 10... it's not hard to believe that they will go with (what they feel is) the "safer" bet.
[6:39 PM] beta_helix:: formula350: u know, you are not far off... because a lot of grant reviewers reject the notion of "gamers"
[6:40 PM] beta_helix: I talked to a Program Manager at the NSF, and he agreed that if I only refer to Foldit players as "citizen scientists", it would be received a lot better! sigh
[6:40 PM] donuts554: --and to have any H-bonds or disulfide formed between amino acids more solid and widened so that it looks like the amino acids are actually stuck together?
[6:41 PM] beta_helix: (Anyway, I can ramble on about this for ever... so I'll stop and answer donuts)
[6:41 PM] beta_helix: @donuts disulfides are a very good example of what you are talking about
[6:41 PM] beta_helix: Because those are some very strong bonds!
[6:42 PM] beta_helix: Seth, jump in here, but I think one of the reasons we didn't go with the h-bond glue (for example) is that we do want you to try different topologies.
[6:42 PM] donuts554: Not glue
[6:42 PM] Susume: given how long some of us have been playing, we've almomst become a "curated collection" of citizen scientists (kinda like being pickled I guess)
[6:43 PM] donuts554: I mean like in this new view protein option, when you pull the two amino acids apart, the solid rod connecting the two charged atoms disappears
[6:43 PM] beta_helix: If you have a protein with a lot of molecules stuck to one another, wouldn't you be less likely to break those?
[6:43 PM] donuts554: Like a normal H-Bond would in cartoon
[6:43 PM] donuts554: No It just looks like its stuck together
[6:43 PM] beta_helix: but that rod would look like it was almost as strong as a carbon-carbon bond, right?
[6:43 PM] SethCooper: well, I think bands can be used to hold h-bonds together if wanted, if that's what you mean
[6:44 PM] donuts554: No not hold H-bonds together, I mean, the same thing as now --
[6:44 PM] donuts554: But
[6:45 PM] formula350: It's all about "spinning" it. Electrons spin, PR spin, as does Grants for funding. In reality, you're not lying by referring to us as Citizen Scientists, or Distributed Brainpower, or Community Think Tank. In the realm of science I can't ever see "Gamer" as being taken legitimately or seriously. Some of us (Susume) have literally been so inspired by Foldit, that they've shifted their life to pursue it. So I think calling Susume a "gamer" is arguably
[6:45 PM] beta_helix: @Susume I like that term... I should use it in a grant be it would go way over their heads! Uh oh... do any grant reviewers hang out in veteran?
[6:45 PM] donuts554: The solidess used to render the bonds holding the amino acids together is used for the H-bond and disulfide bonds as well
[6:45 PM] donuts554: Basically so all bonds look the same
[6:46 PM] donuts554: only with different color of course
[6:46 PM] donuts554: and so that only h-bonds and disulfide bonds can be broken
[6:46 PM] donuts554: thats what I mean
[6:46 PM] beta_helix: Donuts, why would you want them to look the same? The reason we color some orange vs blue (as you just pointed out) is to highlight their difference.
[6:47 PM] donuts554: Yes the color stays different, but the texture looks the same
[6:47 PM] SethCooper: the hbonds and disulfide bonds are rendered a bit differently though
[6:47 PM] beta_helix: The easier it is for players to differentiate between them, the easier it is to manipulate the structure, isn't it? If everything had the same texture (or say was all loop, instead of helix/sheet/loop) then it would be very hard to visualize
[6:48 PM] formula350: Donuts if you would like the Disulfide Bond 'texture', I can make one for you to use instead.
[6:48 PM] donuts554: Yes I mean without the different rendering, same rendering for all bonds, just different colors
[6:48 PM] Susume: I think donuts is saying he wants H bonds and disulfides to be drawn like covalent bonds are currently drawn, as solid cylinders - is that right?
[6:48 PM] donuts554: Like bright cyan solid bond holding as H-bond, bright yellow for disulfide
[6:48 PM] donuts554: Yes! That's what I mean formula
[6:49 PM] donuts554: They are bonds, and they should be treated fairly
[6:49 PM] donuts554: It follows the "bond" definition
[6:49 PM] beta_helix: But they are different bonds...
[6:49 PM] donuts554: It's just their strengths are different
[6:49 PM] formula350: I think you mean Susume? He wants everything to be "solid". Aromatics, Helices, etc. Not this "there's a gap, so something could arguably pass-through it"
[6:49 PM] donuts554: Like you have chemical reactions breaking bonds and making bonds, like in 1855
[6:50 PM] SethCooper: we figured they were different enough they should look different
[6:50 PM] SethCooper: plus, we are often encouraging the creation of hbonds so they are like a "goal" in the game
[6:51 PM] spvincent: Can I ask a question about filter scoring?
[6:51 PM] beta_helix: I completely agree that we are simplifying the reality of proteins... but that is done intentionally, for the exact same reason that you mentioned earlier: Sphere mode (while being more realistic) is almost impossible to fold in.
[6:51 PM] donuts554: This will make it look like there is actual chemistry, breaking and forming bonds
[6:52 PM] donuts554: Oh ok seth nvm about the bonds part
[6:52 PM] beta_helix: Take electrostatics, for example, donuts... or even water, we don't even show those at all! So this really is a cartoon version of chemistry, as opposed to making it look 100% accurate. I hope that answers your question.
[6:52 PM] beta_helix: @spvincent of course!
[6:54 PM] spvincent: I was wondering why the bonus/penalty for filters such as Buns, Core Filter, etc follows a stepwise function as opposed to being continuous. It play havoc with scripts.
[6:55 PM] spvincent: *plays
[6:55 PM] SethCooper: I'm not sure about those specifically, but some bonuses are based on discrete things in the structure
[6:55 PM] donuts554: Well the intent is not for 100% accuracy, but also for less complicated things for the eye to remember and visualize in the brain memory, and to have less rendering so that it is less laggy and therefore less crashes
[6:57 PM] spvincent: In the real world of proteins I'd expect smooth variation. A polar atom isn't either buried or unburied: it can't be that binary.
[6:57 PM] formula350: Yea, like the uh... "IE" filter. I think it's called.
[6:57 PM] formula350: Interaction Energy?
[6:58 PM] SethCooper: for example, if a bonus is based on, say, a number of bonds or a number of residues, then it's more straightforward to score based on that
[6:58 PM] SethCooper: there are some techniques use to smooth them out in some cases
[6:58 PM] beta_helix: @donuts that is a very good point that you bring up, and something that we have struggled with for a long time... especially as we try to post puzzles with bigger and bigger proteins. We actually considered going the opposite direction of your suggestion: removing most sidechain atoms, which is when we tried out Centroid Mode (I know the mention of it just made a few players sick) needless to say, it did not go well.
[6:59 PM] formula350: But I presume that it's based on actual Rosetta scoring, versus most Filters which are ad-hoc and side-calculations that would require more processing power (time) to provide "real-time" output. Is this a fair assessment, Seth?
[6:59 PM] formula350: ("that it's" referring to something like the Interaction Energy filter)
[7:00 PM] SethCooper: yeah, the filters I think can often be simplifications based on counts or thesholds
[7:00 PM] donuts554: Like the convex angles in valine and leucine are intuitvely harder to remember than as rendered as a short, filled orange triangle and a longer, filled orange triangle (cause the vertex point in that position in 3D is located in the inside of the atom according to sphere mode)
[7:01 PM] SethCooper: the early filters/bonsuses were entirely binary, as I recall
[7:01 PM] SethCooper: so adding in the steps in scoring does kind of smooth them out a bit from that
[7:02 PM] spvincent: Its quite frustrating working with proteins when you're on a boundary and e.g. the core bonus keeps flipping from 2900 to 3000
[7:02 PM] spvincent: and back again
[7:03 PM] formula350: Yea. Not having a nice Subscore to gauge what we're doing on, makes thoses, and BUNS, etc, very hard to determine whether what we're doing is "good" or "bad".
[7:03 PM] beta_helix: donuts, you might get a kick out of this (or be mortified) but some of the very early Foldit mockups didn't use actual sidechains at all... they were random shapes (or different fruit) or other cartoony representations!
[7:04 PM] jeff101: I would request the Ideal Loops Filter scores by residues in ideal loops rather than by the # of ideal loops
[7:04 PM]SethCooper: it's quite possible some that are based on thresholds could be smoothed out a bit as well
[7:05 PM] spvincent: that would be most helpful i think
[7:05 PM] formula350: A "glow" factor, or "color transition gradiant" (for BUNS/Ideal Loops, and Core Existence, respectively) would help a lot, if that's doable as well.
[7:05 PM] formula350: Acting much in the same way that the HBond line gets thinner as it grows weaker, or fatter as it gets really strong.
[7:05 PM] Susume: (just try to get funding for something that represents sidechains as bananas!)
[7:05 PM] jeff101: being able to color by specific subscores rather than the total score would be helpful
[7:06 PM] donuts554: Also, showing the glutamine, arginine, asparagine as triangles on top of poles will make it easier to recognize the pattern of the shape of the amino acids, and to make it easier for espcially even children to recognize and categorize and theorize the properties of the amino acids and compare them, intuitively.
[7:06 PM] spvincent: That reminds me: I do not like the way the Buns filter messes with the View Settings.
[7:06 PM] formula350: Yea, they're working on that VIncent :)
[7:06 PM] alcor29: Speaking of ideal loops. Is there some way of knowing whether mutating a single residue would make it ideal, instead of the sometime laborious process of trying to find a better shape?
[7:06 PM] spvincent: oh good
[7:08 PM] donuts554: (ex. Histidine would look like a blue filled pentagon on top of a blue filled triangle on top of a pole., and Asparagine and glutimine would look like blue triangles on top of poles. I think, and even I think myself too, that they make think, --
[7:08 PM] beta_helix: @donuts554 The team is working on a more specific Education version of Foldit that would be aimed more towards schoolchildren.
[7:09 PM] formula350: Trying to smash a square protein through a round target hole with their feverous mouse clicking.
[7:09 PM] beta_helix:That's what the big spiky ball is for!
[7:10 PM] formula350: Who... ME?!
[7:10 PM] formula350: IMAGE: http://fold.it/portal/files/chatimg/irc_817348_1593220214.png
[7:10 PM] beta_helix: hahahahhahahhahahaha Nice!
[7:11 PM] beta_helix: Well thank you all for stopping by Office Hours!
[7:11 PM] jeff101: will someone post a transcript on the Foldit web page?
[7:11 PM] jeff101: thanks for coming
[7:11 PM] beta_helix: We've been trying to hold them at different times, to try to spread the timezones out as much as possible.
[7:11 PM] formula350: Yes, we managed to bring donuts this time.
[7:11 PM] donuts554: "Oh! These three amino acids all have red and blue dots, and all similarly have a blue filled triangle on top of a pole! So, they may think, oh histidine is related to asparagine and glutimine, and histidine may have came from glutamine!'"
[7:12 PM] beta_helix: Yes. We are trying to post the whole transcript, rather than just a discord link. We're still working on that, so thanks for your patience!
[7:12 PM] donuts554: And they think, "So that's why Histidine and glutamine have similar properties"
[7:12 PM] donuts554: You are welcome!
[7:12 PM] beta_helix: Someone should really make an amino acid card game!
[7:13 PM] SethCooper: that sounds like a good idea!
[7:13 PM] formula350: I think Josh made an AA based Board Game (or maybe it was Proteins in general...)
[7:13 PM] beta_helix: In that case, he should make it online so we can all play together!
[7:14 PM] formula350: It's about to be released. It's on his website as Coming Soon
[7:14 PM] donuts554: I have more questions but Ill go to the next office hours
[7:14 PM] beta_helix: Sweet... can I invest in that?
[7:14 PM] beta_helix: Thanks again everyone... Thank you Seth!
[7:14 PM] formula350: @Josh @Gamma_Helix wants to invest in your new boardgame ;)
[7:15 PM] formula350: Yes, thanks Beta and Seth.
[7:15 PM] beta_helix: Thanks donuts, I look forward to them
[7:15 PM] beta_helix: Have a great weekend everyone... stay safe, and keep up the great folding!
[7:15 PM] SethCooper: yes, thank you!

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office hours 7/17- led by beta_helix

[2:01 PM] beta_helix: Hi everyone! Foldit Office Hours are open
[2:02 PM] beta_helix: I wanted to report back from a conference that Seth and I presented at this week (virtually, of course). It was at the biggest Bioinformatics conference, and this special session was called "Bioinformatics outside the lab: How to mobilize online citizen scientists to accelerate research" https://www.iscb.org/ismb2020-program/special-sessions#sst02
Special Sessions
ISCB - International Society for Computational Biology
[2:12 PM] lociOiling:Oiling: hi - not a very chatty crowd today
[2:13 PM] Josh: I think we forgot to advertise this, haha
[2:13 PM] beta_helix: It was posted in discord, but only once I think...
[2:13 PM] beta_helix: perhaps that's not enough.
[2:13 PM] beta_helix: It's ok, I get empty office hours when I'm on campus too!
[2:14 PM] Josh: While we have you, beta, did you get the latest newsletter? What's your scientific opinion on the top scoring results?
[2:14 PM] beta_helix: Did you trick me in here to go over your newsletter?
[2:15 PM] Josh: Nah I just want feedback on the top scores. Here are more of them:
[2:15 PM] Josh: https://foldit.fandom.com/wiki/Puzzle_1862
Foldit Wiki
Puzzle 1862
Puzzle 1862: Coronavirus Binder Design: Round 13
[2:16 PM] Josh: What feedback can you give to the players on these designs?
[2:16 PM] beta_helix: So I'm one of those "scientists" that need to move my protein around in PyMOL or Foldit...
[2:16 PM] lociOiling:Oiling not supported in png
[2:16 PM] beta_helix: It's really hard to analyze them in depth with just 2D
[2:16 PM] beta_helix: hahahahahahahhahaha
[2:17 PM] beta_helix: I'll try to get an undergrad to work on that!
[2:18 PM] beta_helix: We actually have discussed this (just before covid), where ideally we'd have a viewer on the Foldit webpage, and you could view your own models
[2:18 PM] HuubR: Was it ever considered to make a 3D viewer in Foldit (I mean, different views for left eye and right eye)
[2:18 PM] alcor29: Did anything that will affect us come out of the conference?
[2:18 PM] beta_helix: (I don't know if I'm not supposed to spill that secret, but luckily there aren't many people here today!)
[2:18 PM] lociOiling: it's been suggested, HuubR, viewers like JMol have 3D
[2:19 PM] beta_helix: @HuubR but rendering in Foldit already slows us down when it comes to large proteins
[2:20 PM] beta_helix: This would be solved if we can get Foldit VR going(edited)
[2:21 PM] HuubRWould VR reduce the workload needed for rendering?
[2:21 PM] beta_helix: @alcor29 I think it was a very interesting session, first off because it was the first time all of these games have been together in on (virtual) room... which is honestly a shame that it took 12 years!
[2:22 PM] beta_helix: @HuubR great question... I think VR would open up a bunch of new limitations, but how cool would it be to move around in Foldit with it!
[2:23 PM] beta_helix: One of the other cool things about that session is that they invited 3 speakers from the video game industry to talk about how they've incorporated citizen science games in Borderlands 3 and Eve Online.
[2:24 PM] HuubR: @pc (nspc), are you listening in?
[2:24 PM] pc: hello
[2:24 PM] beta_helix: One of the interesting points they brought up was that they've learned not to hide the science in their games.
[2:24 PM] beta_helix: Hello Hello!
[2:25 PM] HuubR: (nspc is a French game developer)
[2:25 PM] beta_helix: At first when they designed these scientific minigames (inside these massive video games) they felt they should try to mask the science as much as possible.
[2:27 PM] Susume: the same kind of nerdiness that makes gamers like to research and memorize meta and lore and character stats is the same kind of nerdiness that makes science fun and addictive
[2:27 PM] beta_helix: but they realized that was the wrong way to go about it, and players were actually interested in understanding the science behind it... that it reinforces the "WHY am I playing this?"
[2:28 PM] beta_helix: For sure, Susume! Another thing they pointed out is how many science-lovers ended up in the game industry.
[2:30 PM] pc: Yes in foldit I often ask this question to myseft (in design puzzles): "Is my protein works in real life ?"
[2:30 PM] beta_helix: C'est une question qui n'est pas facile a savoir!
[2:30 PM] pc: I know score system is not perfect
[2:31 PM] beta_helix: Opps, sorry, I had caps lock on there!
[2:31 PM] beta_helix: It's a question that isn't easy to get the answer to.
[2:31 PM] pc: And add more metrics is complex, because current metrics take lot of computation (like SASA DDG and SG)
[2:32 PM] pc: I saw lof of top score solutions that have just 3 links with target, but not realy shape complementarity
[2:32 PM] beta_helix: That is the Foldit dilemma that's we've faced from the start.
[2:32 PM] lociOiling: gotta watch caps lock on those canadian keyboards
[2:33 PM] beta_helix: We could have had a client that could easily load 1000 amino acids (like PyMOL), but you wouldn't be able to produce anything useful with it.
[2:33 PM] beta_helix: @lociOiling so true
[2:33 PM] HuubR: Now that we're talking about scoring, there's a question that has been in the back of my mind for some time
[2:33 PM] lociOiling: don't get me started on swiss
[2:34 PM] beta_helix: There is also the balance between letting you fold without filters, but then when you turn on the filter you find out all your work was useless
[2:34 PM] beta_helix: Which is worse?
[2:36 PM] beta_helix: Right now, you can think about it that we tell you this information after the fact (when we analyze your solutions). It has to be better to give you that ability... but it is still frustrating to have your hard work invalidated if you click a "check my protein-ness" button.
[2:38 PM] HuubR: I'm sure the present scoring (is it Rosetta?) is very well suited for prediction puzzles, but isn't it so that for design puyzzles, it might favor certain amino acids over others?
[2:38 PM] beta_helix: @HuubR great question
[2:39 PM] HuubR: Or certain secondary structures, for that matter?
[2:39 PM] beta_helix: It is Rosetta (there are just many different flavors of Rosetta). You can imagine that it is a lot harder to come up with an accurate score function when dealing with design.
[2:39 PM] pc: For computation we can use GPU acceleration or multithreading (but it is complex). An other idea can be : for some metrics, dont include them in ojectives but just have some recipes availables and run them to check our protein. (or a lua fonction for thoses metrics)
[2:40 PM] beta_helix: @HuubR Interestingly, Foldit players have helped the Rosetta community improve its energy function (particularly for protein design). One current example is: https://fold.it/portal/node/2010013
Join the author list on an upcoming research paper
Join the author list on an upcoming research paper
[2:41 PM] beta_helix: But even going back to the earlier days of protein design, there were some very big loopholes in the energy function that Foldit players uncovered.
[2:42 PM] beta_helix: Off the top of my head, the diels-alderase Nature Biotech paper (http://www.cis.umassd.edu/~fkhatib/Papers/NatureBiotech.pdf) had a crazy case:
[2:42 PM] pc: Yes, rosetta is for only one protein if I understand. So, without objectives, in design puzzles, when a recipe try to find correct AA between our protein and target, it is like it search AA between 2 Helixies of the same protein.
[2:43 PM] beta_helix: The helix that came back from Foldit had a ton of TRP residues in a row, one after the other, because the energy function rewarded you for residue-residue contacts.
[2:43 PM] pc: BUNS are good to resolve that problem, but lot of players minimise contact with target (so less buried AA).(edited)
[2:44 PM] beta_helix: It revealed many flaws with Rosetta at the time. The same happened when we introduced Foldit symmetry: for some reason Foldit players were trying out things that the automated Rosetta algorithm would never do!
[2:45 PM] HuubR: I have really no idea how the scoring works, but is it possible that some AAs have an intrinsically higher score than others?
[2:45 PM] beta_helix: Pi-helices are another great example. Rosetta never tried breaking an alpha-helix into a pi-helix... but Foldit players did for sure, and that uncovered a bug that had been in the code for years!
[2:45 PM] beta_helix: @pc how is BUNS working for you so far?
[2:47 PM] beta_helix: @HuubR yes, there were definitely many artifacts in there. It's worth remembering that this energy function was initially based on the know proteins that had been deposited in the PDB. Design throws that right out the window!
[2:48 PM] HuubR: That is exactly my point. Design and prediction are completely different concepts
[2:48 PM] beta_helix: Proof: left-handed helices used to be "impossible" then the Baker Lab made one: https://www.rcsb.org/structure/5KX0.
[2:48 PM] beta_helix: I completely agree with you, HuubR.
[2:48 PM] pc: BUNS is a score that is nice to force players or recipes for making strong bouds with targets. But there is a other metric missing. Because lot of players avoid contact with target to reduce BUNS, and it gives lot if points.
[2:49 PM] beta_helix: That is true, pc... and this is the balance we need to find:
[2:50 PM] beta_helix: providing you with the metrics that we use "after the fact" to check the validity of your models... but not giving you required metrics that slow your gameplay to a halt.(edited)
[2:51 PM] pc: maybe we need a metric that give points when we have the orange part (the core existance objective), near the target (and not only with our protein)
[2:52 PM] pc: there is no Shape Complementarity metric eigther, and players make a lot of voids too in hight score solutions(edited)
[2:52 PM] beta_helix: @HuubR The one thing I learned about proteins when I was in grad school is that "there are exceptions to every rule". When I started: knotted proteins were believed to be impossible (and any reported knots were errors or mistakes). If you submitted a knotted model in CASP, it was immediately discarded. Then a bunch of deeply knotted proteins were discovered and deposited in the PDB... and at the next CASP one of the targets had a knot!
[2:53 PM] alcor29: Re SC and rules. Greg Bowman's work at the U of Washingotn, MO seems to show that the covid19 spike acts like a jaw which closes in on its target. In such a case might designing something which blocks that action work even though it has poor shape complimentarity?
[2:53 PM] brunokestemont: IMAGE: http://fold.it/portal/files/chatimg/irc_447652_1595019200.png
[2:53 PM] pc: interesting alcor29 ^^
[2:54 PM] beta_helix: @pc Great points... and indeed a Shape Complementarity is essential
[2:55 PM] beta_helix: @alcor29 Thanks! Can you send me Greg's work? I'll share it with the rest of the team.
[2:55 PM] alcor29: I'll see if I can find it.
[2:56 PM] beta_helix: My opinion w.r.t covid is that we need to try to tackle this problem from all possible angles, so any promising leads are worth pursuing!
[2:56 PM] beta_helix: Cheers, alcor29
[2:57 PM] Susume: what other takeaways did you have from the conference, beta?
[2:58 PM] pc: To try something very different, maybe we need a simulmation tool. Because we never know if something too different will work. And to make points, we often just make a tripple helix
[2:59 PM] pc: objective are usefull to make limitations too. It is true when a laboratory can't make too big proteins for exemple
[3:00 PM] beta_helix: I think the session was quite eye-opening for the computational biology community. I know they've all heard of Foldit, and maybe a few of the other games there, but to see and hear how successful these various projects have been (and how there are all still going on strong) lead to many positive comments (from regular conference attendees) at the end of the day.
[3:00 PM] pc: But too much limitation can limit creativity sometime too (it is difficult I know ^^)
[3:00 PM] beta_helix: Indeed, it's a fine line and a difficult balance to find, @pc
[3:01 PM] lociOiling: @beta_helix will there be a YouTube of your talk and Seth's?(edited)
[3:01 PM] beta_helix: Great question! Yes for sure.
[3:01 PM] beta_helix: and the organizer's hope was to post all the talks online.
[3:02 PM] lociOiling: we will await...
[3:02 PM] brunokestemont IMAGE: http://fold.it/portal/files/chatimg/irc_447652_1595019739.png
[3:02 PM] beta_helix: @Susume I think one thing I learned after reflecting a bit about the session was how it's unrealistic to expect all these games to have thousands of active players.
[3:03 PM] beta_helix: All of these projects reported ~50-200 players at a given time... so that was nice to hear.
[3:04 PM] alcor29: The link to Bowman's demo on Covid 19:
[3:04 PM] alcor29: https://medicine.wustl.edu/news/foldinghomes-fight-against-covid-19-enlists-big-tech-gamers-pro-soccer/
Washington University School of Medicine in St. Louis
Julia Evangelou Strait
Folding@home’s fight against COVID-19 enlists big tech, gamers, pro...
Over 4 million computers worldwide aiding coronavirus research
[3:04 PM] beta_helix: Awesome... thanks
[3:04 PM] beta_helix: Oh, this is Folding@home! Great.
[3:05 PM] clark92: Hi beta_helix
[3:05 PM] beta_helix: Hi Clark92
[3:05 PM] Susume: I'm glad we got some cred with the mainstream bioinformatics folks
[3:06 PM] beta_helix: For sure
[3:06 PM] lociOiling: (Clark92 started in February, and had a 1st place on the last puzzle....)
[3:06 PM] beta_helix: Congrats!
[3:06 PM] beta_helix: That is awesome
[3:06 PM] beta_helix: I generally get last place on all of them
[3:07 PM] clark92: The link above says there are over 4 million computers, why can't we make them become folders?
[3:07 PM] beta_helix: Clark92, may I ask how you find out about Foldit?
[3:07 PM] beta_helix: Great question, Clark92!
[3:07 PM] Susume: I wonder if it would be productive to have an online meeting of players from all the games
[3:08 PM] beta_helix: Funny story: when Foldit started we got a ton of players that came from Rosetta@home (since that was easy advertising).
[3:08 PM] beta_helix: Most of them tried the game as said "this is way too hard" and went back to donating their CPU hours(edited)
[3:09 PM] clark92: Thanks for the compliment; Foldit went viral on Reddit. That's how I found out about it.
[3:09 PM] beta_helix: Some of those people are still playing Foldit though
[3:09 PM] beta_helix: ahh... gotta love Reddit!
[3:10 PM] beta_helix: @Susume that is a great suggestion!
[3:10 PM] Susume: maybe when the videos of the talks from your session come out, we could have some kind of online conversation (even if it is threaded forum rather than real time) between players from the games
[3:10 PM] beta_helix: @Josh has organized meetings with the devs from EteRNA and Eyewire so far... but that is a great idea!
[3:11 PM] beta_helix: I will bring that brilliant idea up to the head of the session (the creator of Phylo)
[3:11 PM] Susume: actually we could probably host it here on the discord
[3:11 PM] beta_helix: For sure! Discord would be perfect
[3:12 PM] beta_helix: I know the Stall Catchers group offered their platform to host all the videos already as well
[3:12 PM] beta_helix: Well thank you all again for swinging by Foldit Office Hours!

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office Hours 7/30/20- led by Jflat06

[12:59 PM] jflat06: Hey all, we can go ahead and get started
[1:01 PM] jflat06: I don't have a particular focus for this, so we can talk about pretty much anything to do with Foldit development
[1:01 PM] jeff101: hi jeff
[1:01 PM] jflat06: Hello!
[1:01 PM] jeff101: any new ED tools in the works?
[1:02 PM] jflat06: I believe we have recently put out a few improvements for ED. We are also having internal talks (no development yet) about some other improvements.
[1:02 PM] jeff101: what sort of puzzles do you expect to post for the next few weeks?
[1:03 PM] jflat06: The puzzle posting isn't really my area of expertise, but I think we are trying to push for the next round of BUNS filters, using the new devprev which (hopefully) fixes any remaining issues with those
[1:03 PM] nspc:: the coronavirus with new reaction design tool ^^
[1:05 PM] jflat06: We have been holding off on the science side of things for a bit because we were seeing a lot of reports of crashing with the previous devprev, so we haven't been able to use the new filters.
[1:05 PM] nspc:: or maybe adding new metrics for design puzzle ^^
[1:05 PM] jflat06:Yeah, we are definitely working on new metrics (including the BUNS filter)
[1:06 PM] jflat06: we have more metrics in the pipeline, too, but they are very slow, so they will require a rethinking of how we use them to help you fold
[1:08 PM] malphis: h @Susume and I are still experiencing remix crashes in the new devprev on symmetry puzzles. It is a little better than before, but definitely still there.
[1:09 PM] pc: some low computation metrics can be usefull to check if we are in a good way. A player created a DDG recipe this week. This kind of tool can be very usefull in cpp, and usable in lua
[1:12 PM] jflat06: hopefully the metrics we give you will be much faster than the ones people code up in LUA (and hopefully more accurate, too)
[1:12 PM] pc: cool
[1:13 PM] jflat06: The remix crashes theoretically shouldn't be happening, since we essentially undid the changes that were in the previous devprev'
[1:14 PM] jflat06: It's possible our "undo'ing" wasn't completely clean, we can look into it
[1:14 PM] jeff101: please do
[1:15 PM] pc: after a round, I have often 20 saved solutions. I just select the top score of each alternative solutions. but with a low computation tool that give a more accurate score, I can do a better selection for scientist share
[1:15 PM] jeff101: could you raise the # of shares to scientist to say 10 per puzzle?
[1:16 PM] jflat06: We should be able to do that - from a talk we had in a meeting the other day, it sounds like if you just overwrite some of existing scientist shares, we'll still get them all
[1:16 PM] jeff101: in the ligand puzzle 1855, having more shares to scientists would been helpful
[1:17 PM] jflat06: But I think it should be really easy to change the number of shares
[1:17 PM] jeff101: in those sketchbook puzzles with limited # of moves, having more shares to scientists makes sense
[1:18 PM] jflat06: Yeah definitely
[1:18 PM] jflat06: (remember that we also still get all of your solutions, even if you don't explicitly share them)
[1:19 PM] pc: mabe solutions we save in local can be get in priority too
[1:19 PM] jeff101: I often count on that, but I was told that only 5 share w/scientists + best solo + best evo are actively looked at by the scientists ... the rest are like backups for global automated analysis
[1:20 PM] jflat06: From the sound of it at the meeting, it sounds like we still keep around any of the 'replaced' ones, so we actually do look at them all
[1:21 PM] jflat06: I'm taking a look at the client code right now to see if the 5-limit is coded in there, or whether it is completely server side
[1:21 PM] jflat06: I'm pretty sure it's server side, in which case I can easily tweak it without needing a client change
[1:23 PM] pc: some players use external tool to make a protein analisis in design puzzle. Maybe some can be in game
[1:23 PM] jflat06: That's actually the main focus of our development plans for design puzzles right now
[1:24 PM] pc: we need to experiment too. Sometime I ask me question "is it stable," "is the protein move or stay stick?"
[1:25 PM] pc: I know there is no physic simulation in game, but sometime it can be usefull to experiment and discover and try something new
[1:25 PM] pc: For beginers, it is often better to make a safe triple helix because we know it works, but we dont realy see why at begin.
[1:26 PM] HuubR: nspc, you sometimes speak about DDG. Do you then just move the monomer away and check how much the score drops, or is there more to it than that?
[1:27 PM] pc: yes
[1:27 PM] pc: often, the score stay high
[1:27 PM] pc: for some bad solutions with buns, score can be better too
[1:28 PM] pc: this is one of the hack in puzzle design : some top score solutions are very far and not connected to virus
[1:29 PM] HuubR: Would there be any way in which we can make DDG count for binder designs? Or is that just what the devs are working on?
[1:29 PM] pc: I know that, so I try to stay in contact with virus and try to make no additional buns
[1:29 PM] pc: we need first some metrics that fix some score hacks ^^
[1:30 PM] HuubR What exactly do you mean by score hacks?
[1:30 PM] pc::
[1:30 PM] pc: this ^^
[1:30 PM] jflat06: Ok, I've increased the number of scientist share slots to 10
[1:31 PM] pc: cool
[1:31 PM] malphis: I don't think hack is the right word, it's just what foldit and the script we have tend to do.
[1:31 PM] jeff101: :)
[1:31 PM] HuubR: Shall we try it now? :D
[1:32 PM] jflat06: Go ahead! From what I can tell, there shouldn't be a problem
[1:32 PM] MikeCassidy: I am old 'hack' means to me: 'Quick and dirt'; UNIX was originally Quick and Dirty
[1:33 PM] pc: sorry I am french, maybe i dont use the better words ^^
[1:33 PM] MikeCassidy: I always wondered how it got hijack to mean braek ing into computer systems
[1:34 PM] jflat06: That one I think is pretty close to the actual definition
[1:34 PM] malphis: It's just something that isn't restricted by the objectives and scoring functions. In my experience, it is difficult to make bonds at all.
[1:34 PM] jeff101: I use bands to force bonds to form
[1:35 PM] jeff101: h-bonds and disulfides h have "ideal" distances and directions, so often I need 2 bands to form 1 hbond and 3 bands to form 1 s-s bond
[1:36 PM] jflat06: I wonder if we can make a special kind of rubber band that includes constraints on h the distance AND directions...
[1:37 PM] HuubR: When you use bands to create H-bonds, does it help to do that in Stick view, where you can attach the band to the individual atoms?
[1:37 PM] jeff101: the web page for the recipe bandsome gives more details
[1:37 PM] jeff101: stick view and cpk coloring helps
[1:37 PM] pc: sometime I create a triple helix with little hydrophilic sidechains outside, and wiggle in a zone. I test multiple zones and manualy change some sidechain. I mutate, try again.. and sometime it found some good bouds.
[1:37 PM] jeff101: bandsome asks for specific atom #'s to connect
[1:38 PM] HuubR: When I then switch back to Cartoon, the band seems to shift. Is that really so, or is it just the visualisation that changes?
[1:38 PM] jflat06: it should just be a visual change
[1:38 PM] jeff101: I think the visualization just changes
[1:39 PM] malphis: maybe I should have qualified that with as little force as possible. I do place the monomer at what I think is a good docking spot, but then just vary CI, wsm.
[1:39 PM] jflat06: the bands dont even know about the visual representation of the protein - they are using a different complete "full-atom" version of the protein model
[1:39 PM] jeff101: sometimes changing cartoon/stick etc helps see the desired bonds ... often the bands hide the desired bonds
[1:40 PM] jflat06: yeah, we can probably do some work to increase clarity for these types of situations
[1:40 PM] jeff101: if you're going to make a new kind of band, please make a band that is 1-directional
[1:41 PM] jeff101: like a bumper on a car (it only repels when its distance is below a certain cutoff)
[1:41 PM] jeff101: or a spring attached to a long string (it only contracts when it is stretched beyond a certain distance)
[1:42 PM] jflat06: The repulsion should be handled automatically by the fa_rep term in the scorefunction
[1:42 PM] HuubR: Can you push things apart with a band?
[1:42 PM] jeff101: yes
[1:42 PM] jflat06: yeah
[1:42 PM] jflat06: a band is just a "constraint" - it tries to make a value be another value. So if the desired value is greater, then it will "push"
[1:42 PM] jeff101: https://fold.it/portal/node/996372
[1:43 PM] jeff101: 1-directional bands ^^
[1:44 PM] jeff101: foldit bands right now are 2-directional: they contract if stretched too far and they repel if their lengths get small
[1:44 PM] jflat06: I think rosetta in general can use pretty much whatever function you want underneath for determining the constraint behavior
[1:44 PM] jeff101: yes, but we can't code rosetta in LUA
[1:44 PM] jflat06: I'm not sure how hard or easy it would be to have multiple options available in the GUI/LUA
[1:44 PM] jeff101: the FOldit devs could give us more tools
[1:45 PM] jeff101: https://fold.it/portal/node/996372#comment-38712
[1:45 PM] jeff101: above shows some rosetta pages about constraints
[1:45 PM] jflat06: Would be a question for the more rosetta-fluent of our developers. I can ask around
[1:46 PM] jeff101: I think 1-way bands can already be done directly in rosetta
[1:47 PM] BletchleyPark: Is there a programmatic interface for pushpins ?
[1:48 PM] jflat06: The main issue in Foldit would be that we don't currently track anything other than the fact that there is a band, so we'd have to restructure the internal state of our protein model as well as save formats, etc. Depending on how engrained the idea that "a band is just a band" is, it could be easy, or it could be hard.
[1:48 PM] jeff101: giving us view options like show bands / hide bands would help us see if our bands have made the bonds we want. transparent or translucent bands would help in the same way
[1:48 PM] jflat06: @BP I believe that was either added recently, or else is being added soon
[1:48 PM] BletchleyPark: Ok, thank you, how would I know / how can I find out ?
[1:48 PM] Xartos: jeff101: you can simulate 1 way bands with selecting only the part that you want to move ?! or am i wrong?
[1:49 PM] jflat06: If it isn't in the release notes, then it's likely still in the works
[1:49 PM] jflat06: from what I can recall, I think we haven't pushed that one out yet
[1:49 PM] BletchleyPark: How do I get the release notes ?
[1:50 PM] jflat06: News posts are the most reliable source. The new site will have them automatically attached to the releases, so it will be much more organized
[1:51 PM] BletchleyPark: ok, thank you ! I'll be afk in a few minutes, have a great day.
[1:57 PM] jeff101: one thing I've ask for in the past is a version of the Nelder-Meade Simplex Direct Search method implemented in FOldit
[1:58 PM] jeff101: it seems like something that could be coded in LUA, but if it were a built-in FOldit LUA function, that would be great
[1:59 PM] jflat06: I'm not well versed on that, but I'd be curious to see if anyone's put it into Rosetta already
[2:00 PM] jeff101: the post below talks about using it to improve the Align Protein to Density button for ED puzzles
[2:00 PM] jeff101: https://fold.it/portal/node/993959#comment-32465
[2:01 PM] jflat06: The gradient descent algorithms that the minimizer uses are pretty good at what they do. The main thing that looks interesting about the Nelder-Mead is that it doesn't require derivatives, which could be useful for things like our Filters that aren't currently included in minimization (since gradient descent requires derivatives)
[2:01 PM] jeff101: right. NM doesn't need derivatives.
[2:02 PM] jeff101: if NM were built into FOldit, you could use it with xyz coordinates for the atoms to make the Align Protein to Denisty Button work better
[2:03 PM] jeff101: it would be harder for a player to code as a recipe because we don't have access to all the coordinates and density data we would need
[2:03 PM] jflat06: I think what you'd want to do is just set up a 6DOF space with the location/orientation of the entire model and apply NM to that
[2:05 PM] jeff101: I've seen NM work well in optimization problems up to about 20 variables
[2:05 PM] jflat06: Yeah, it might work for alignment
[2:06 PM] jeff101: I don't know how many variables will make it perform worse
[2:06 PM] jeff101: generally more variables takes longer
[2:06 PM] jflat06: the actual optimization space of general folding is significantly more complex, though. thousands of degrees of freedom.
[2:07 PM] jflat06: The main problem with the align-to-density tool is that a lot of proteins (especially helical bundles) tend to look pretty similar even when rotated about their axis
[2:07 PM] jeff101: but for taking the protein as a rigid body and finding the best position and orientation for it to match the ED, you only need xyz for translation/position, and 3 euler angles for direction
[2:07 PM] jflat06: right
[2:08 PM] jflat06: it's a 6DOF space, so it's much more manageable
[2:15 PM] jflat06: Thanks all for showing up! - I am going to go grab some lunch now.
[2:15 PM] jeff101: thanks for coming
[2:15 PM] jeff101: these meeting are helpful
[2:16 PM] jmbrownlee333: Have a nice lunch. I second what jeff says.
[2:16 PM] HuubR: Yes, it was an interesting hour!

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office Hours 8/14/20- led by Sciren and rmoretti

[1:04 PM] sciren: Hello everyone! I hope everyone is having a great Friday so far. As @agcohn821 mentioned I am a developer for Foldit and have been working on Ligand based tool sets for the game.
[1:04 PM] donuts554: hello how are you?
[1:05 PM] sciren: I've never had a virtual office before, but then again I've also never had my own office so I think this works well!
[1:06 PM] sciren: I am doing well Donuts544. Its been a little rainy lately, but I like that. How are you?
[1:07 PM] donuts554: I am good, thanks for asking!
[1:07 PM] donuts554: I have a question
[1:07 PM] sciren: Sure thing!
[1:08 PM] donuts554: Why is there a limited number of choices in each column of the Reaction Design tool?
[1:10 PM] sciren: Let me make sure I understand your question. Are you talking about the "reagent" choices for the tool?
[1:11 PM] donuts554: Yes the reagent reactant choices
[1:15 PM] sciren: Perfect. Each puzzle that we release with the Reaction Design tool will have a varying number of reactant choices. For the latest puzzle we had a specific series of designs that could be made to suit the problem at hand. This will not always be the case. The tool is designed to only display 10 choices at a time, but it can pull from a much larger dataset. When this happens, you will still only have access to a those 10 choices, but there is other functionality to propagate other reactant choices.
[1:15 PM] jeff101: @sciren: have you read any of the feedbacks below?
[1:15 PM] jeff101: https://fold.it/portal/node/2009896
[1:15 PM] jeff101: https://fold.it/portal/node/2009887
[1:16 PM] sciren: I have indeed jeff101:. I actually have them bookmarked and am using them to update the tool.
[1:17 PM] pc: The main problem I had when I used liga the tool is : When I want to change a part of the molecule, it moved all or changed part I choosed. I needed somethink like protein design : base molecule that not move (like a backbone that not move duruing edition), and I can add a new part with a contextual menu (like I choose a new sidechain or add a new backbone part). It is important to keep the other parts in same position to keep the bouds with target.
[1:17 PM] jeff101: that's good. what things have you been working on?
[1:17 PM] sciren: For me, I get to see the inner working of the tool and have stared at it for many many hours. So it is was really amazing to have all of you look at it with fresh eyes so that I could see areas that needed improving.
[1:19 PM] sciren: PC are you saying that what you were wanting was to change a portion of the ligand at a time?
[1:19 PM] jeff101: I like 06pc's idea above. Perhaps make something like the Blueprint Tool with Building Blocks
[1:20 PM] jeff101: or the mutate buttons
[1:20 PM] pc: Yes maybe add parts on the molecule with only contextual menu for the exreminity with selected
[1:21 PM] jeff101: but I guess connectivity is different for small ligands ... they aren't just chains of amino acids ... they can have branches and rings too
[1:22 PM] jeff101: more like a 3-d object than a 1-d object
[1:22 PM] rmoretti: One limitation of the Reaction Design tool is that it's designed for compound libraries where you have a sort of fixed enumeration of reagents. We're hoping to keep the tool close to that approach such that the molecules we get out of it can be synthesized by those techniques. (Which are distinct from that sort of arbitrary modification.)
[1:22 PM] sciren: jeff101: It never occurred to me that having a number or some other sort of representation on the reactant choices would be something that would be beneficial for collaboration. Also adding additional color dependent on which reactant column is chosen was a great idea as well. These are two of the addition/updates that I am working on.
[1:23 PM] rmoretti: That said, we do have another person who is looking at different (better) ways of lining up the old and new molecules such that they "match" better and more directly.
[1:24 PM] rmoretti: (Also, the concept of adding other tools which will allow that sort of targeted modification is something that's on the todo list.)
[1:25 PM] jeff101: I've been playing with 1855 (the Y1 receptor ligand design puzzle) since it ended. It would help a lot to have a tool that counts the # of h-bonds the ligand makes to the protein
[1:25 PM] Josh: I wonder if that's something a recipe could do?
[1:26 PM] sciren: That's a good point Josh(edited)
[1:26 PM] pc: Maybe we can start with a complex base molecule part, and add more simple after with a contextual menu. But yes, it is not a simple chain. For complex branchs, contextual menu or 3d visualisation still possible I think
[1:26 PM] jeff101: it would also help to have a function like GetAtom() to list the atom type (N H O C etc) for atoms #'s on the ligand or other sidechains
[1:26 PM] Susume: rmoretti: when you say compound libraries, do you mean we are choosing from a list of reagents that scientists already know how to make in the lab?
[1:28 PM] rmoretti: Susume - Yes, more or less. There's particular reactions the organic chemists know how to run (across the top) and then there are reagents they have access to (in the columns) and running the reaction is as "simple" as picking the reagents to use in the reactions they can run.
[1:29 PM] LociOilingIRC: @Josh, a recipe can't count ligand bonds, since the chemistry of the ligand is a black box - can't tell C from O or N
[1:29 PM] rmoretti: Different projects will have different sets of reactions and reagents the collaborators will be able to work with.
[1:29 PM] donuts554: From what jeff said, I think the reaction design tool should be more about modifying the bonds for each and every heavy atom
[1:30 PM] jeff101: on my screen right now, I see 3 hbonds giving a bonding score of 14.4-15.1 (it varies) for the ligand in 1855
[1:31 PM] donuts554: Such that there are internal restrictions as to which bonds can form and which bonds can't form/ not allowed to form
[1:32 PM] donuts554: Why isn't this the case?
[1:32 PM] sciren: So that is a different design process donuts554. For this tool, as rmoretti has mentioned, the idea is be able to create ligands that are either readily reproduceable, or at least follow known methods of synthetic production.(edited)
[1:32 PM] jeff101: now I see 18.4 for 3 hbonds
[1:33 PM] Susume: @LociOiling would it useful (once the tool stabilizes enough to add lua functions for it) to be able to ask what element a given atom of the ligand is?
[1:34 PM] sciren: jeff101: I'm not certain about the varying bonding score. It is something that I can look into though.
[1:34 PM] jeff101: @Susume I would say definitely. It would help for amino acids too, like how to tell which type of histidine is present
[1:35 PM] robgee: Just label those reagents and i'll be happy :)
[1:36 PM] rmoretti:: @LociOiling What's the process by which a recipie can get the info for an amino acid? If that consults a Lua call, there's theoretically no reason why that shouldn't also work for a ligand. (If it's using an internal lookup table in the recipie, then it wouldn't be, though I'm guessing it would be possible to convert the table to a lua call, if such a table is generally useful for recipies.)
[1:37 PM] sciren: robgee, well then I hope you will be happen when the tool gets updated!
[1:37 PM] donuts554: Yes sciren, but I mean that following the known methods of synthetic ligand reproduction, there should be a list of available "functional groups" that can be bonded to each heavy atom
[1:38 PM] jeff101: for amino acids, you can use stick view and cpk with a recipe like bandsome to tell which atoms are which. then you make a table, like brow and perhaps others have made
[1:39 PM] donuts554: And there can be some sort of restriction that follows the known methods of synthetic ligand reproduction as to which functional groups can be --added/removed-- to each heavy atom
[1:39 PM] jeff101: but amino acids are generally the same from puzzle to puzzle. ligand puzzle reactants seem to vary more, and how they fit together and get numbered is mysterious
[1:41 PM] Susume: maybe @Josh and @sciren could work together to give us lua better visibility of atoms in h proteins and small molecules - right now I think atom number is the only way we have to access individual atoms, and as jeff says we have to back into knowledge about the atoms from there
[1:42 PM] sciren: I'm not saying that that isn't a possibility. However, currently to be able to design reactions that are reproduceable in the lab requires a lot of chemical knowledge.(edited)
[1:44 PM] Susume: it kinda makes sense to let us demonstrate we can do something useful with a limited set of legos before letting us design our own legos (which would require coding in the rules for how to design valid legos)
[1:45 PM] sciren: That's actually a really great analogy Susume!
[1:45 PM] jeff101: so far in these puzzles, if you pick a certain set of reactants, do you always get the same ligand product?
[1:46 PM] jeff101: I know some things in organic chemistry have probabilities to form different products ... like adding functional groups to rings ..you worry about ortho para and meta (I think).
[1:46 PM] rmoretti: The original ligand design tool allowed much more flexibility in how ligands were constructed. The issue there was that Foldit players took full advantage of the flexibility to create molecules which were ... interesting. That's our (the developers') fault there, as we didn't put in rules which limited things to be reasonable. The problem is that coming up with such rules was harder than we anticipated, so we tried a different tactic with the Reaction Design tool.
[1:46 PM] Susume: I think one column was for chirality - different shapes that can be made from the same reactants?
[1:47 PM] donuts554: Because there is a wide variety of known organic chemical reactions possible, why is there a limited number of organic chemical reactions in the top row?
[1:48 PM] rmoretti: Susume - Right. One of the reagent choices for the first puzzle was either a "D" or an "L" arginine, which were two options in how the library we were targeting was constructed.
[1:48 PM] sciren: jeff101: if you choose the same set of reactants, you will get the same products
[1:50 PM] sciren: donuts544 we were only looking at those particular reaction types, which will be the case for all of the puzzles. Though for puzzles that we are using a much larger chemical library, those reaction types will increase.
[1:51 PM] pc: Yes I want to play lego ^^
[1:51 PM] sciren: Same pc. Same.
[1:52 PM] donuts554: Based on what rmoretti: said about limiting ligands to be reaosnable,
[1:53 PM] donuts554: I think I found a list of rules that have already come up in organic chemical reactions
[1:53 PM] donuts554: And have been put into an online application
[1:54 PM] donuts554: alchem.ie/mechanisms
[1:55 PM] sciren: Where was this when I was taking organic Chemistry??
[1:56 PM] donuts554: Is there any constraint in the Rosetta software that doesn't allow for this, or is this technique as mentioned above possible?
[1:57 PM] donuts554: I also think that this techinique has the portential of being used in FOldit Education mode to teach chemistry as well as to design ligands at the same time
[1:57 PM] donuts554: It sounds very ergonomical
[1:57 PM] jeff101: if you let us pick which item to use in each column using LUA, we can make recipes that automatically go through all the combinations ... like a way to screen the ligands
[1:58 PM] donuts554: and efficient
[1:58 PM] sciren: One of the things that rmoretti also mentioned was that the implementation of these rules proved much more difficult to implement than originally thought. That is reason behind the reaction based approach.
[1:59 PM] donuts554: oh ok, I just thought that a functional-group based approach or a electron-based approach would work well
[1:59 PM] sciren: That is something we can definitely take that into consideration jeff101.
[2:01 PM] donuts554: One thing I like about the reaction design tool is that when it resets your ligand to a default position when you choose new reagents,
[2:02 PM] donuts554: It seems that it eliminates the possibility where there the ligand is inimpossible conformation &/or position
[2:02 PM] jeff101: does that mean if you use the same set of reagents again, you will get the same product formed in the exact same position with respect to the protein and so get the same Foldit score?
[2:03 PM] donuts554: for me, yes
[2:03 PM] sciren: I want to take a moment and thank all of you. Not only for all of your hard work on these puzzles, but also for your feedback. This is invaluable to designing and improving toolsets for all of you. I have also very much enjoyed getting to watch this amazing community of Citizen Scientist do such amazing science!
[2:04 PM] donuts554: You are welcome!
[2:04 PM] robgee: Cheers sciren
[2:04 PM] jeff101: thanks sciren and rmoretti for coming

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office Hours 8/27/20- led by Milkshakeiii

[2:34 PM] milkshakeiii: Hello I'm milkshake and I'm hosting this week's office hours. I can talk about foldit performance and crashing, a topic which I know many are interested in, haha.
[2:36 PM] skippysk8s:k8s: What can we do to fix dremix on the trimers. A lot of us would like to use it for early game, but it always crashes
[2:37 PM] milkshakeiii: We believe there is a crash related to remix itself
[2:38 PM] milkshakeiii: There are a lot of crashes caused by bugs in the code
[2:39 PM] milkshakeiii: Haha
[2:39 PM] susume: have you been able to reproduce a crash using remix?
[2:40 PM] milkshakeiii: Sort of, I think I am starting to get a lead on it
[2:40 PM] milkshakeiii: The reproduction steps for crashes that you have sent us have been really helpful Susume
[2:40 PM] susume: thx, I love to break things
[2:40 PM] milkshakeiii: Haha
[2:42 PM] pc:When we send log after a crash. it looks there is often net very usefull log information. Maybe we need more log, or enable a "more log option" for user who want help?
[2:43 PM] pc: The log file when we restart foldit is cleared by the new one. Maybe we can keep the old after a crash (we can detect foldit didnt closed normaly). and maybe auto send a zip with crash report (with user activated this option ?)
[2:45 PM] milkshakeiii: Unfortunately backtraces get lost on segfault, which means for many of the crashes the log file just terminates when the program crashes
[2:46 PM] milkshakeiii: We do actually get automated uploading of log files, which is nice though, so in many cases you don't actually have to do anything and we will get a report
[2:46 PM] pc:oh ok thanks
[2:47 PM] jeff101:Hi Milkshake,
[2:48 PM] jeff101:I even found Rosetta commands that could be used to implement them.
[2:48 PM] pc: for the last metrics, there needed many software update to have something working. It is strange that is was not tested enough, or there is difficulty to know if something work before ? is it because you need more end game protein to test, or something else ?(edited)
[2:49 PM] milkshakeiii: You mean the new binder metrics like DDG and such, right?
[2:49 PM] pc: yes. but now it works ^^
[2:50 PM] milkshakeiii: Yeah there were a lot of devprev pushes for that
[2:50 PM] milkshakeiii: It was a combination of things
[2:52 PM] pc: maybe there is something to improve in production process. (at my work we have a tester team, its help of course ^^)(edited)
[2:53 PM] jeff101:https://fold.it/portal/node/996372
[2:53 PM] jeff101:above is one feedback about 1-way bands. it lists the rosetta code.
[2:54 PM] jeff101:https://fold.it/portal/node/996372#comment-38712
[2:54 PM] milkshakeiii: 1-way bands, interesting
[2:54 PM] devjosh: having more testers would be great if we had the money for it haha
[2:55 PM] pc: yes ^^
[2:58 PM] jeff101:Milkshake, what aspects of Foldit have you worked on? Does you Math background ever come in handy?
[2:58 PM] pc: for new metrics, so you work on something that compute faster ?
[3:00 PM] milkshakeiii: I started earlier this year and I've worked on small UI changes as well as crash fixes and improving the performance of rubberbands, and some new features for new objectives
[3:01 PM] milkshakeiii: I haven't gotten the opportunity to apply much math to foldit yet sadly but I have gotten to learn about some interesting techniques used in Rosetta for surface area calculations for example
[3:01 PM] MikeCassidytoo: New objectives other than DDG, SASA Score and Shape Comp?
[3:01 PM] jeff101:one way to speed programs is to look for multiplications by zero and rearrange loops to avoid multipling by zero
[3:02 PM] pc: There is lot of same recompute maybe too
[3:03 PM] milkshakeiii: yeah I focused a lot on reducing recomputes during my first couple months
[3:03 PM] pc: cool
[3:03 PM] milkshakeiii: I did manage to eliminate some recomputation related to rubber bands and undoing
[3:03 PM] milkshakeiii: Not sure if anyone noticed though, haha
[3:03 PM] jeff101: the time saved probably got used by another feature
[3:04 PM] skippysk8s:k8s maybe the rubber bands he he
[3:04 PM] jeff101:if you understand the math behind something, you can reformulate it to be more efficient
[3:05 PM] pc: do you keep all solutions database of each puzzle round?
[3:05 PM] jeff101:one professor joked about someone not realizing they could use a sine function for what looked like a complicated program
[3:06 PM] milkshakeiii:I haven't actually touched the solutions database yet, I know that we definitely keep the ones that people click "submit to scientist" on, and even express them in the lab sometimes, that's bkoep's area
[3:07 PM] pc: do you keep the all other solutionsnot just shared to scientist?
[3:07 PM] Formula350: Yea Skippy, band creation in Recipes really are a major processing hog and I could see people not being able to notice any gain, or gain elsewhere, due to that.
[3:07 PM] pc: it is because that can be usefull in deep learning to compute metrics
[3:07 PM] Dudit: faster new metrics using GPU ?
[3:08 PM] milkshakeiii: I haven't looked at GPU yet, have been looking at multithreading
[3:08 PM] Formula350: I also wonder then, if the major delay while tools are running (wiggle specifically), and trying to create bands where it ends up having a delay on mouse click release, is related to refactoring the bands code? :S(edited)
[3:08 PM] pc: with a large database with can apply a fast compute using simple game elements than void, position.. and compare with a real metric, and find the correspondance(edited)
[3:09 PM] milkshakeiii: formula350, I'm glad you noticed the delay on mouse click release
[3:09 PM] milkshakeiii: that's exactly what I tried to get rid of
[3:10 PM] Formula350: Oh, it actually cropped up. Was the same release where the double-click to freeze on Objectives puzzles was fixed. If that helps you narrow it down. (So, roughly 2mo ago)
[3:11 PM]jeff101: do you have a debugging tool that lists the fraction of compute time spent on each line of code. that can be very helpful
[3:11 PM] milkshakeiii: i'll look into it, ideally adding a rubber band should not cause a noticeable delay
[3:12 PM] milkshakeiii: are there any puzzles currently active that you're seeig it on?
[3:12 PM] pc: jeff : there is get date function for that in c++ yes ^^
[3:12 PM] Formula350: I think I linked Josh a bunch of those kind of polling programs back in late-march, Jeff. He passed 'em along to the rest of the team (but as a non-coder, who knows if they were actually usefull or not already known about lol)
[3:12 PM] milkshakeiii: jeff101 yeah we do have that, c++ has a lot of debugging tools which is nice
[3:12 PM] robgee: ollydbg ftw :)
[3:13 PM] Formula350: I'll talk with you about it after your Office Hours, so I don't tie you up with this Bug Talk lol
[3:13 PM] milkshakeiii: roger
[3:14 PM] jeff101:roger roger (as in Star Wars: The Clone Wars series)
[3:14 PM] jeff101:or that movie Airplane?
[3:15 PM] milkshakeiii: lol, I actually love sw the clone wars
[3:15 PM] milkshakeiii: surprised to see it mentioned here haha
[3:15 PM] jeff101:there's supposed to be a new season on Disney+ soon
[3:15 PM] milkshakeiii: that's actually out!
[3:16 PM] milkshakeiii: an epic conclusion with lots of Ahsoka Tano haha
[3:16 PM] milkshakeiii: you should watch it
[3:16 PM] jeff101: gotta have Disney+
[3:17 PM] pc: Do you think we can use some simplier metrics : for exemple just "number of sidechain near target" to have score, instead of real SASA ? or you think it make a real difference to use the real metric?
[3:17 PM] Formula350:And Ahsoka is getting a Live Action appearance in one of SW shows, too
[3:18 PM] pc: a simple metric that increase score with contact with target, avoid the problem that some recipes make the protein far to save some buns for exemple
[3:18 PM] milkshakeiii: these metrics were mostly designed by protein scientists and they are quite advanced from what I understand, if you get a high enough score on those three metrics you can actually be fairly sure you've designed some sort of binder
[3:19 PM] milkshakeiii: i'm not a protein scientist though
[3:19 PM] Dudit: in the latest main Foldit build, the monitor screen did not turn off (even when I set it to turn off after 10 minutes of inactivity)
[3:20 PM] jeff101:maybe a hacked together approximate version f the metrics would give similar values but calculate faster
[3:21 PM] jeff101:like use step functions instead of Gaussians
[3:21 PM] jeff101:or linearize things
[3:21 PM] pc: so, good and fast metrics need scientist and programmer / maths skills in same time.
[3:21 PM] jeff101:Pade Approximation? power series? polynomials?
[3:21 PM] milkshakeiii: i am actually working on something that should make those metrics much less annoying to work with
[3:22 PM] milkshakeiii: at the risk of foreshadowing ha
[3:23 PM] jeff101:multipole expansion? fourier transforms? laplace transforms?
[3:23 PM] Formula350: I personally love their current non-scoring, user-triggered nature ^_^ Alas, I know that won't remain how they'll function lol (which I also understand why they won't)
[3:24 PM] jeff101:Ewald summation?
[3:24 PM] pc: When I play, I somethimes try to ignore the score system, and focus on what happen in reality. but when we use recipes, only score is used for heuristic, and we can be stuck in a bad hightscore solution.
[3:25 PM] Formula350: Time dilation. Einstein-Rosen Bridge. Dodecahedron. SEE Jeff, I can do that too! /sarcasm lol
[3:25 PM] pc: the next puzzle with score in new metrics will help a lot I think ^^
[3:25 PM] milkshakeiii: haha, math name games are always fun
[3:26 PM] jeff101:or for common calculations, make like a graph/table of x and y values and use them instead of the raw functions
[3:26 PM] milkshakeiii: cacheing is indeed a good speedup method
[3:27 PM] Formula350: Yea PC: I hear that! I loaded one of our teams highest score on Corona to see what the Metrics looked like in comparison to my massively poorer scoring solution, and I was surprised that my New Metrics were substantially better <_>
[3:28 PM] pc: you had bas score but new metrics were goods?
[3:28 PM] jeff101:way to go F350!
[3:28 PM] Dudit: is there a minimum system requirement in Foldit for running new metrics?
[3:28 PM] pc: In our better solutions we are very near of the 1500 SASA(edited)
[3:29 PM] Formula350: Wonder if, thanks to the larger L3 cache that CPUs of the last 5yrs have had, that it could somehow be leveraged to speed up Foldit a bit. Maybe store the entire Recipe in the L3 cache? (I'm out of my element here, and just shooting spitballs at the ceiling to see what sticks)
[3:30 PM] jeff101:brainstorming splatter on the walls
[3:31 PM] pc: do you think deep learning can be used to compute faster metrics, using the large database of all previous solutions?
[3:31 PM] milkshakeiii: that sounds a little bit like bitboards, which are also fun speedup techniques
[3:31 PM] Formula350: Dudit, Foldit is a curious beast when it comes to what it can run on, and what it runs faster on. It very much still feels like a 2010 program, and so the better the system you have, reaches its limits for speeding up what we do. (in my experieence, going from a 2015 laptop to my desktop)
[3:32 PM] milkshakeiii: game programming doesn't really work on the level of processor caches thogh
[3:32 PM] pc: Or do you think we can reuse a part of some previous solution,s to help recipe to find the best sidechain for a position
[3:33 PM] Formula350: I personally don't look at Foldit in the "it's a game" sense. I liken it closer to CAD, or Photoshop, or perhaps like Universe Simulator (which is much more borderline "game" but just as much "educational tool" like Foldit)
[3:33 PM] pc: for that I thinked of a feature : In each round we create a database with : some sidechains at a position with a bound with target. And we can use them in some recipes to help find best boud with target(edited)
[3:34 PM] pc: yes fodit is not a game, it is a scientific software, that.. look like a game sometimes
[3:35 PM] milkshakeiii: that's quite true haha
[3:35 PM] Formula350:That's similar to something I'd requested PC:, but actually applies it a bit in a different way. I've wanted a sort of "personal storage" system for portions of Poses, that we can access like Building Blocks, to paste into the game.
[3:36 PM] jeff101:maybe break up the metrics a bit like we vary the wiggle power ... lo wiggle power is faster but doesn't adjust all the bond lengths and angles ... hi wiggle power is slow but varies some bond lengths & angles
[3:36 PM] pc: for my suggestion, the idea is to have a progression in each round. Something we can reuse to have better solutions that work in lab ^^
[3:36 PM] pc: I think in each round end we need a simple "best sasa, best ddg.." score to compare
[3:38 PM] milkshakeiii: that would be useful
[3:38 PM] Dudit: or maybe best total objectives score to compare?
[3:39 PM] Formula350: Maybe being able to see the metrics with a very blurred image of the solution, to not give away details, but still sort of give you an idea of what was needed to attain that (ie, sidechains wouldn't have enough detail to tell which they are). Would still give us feedback, but then still permit design diversity for the science side.
[3:40 PM] milkshakeiii: i think there's a high probability bkoep might post something like that a little later
[3:40 PM] milkshakeiii: but not sure exactly what his plans are
[3:40 PM] Drum: these novice puzzles are actually harder to get a high rank
[3:40 PM] jeff101:regarding calculating the metrics, use cutoffs so that if things are too far apart to give significant scores, give them scores of zero instead of doing the full calculation
[3:41 PM] Formula350: I like that Jeff.
[3:41 PM] skippysk8s: Jeff's suggestion is good. often we build a monomer to get past all the other filters before docking it
[3:42 PM] pc: End round statistics can help us to know what was wrong in last round. In lot of rounds before new metrics, I saw hight score solutions with not a lot of contact with target. With some score détails (like low sasa in most of solutions) we can can understand what we have to do better in next round
[3:42 PM] Formula350: Not just the New Metrics, but the BUNS, too. Yes, for just the reason Skip said. Would let you design further out in space, still have the objectives turned on, but due to distance would not calculate some since it's pointless.
[3:44 PM] Formula350: PC: that's exactly why I liked Skippy's suggestion a couple days ago (I pinged bkoep but didn't hear back), to allow us to load in our personal designs from all past rounds of that puzzle. So we can run the New Metrics on them and see which of them may be viable with tweaks.(edited)
[3:44 PM] pc: yes, the buns number change when I move a very far protein from target
[3:44 PM] pc: I mean, the protein that was already far
[3:46 PM] Formula350: Yea the BUNS filter I personally think, might need its sensitivity dialled back JUST a tiny bit. Being able to wiggle 1 single selected segment and have it cause 4 BUNS on the target is an eyebrow raiser for me. haha
[3:46 PM] jeff101:if several teams contribute to a metric, try to estimate which term will be the largest, calculate that, and leave off the others. this works well if one term is dominant (like 90% of the total)
[3:46 PM] robgee milkshake, do you view the asm when you debug or is that not necessary ?
[3:47 PM] milkshakeiii: gdb is my go-to tool when the simplest methods fail
[3:48 PM] milkshakeiii: viewing compiler output is not good for the brain
[3:49 PM] HuubR About loading previous solutions to check how they score on the new metrics: would it be possible to create a sandbox for that, with all the filters + metrics available?
[3:50 PM] pc:I hope we will have score in new metrics soon. it is realy what its missing to have more chances to have working solution in lab
[3:52 PM] skippysk8s: agreed. One of our team had a nice shape to hook around on a binder. It was a really great shape. Perhaps we can do a contest if we can't get things into the sandbox. Each team can nominate some candidates
[3:52 PM] pc: for you idea huubr, we can create a menu too, accessible everywhere that can run some protein analysis (like metrics).
[3:53 PM] pc: a "protein analisys" menu. So we can open old rounds and run new metrics on them
[3:53 PM] Dudit: are Foldit top solution being used in Rosetta@home ?
[3:53 PM] pc: or of course, if new metrics are accessible in lua, we just have to run a recipe that call one function
[3:54 PM] milkshakeiii: Rosetta@home is used by scientists to design the own "solutions," basically
[3:54 PM] milkshakeiii: their own*
[3:55 PM] milkshakeiii: as well as some other stuff that I don't understand
[3:55 PM] milkshakeiii: they use some fairly brute-forcey approaches, not like foldit players who guide things by hand
[3:55 PM] milkshakeiii: which is why they need all that processing power
[3:55 PM] milkshakeiii: FoldIt actually gets around the need for something like Rosetta@home, ha
[3:56 PM] milkshakeiii: at least that's the way I think of it as someone who doesn't know a lot about it
[3:56 PM] Josh: HuubR About loading previous solutions to check how they score on the new metrics: would it be possible to create a sandbox for that, with all the filters + metrics available? HuubR that's a good suggestion for @bkoep , he might be able to make some kind of sandbox for binding practice, but not sure what you'd be binding to
[3:57 PM] MikeCassidytoo: A sandbox would be good
[3:58 PM] Josh: do you keep the all other solutionsnot just shared to scientist?
@pc: not sure if your question ever got answered. Yes we keep all solutions, and we collect autosaves about once every 5 minutes.
[3:58 PM] pc: A recipe that can call new metrics in lua can be very usefull too, and do the same result
[3:58 PM] pc: thanks ^^
[3:58 PM] pc: cool
[3:59 PM] HMK: Hello to everyone: btw Sandbox, will there be the option to create frozen parts of the puzzle like in 1880 (CV Binder Design) by players?
[4:00 PM] Josh: Hi HMK, I think a binder sandbox is still just a thought right now, nothing confirmed yet
[4:01 PM] HMK: would be great! :)
[4:04 PM] Formula350: While I think a sandbox puzzle would be nice. I think the main thing that would be the most help (to me, I don't know about others; our brains all learn differently lol) is the ability to not only be able to load my pre-New Metrics designs, but also still be able to use them in the Science puzzle to modify. After all, we've put in a week's worth of computational time on them. So having to try and re-create it by hand would be... well impossible heh
[4:05 PM] Formula350: (Which is why it sucks that the one "protein copier" recipe from many years ago that I found the listing for, was blacklisted, as it'd be perfect for this legitimate use. But I understand why it was removed, as it could get leveraged for cheating/stealing designs.)
[4:08 PM] jeff101:if the new metrics give drastically different scores, I think letting us reuse solutions from old puzzles won't skew the rankings a lot ... it will be a bit random how all the old solutions we saved will score with the new metrics
[4:09 PM] Formula350:I agree, because I'm also fully expecting that these old solutions would also, most times, be completely not-viable. It'd be on us whether we felt up to spending that time looking at them though
[4:10 PM] Formula350: (the ???? was a grinning emote, for those in Foldit) Opening another client window while other ones run a recipe, as I look at old designs, just makes for better use of resources.
[4:12 PM] susume: when they do Rosetta testing of foldit designs I think they do use R@h - because they have to run thousands of models to get those "funnel" graphs
[4:12 PM] jeff101:do solutions we find after a puzzle expires get stored on the server? do the scientists ever look at those too?
[4:13 PM] jeff101:I've seen many barrels lately on R@h
[4:19 PM] jeff101:Thanks to Milkshake and Josh for coming today. I enjoy these Office Hours.
[4:22 PM] pc: thanks ^^
[4:22 PM] robgee Thanks josh and milkshake
[4:23 PM] skippysk8s:k8s: yes, thanks
[4:23 PM] Dudit: thank you so much!
[4:27 PM] milkshakeiii: Thank you!

agcohn821's picture
User offline. Last seen 3 hours 12 min ago. Offline
Joined: 11/05/2019
Groups: Foldit Staff
Office Hours 9/11/20- led by Neilpg628

12:59 PM] neilpg628: Hello All, its neilpg628 , ready to answer some questions!!
[1:00 PM] formula350: Hello Neil :D
[1:00 PM] bletchleyPark: Hi Neil
[1:00 PM] LociOilingIRC: hi
[1:01 PM] Dudit: Hello
[1:01 PM] bletchleyPark: Is there news on the paper version of the energy landscape paper ?
[1:01 PM] pc: hi
[1:01 PM] pc:There is many type of buns in the game : on loops when making tripple helix sometimes, in helix, and between player protein and target. Are they the same impact in real protein in lab? If yes, do you think we can need a different score for each?

[1:03 PM] neilpg628: To answer BootsMcGraw's question (I don't know if he is around), BUNS sometimes appear when proteins are rotated because the calculator determining if an atom is buried in the core of the protein is only an approximation, which can sometimes be tricked with a simple rotation.
[1:06 PM] neilpg628: @pc The different types of BUNS could have different levels of importance, but the ones on the interface are probably the most important, because if they are present, there might not be much of a reason for the protein to bind.
[1:07 PM] formula: Just something that came to mind... Would this be something the team might consider: adding a slower BUNS filter that had maybe twice the accuracy (but didn't contribute to points, to keep everything fair for slower system), and which operated like the current "New Metrics" that are Self-Toggled to calculate? My thought was this could at least help us refine our designs by showing which BUNS have a much higher chance of being "real" since they'd disappear compared to the Real-Time filter.
[1:08 PM] formula: (mind you, I know the New Metrics being self-toggled is not the end goal)
[1:08 PM] neilpg628: BletchleyPark, do you mean this one? https://www.biorxiv.org/content/biorxiv/early/2020/07/24/2020.07.23.218917.full.pdf
[1:09 PM] bletchleyPark: Probably, as I cannot open this link in the foldit client
[1:10 PM] formula350: I'll make a small link for oyu Bletch
[1:10 PM] bletchleyPark: The energy landcape paper, where the foldit players are co-authors
[1:10 PM] bletchleyPark: I know where to find it on biorxv, I would expect there to be a paper print version somewhere
[1:10 PM] LociOilingIRC: https://tinyurl.com/y4czb5g6
[1:10 PM] bletchleyPark: thank you
[1:10 PM] LociOilingIRC: there you are
[1:11 PM] formula350: shakes fist LOOCCIIIII!! lol
[1:11 PM] formula350: But that does have Foldit Players as authors. Titled: "Protein sequence design by explicit energy landscape optimization"
[1:11 PM] LociOilingIRC: there's also a forum post from Beta Helix today on another paper
[1:12 PM] bletchleyPark: I know the biorxv is a preprint version, is a journal version expected ?
[1:13 PM] neilpg628: It will probably me published at some time, I'm not sure when.
[1:13 PM] Dudit: Can we have an additional 'UnBun' Action button in Foldit? (similar to UnBun recipe script) ?
[1:14 PM] bletchleyPark: ok, thanks I ask because it refers to the supplemental info which I have not found online yet
[1:15 PM] neilpg628: @Formula350 The problem with a more accurate BUNS filter is that the addition that would make it more accurate is code that would be hard to integrate into Foldit (It's in Fortran, vs Foldit's C++). Including a Fortran executable in the Foldit package could be problematic
[1:15 PM] neilpg628: I have not found the Supplementary Info either, but it might not come out until the paper is fully published.
[1:16 PM] pc: I prefer a more accurate metric than a faster one sometimes ^^
[1:16 PM] formula350: OOoh ok. I thought the filter we had was capable of higher accuracy, but had been toned way down for speed. Thanks :)
[1:17 PM] pc: In the LCB1 puzzle, the solution, that work in lab in spike protein, have 30 BUNS. Maybe there is some false positives, but do you think it means some BUNS can be acceptable ? Or maybe metrics like SASA DDG and SC are good enough to find a good solution with some BUNS ? (solution have 2000 SASA). Is a large contact areas or good DDG allow more BUNS ?
[1:17 PM] bletchleyPark: @Neil I'll await the print version, thanks
[1:17 PM] neilpg628: @Dudit An 'UnBun' action might be possible, but I'll have to get back to you on that.
[1:18 PM] bletchleyPark: #fortran; can't it be rewritten in c++ ?
[1:19 PM] nspc: IMAGE: http://fold.it/portal/files/chatimg/irc_959223_1599851953.png

[1:19 PM] nspc: (a snapshoot of the LCB1)
[1:19 PM] LociOilingIRC: then there's the whole punch cards thing
[1:20 PM] nspc: (I am pc)
[1:21 PM] Jumper2: whatever the language used, just PLEASE tell me you aren't using any code from that horrible Numerical Recipes book
[1:22 PM] neilpg628: @pc Some BUNS could be false positives, but there is a point to be made that just because a particular structure (that we know folds) has BUNS doesn't imply that when designing new structures BUNS should be fine from the start. Some structures might just be exceptions, and a lot of the BUNS on the target look like edge cases that could be solved with the Fortran Code.
[1:22 PM] Dudit: Is it possible to have 0 BUNS on every puzzle?
[1:22 PM] neilpg628: There are also some licensing requirements (I think) that complicate issues.
[1:23 PM] pc: thanks
[1:23 PM] bletchleyPark: ok, thanks
[1:24 PM] neilpg628: @Dudit I can't say it's not possible, but we just want to minimize that value. Also we just revamped the BUNS filter (should be in devprev soon) that will (among other things) allow us to only penalize BUNS on particular regions of the structure. Then we can simply tell the filter to ignore the target which you cannot modify.
[1:24 PM] pc: for the LCB1 we think that some BUNS in game are bouds in lab, and this is just in the game that the fold was not perfect.
[1:25 PM] pc: (but here there is 30 buns, it is a lot)
[1:26 PM] jeff101: can you make Foldit LUA commands like isdonor(seg#,atom#) or isacceptor(seg#,atom#) to help us find donor and acceptor atoms from within recipes?
[1:26 PM] Jumper2: ooh, cool idea
[1:27 PM] neilpg628: @pc All we can say is that structures without BUNS are more likely to fold. That doesn't imply that if it folds it has no BUNS. (Affirming the consequent)
[1:27 PM] bletchleyPark: #quicksaves; can you increase the number of quicksaves to 1000 or so ?
[1:27 PM] neilpg628: @jeff101 Possibly, I am not as familiar with the Lua stuff, but probably Josh would know
[1:28 PM] jeff101: I guess isdonorH(seg#,atom#) would help too ... seems h-bonds need hydrogen atoms to be attached to the donor atoms
[1:28 PM] pc: Ok, so LCB1 is like a solution that is good enough with other metrics to have buns, but it is better without buns
[1:29 PM] formula350: Might be handy to create a Sandbox Puzzle (no Target) with Multi-Start, that has the BUNS filter and all of the Multi-Start proteins are ones that have sucessfully folded in the lab. So we can see how BUNS are on them, to get a rough idea of what MAY be "Acceptable" in our own designs.
[1:29 PM] jeff101: I like this idea Formula350
[1:30 PM] Jumper2: It may also yield clues as to where or in what particular cases buns are not a big deal or are critical
[1:30 PM] formula350: (clarification: the multi-starts would be ANY man made protein, not specifically ones players made)
[1:30 PM] Jumper2: which could then be fed back into scoring algorithms
[1:30 PM] neilpg628: @pc I suppose, but it works, so we don't touch it
[1:31 PM] formula350: That's exactly my thinking Jumper
[1:31 PM] Jumper2: As a teacher of mine use to say: "Great minds think alike, and fools never differ."
[1:31 PM] neilpg628: @pc See this paper: https://science.sciencemag.org/content/early/2020/09/08/science.abd9909
De novo design of picomolar SARS-CoV-2 miniprotein inhibitors
Targeting the interaction between the SARS-CoV-2 Spike protein and the human ACE2 receptor is a promising therapeutic strategy. We designed inhibitors using two de novo design approaches. Computer generated scaffolds were either built around an ACE2 helix that interacts with t...

[1:31 PM] formula350: Jeff, I'll also add your suggestion to the Requests Doc, cuz it is good as well.
[1:32 PM] jeff101: thanks. there is probably a feedback about it already
[1:32 PM] neilpg628: It was just published in Science and details the binders designed via computational methods at the IPD.
[1:33 PM] pc: in the same idea as formula, we can have some objectives available in lua, and can create a recipe to run them in any puzzles. So we can test new metrics, or more experimental or low compute metrics in any puzzles. And of course use them in recipes like "wiggle mutate" to have better heuristic.
[1:33 PM] neilpg628: That sounds like a good idea Formula350. It might be useful/
[1:34 PM] bletchleyPark: #loading and saving times; I notice that loading and saving has become slower in the clients, could this be improved ? #default tracks windows open and default recipe open ?
[1:36 PM] Dudit: I think it is a good idea to add a 'Successful Solution' in Foldit Achievement page whenever a Foldit player solution really works in the lab
[1:36 PM] neilpg628: Either @Josh or @bkoep might be more familiar with the loading and saving issues, but I personally don't know what could have caused it to change.
[1:37 PM] bletchleyPark: more complex structures and more rules perhaps ?
[1:37 PM] bletchleyPark: How about leaving the tracks windows and recipe window at their last position and poses (opened) ?
[1:38 PM] neilpg628: @Dudit That would be a very cool award to win! I think it would be easy to implement. We'll see.
[1:38 PM] Jumper2: On Dudit's suggestion, maybe even splitting it into a set of achievements for creating solutions that got physically produced for lab testing, and the higher achievement of not only being synthesized, but actually performing well in tha lab
[1:39 PM] Jumper2: So the 99 players that got something synthesized for the spike binder would get the lower achievement even though none worked, still leaving the higher level achievement to strive for
[1:40 PM] Jumper2: It's like 3d printing, it could be cool know how many times something you designed got physically produced
[1:40 PM] Dudit:
That's a great idea, Jumper
[1:40 PM] formula350: Bletchley that is how Foldit is designed, but there's a bug in the Options.txt saving right now that is preventing changes from being commited to the file. :\ (Though I suspect you may be asking for a bit more, such as loading of the track that client left off on, too?)
[1:41 PM] bletchleyPark: not necessaruly the latter, just opening the tracks window and recipe window would help already
[1:41 PM] neilpg628: Jumper, that could work as long as we have the awards infrastructure to do it.
[1:41 PM] formula350: At least, with Selection Interface... I assume that the window's location and open/closed info in the file relates to Original Interface as well... But I'm not sure <_>
[1:42 PM] Jumper2: The achievement pages generally deal with a threshhold value being crossed. I'd really love to see an actual count of got synthesized and worked in lab
[1:42 PM] bletchleyPark: I only use original interface, been here a long time ;-)
[1:42 PM] Trigger: Howdy, y'all. I joined the chat late. Is there going to be a transcript made available?
[1:42 PM] pc: yes trigger
[1:43 PM] formula350: Yep, that I know! I figured that you may use that still, afterwards hence my followup. I definitely understand hesitation to use Seleciton Interface... I was there with you. But Selection is so much more powerful that I forced myself to deal with the annoyances it has lol
[1:43 PM] nspc: https://fold.it/portal/node/2009805
[1:43 PM] nspc: you can check here trigger ^^
[1:44 PM] Trigger: Thank you, nspc:.
[1:45 PM] formula350: Neil, more a QoL quesiton that may not be applicable to you: Is it possible to have a way to highlight NEW BUNS that are present after a tool stops running?
[1:45 PM] formula350: I often have to Undo-Redo-Undo-Redo with Shown enabled in order to painstakingly find where the new BUNS(s) are. So having it turn bright-green to pinpoint them, would be great...
[1:46 PM] neilpg628: Jumper, there are achievements like 'The Write Stuff' that are managed separately from Foldit thresholds, so I think it would be possible to give players individual awards for synthesized/folded solutions.
[1:46 PM] pc: I use selection interface sometimes. It can be usefull to change only some sidechains
[1:47 PM] Xartos: oh stumbled into office hour xD hi @all
[1:47 PM] nspc: hi xartos
[1:47 PM] Jumper2: Oh, that reminded me of another wishlist item. Bands are not transparent enough. When using a band to try and create a bond between two atoms, sometimes the band hides the bond completely. I find my self doing the "R" remove bonds, then Ctrl+Z to put them back to see the bonds underneath
[1:48 PM] formula350: *waves* Hellow Xartos. Neil is the "BUNS" man, if you have any questions for him (though, if it's a general on, it probably had been asked so might need to wait till after and we can quote the answer for you)
[1:48 PM] alcor29: Is there any knowledge on how many BUNs does it take to destabilze a given quantity of hodrophobe attractions?
[1:49 PM] formula350: Jumper, are.... You a ME, transported from another multi-verse? I've made that SAME suggestion for the SAME reason!
[1:49 PM] alcor29: *hydro
[1:49 PM] formula350: (I've made so many requests prior to me keeping the doc, that half aren;'t even on it lol)
[1:50 PM] Jumper2: LOL
[1:50 PM] Jumper2: quickly hides the time machine
[1:52 PM] Skippysk8s: I'd like to see the sidechain of the bun go a different color. Half blind lol. But Neil, per PC's question -- should we worry most about buns in the center or near the binding site rather than at the edge? Are there other things you think might help us make good folds?
[1:52 PM] Xartos: main question not regarding buns or stuff... is there any attempt/aspiration to implement multi core support for foldit?
[1:52 PM] Xartos: sry old question
[1:52 PM] neilpg628: @alcor29 I don't think there is a specific ratio, but obviously fewer is better
[1:53 PM] Susume: do you have a sense of how many kJ/mol difference a BUN makes in the interface? it doesn't seem like foldit gives us that info
[1:54 PM] neilpg628: Especially for binder puzzles, BUNS near the interface are more important because in the unbound state, those polars could just bind to water. Hence there is little reason for them to be happy in the bound state when they don't have a polar partner to interact with
[1:54 PM] pc: When we use recipe like GAB, (that move protein and mutate and improve score), we have often a problem : When there is some buns between player protein and target, the recipe often detach the protein a little (and create some voids) and mutate to hydrophobic.
[1:54 PM] pc: It makes highter score, but it decrease the contact area with target ,and make a protein that have less chances to work in the lab. Do you think that add the SASA in score can resolve this problem? or there is something else usefull too?
[1:54 PM] Jumper2: Here's an idea that just popped into my head.. May be crazy, but what if you could create an alignment template from a track (you can do the opposite by creating a track and matching the template in it) It could help with BUNS if you have different tracks that satisfy different BUNS and want to create a hybrid to see if you can combine them (or just make the protein blow up most likely)
[1:55 PM]
formula350: Jumper "Load as Guide"?
[1:55 PM] Xartos: xD sounds fun
[1:56 PM] jeff101: Foldit shows the hydrogen bonds between atoms, so it must know how binds to who. Can you make LUA commands to list which seg# and atom#'s are connected by hydrogen bonds?
[1:56 PM] Jumper2: not sure about load, I was just thinking that working different tracks to attack different areas of the puzzle it would be nice to be able to combine the tracks into a possible final answer
[1:56 PM] jeff101: Also, on ligand puzzles, it would sure be nice to know how many hydrogen bonds go from the protein to the ligand. I often find myself counting them by eye, which is kind of tedious.
[1:57 PM] neilpg628: @Xartos Multicore support would be very useful, but might not be worth it, especially for Monte Carlo actions. See here: https://www.rosettacommons.org/content/abinitiorelax-using-multiple-cores-single-machine-without-mpi
[1:57 PM] bletchleyPark: @Xartos, you can run multiple clients each on their own core in a multicore machine
[1:57 PM] Jumper2: jeff: I find myself doing that too, though if available in lua, could have a quick recipe report it for you
[1:57 PM] neilpg628: tinyurl.com/y3f3enlp
[1:57 PM] formula350: @neilpg628: This sort of piggybacks off of Skippy's Q: If there's one mistake (or good thing) you keep seeing in all our results that you'd like us to be aware of so that we can stop (or continue!) doing them... What would it be? :P
[1:58 PM] Jumper2: For multi core, I was thinking of not so much speeding up computation, but for smoothing out the GUI interaction of moving the protein on the screen
[1:59 PM] neilpg628: @Xartos BletchleyPark's method is the only real way to do it. It's described in the website as well, but it's not the ideal way
[1:59 PM] Dudit: or maybe iGPU support?
[1:59 PM] Xartos: @BlechleyPark @neilpg628: doing so
[2:00 PM] bletchleyPark: I do so
[2:00 PM] bletchleyPark: have done so for many years
[2:00 PM] jeff101: The Bonding subscore gives a clue about which segments are involved in hydrogen bonds, but it doesn't say how many hydrogen bonds are made.
[2:01 PM] formula350: I think that's less an issue of Foldit being quite serialized still, but perhaps scheduler/event related? The UI waits for a certain thing to occur, and then updates. Though it could be that it's indeed handled by the same game_library thread that schedules other actions, and is indeed overworked. *shrug*
[2:02 PM] bletchleyPark: @jumper; this is already done, one thread is dedicated to GUI, the other to calculations for the main program
[2:02 PM] Jumper2: Shouldn't the bonds existing be part of the pose?
[2:02 PM] jeff101: 06pc, one way to keep the binder near the target is the recipe BandOntoSphere. Using bands with trget length 0 and strength 0.1-0.3 seems to work well.
[2:03 PM] neilpg628: @Dudit Again, that would require changes to the underlying Rosetta code. Rosetta doesn't have a lot of protocols that are open for the parallelism that a GPU offers, and I don't think anyone has tried incorporating CUDA or OpenCL into the project.
[2:03 PM] pc: thank jeff
[2:03 PM] pc: oh it is a jeff recipe
[2:03 PM] jeff101: yes
[2:04 PM] bletchleyPark: One way to speed up GUI is to disable filters while moving things around, then re-enable after positioning
[2:04 PM] formula350: It'd have to be OpenCL for the sake of universally being supported. (There's tttons of tools to port over C++ to OpenCL code available. AMD offers a ton of them for free on their Developers page)
[2:04 PM] Jumper2: I think the only possible reason for incorporating CUDA or OpenCL into Rosetta would be running on GPU farms (like certain instances that are available in the Amazon AWS cloud)
[2:05 PM] Jumper2: Granted in the supercomputing crowd, that might be appealing enough to find funding for the effort
[2:05 PM] formula350: Bletchley, while true, that's also only trying to excuse the underlying problem by coming up with work-arounds lol Really, there just shouldn't BE the level of hitching going on with the UI as there currently is.
[2:07 PM] bletchleyPark: I speak for myself, it works for me
[2:07 PM] Skippysk8s: @neilpg628, do you have advice about what is better for binding? A long connection with a bun at an edge, or a short connection that is bun free? Often we design limited interface proteins to minimize buns
[2:09 PM] bletchleyPark: I'm off to bed. I'll read up in the transcript, later, thanks Neil !
[2:09 PM] bletchleyPark: bye for now
[2:10 PM] argyrw: hi
[2:10 PM] formula350: Thanks Neil!
[2:11 PM] formula350: Oh I mistook Bletchley's message LMAO
[2:11 PM] neilpg628: @Skippysk8s I think shorter connections are better in general, because Rosetta has more difficulty predicting long loops because of the huge number of possible conformations. However, obviously too short is bad too.
[2:11 PM] formula350: IGNORE me :>
[2:14 PM] neilpg628: @Formula350 OpenCL is possible of course, and the tools do exist, but
[2:15 PM] pc: @neilpg628: When we use recipes in design puzzles, they often make a lot of hydrophobic near the target. And it is not good for the protein when it is alone in lab. Do you think we need a metric that allow hydrophobics in player protein only when : they are only inside the protein (indide a tripple helix) or near an other hydrophobic in the target ?
[2:15 PM] neilpg628: 1. Rosetta in general is not written to take advantage of mass parallelism. This is mostly because of the Monte Carlo usage, which would require most parallelism to only be don't at the highest level
[2:16 PM] argyrw: :)
[2:16 PM] neilpg628: 2. Adding OpenCL support would require the support of the entire Rosetta community (as Foldit is built on top of Rosetta) and not just us
[2:17 PM] argyrw: how you take the points of the hydrogen bond network
[2:17 PM] argyrw: ?
[2:17 PM] argyrw: I will broke it
[2:17 PM] argyrw: :)
[2:18 PM] neilpg628: Also, @milkshakeiii will be adding support for asynchronous filters soon, so the UI thread can run without blocking on filter calculations
[2:19 PM] formula350: Thankfully I'm finding stuff on Parallelizing Monte Cartlo Simulations (not by "some dude on the internet" either lol Respectable places like Oak Ridge Nat'l Labs). Using "OpenMP" was what they suggested, which was something I'd made to Josh at one time. Along with M$'s AMP
[2:20 PM] formula350: Having the UI run smoother, which it sounds like will be done with Hentry's work, will be a huge QoL improvement! :D
[2:20 PM] formula350: Henry's*
[2:21 PM] neilpg628: With regards to parallelism, I think the fundamental fact is that it would require more than just the Foldit team to agree to such a thing. There are likely other reasons for why it's not common that I'm unaware of
[2:22 PM] formula350: Yea I can recognize and appreciate how much of it is due to what can't be touched (on Foldit's end) in regards to being Rosetta code :(
[2:23 PM] Dudit: foldit multithreading?
[2:23 PM] formula350: (I'll secretly blame Moretti lol)
[2:25 PM] neilpg628: Also in some ways it might be an even harder problem on the Foldit end because we have to work on far more combinations of OS/Computer/GPU than Rosetta has to
[2:27 PM] argyrw: when you end tell me to speak to you if you want
[2:27 PM] argyrw: thank you
[2:28 PM] neilpg628: Yeah, I'll probably end now! Thanks for coming! (I'll answer one more question if anyone has one)
[2:28 PM] formula350: That's why it's important to select a universal code set, like OpenCL, OpenMP, and like is already used: OpenGL (for graphics). Then target the vendor with the fewest supported instructions (probably Intel with their older GPUs having minimal OpenCL). Integrate from there as best as possible. But that's why I think trying to parallelize the Objectives, if possible through OpenCL, would be the best 'start'.
[2:28 PM] formula350: Thanks Neild. I gained lots of insights :D

Sitemap

Developed by: UW Center for Game Science, UW Institute for Protein Design, Northeastern University, Vanderbilt University Meiler Lab, UC Davis
Supported by: DARPA, NSF, NIH, HHMI, Amazon, Microsoft, Adobe, RosettaCommons