Developer Chat

@inkycatz Hi everyone! 13:00
@inkycatz Let’s get started. :) 13:00
@inkycatz SCIENCE CHAT time. 13:00
@inkycatz (in all time zones, yes.) 13:00
@inkycatz A few quick housekeeping items. First of all, chats are logged, so if you have to go don’t worry we post the log afterwards 13:00
@inkycatz 13:00
@inkycatz for all your previous chat needs and future chat needs 13:01
@inkycatz You should bookmark it if you haven't 13:01
@inkycatz As always, thanks jflat for the logging! 13:01
@inkycatz Additionally, our Demographic Survey is open until 8 October: 13:01
@inkycatz Please tell your teammates and whatnot 13:01
@inkycatz It helps us the more people fill this out 13:01
@inkycatz Thank you to everyone who has participated so far. 13:02
@inkycatz Today’s topic and question list is pretty full, with seriously multipart questions. Please be patient while we type! :) 13:02
bkoep Hi everyone! 13:02
@inkycatz I see we have bkoep and free_radical today! hi guys 13:02
free_radical hi! 13:02
@inkycatz Since the questions are really multipart - feel free to give me a “done” when you’re done 13:02
@inkycatz so I can move onwards 13:03
@inkycatz first one’s for you, free_radical! 13:03
@inkycatz "I'm curious how the new Drug Design tools are progressing. Could there be puzzles this fall? Will those puzzles be a new category or something else?" 13:03
free_radical I can take this 13:03
@inkycatz (I sure hope so!) 13:03
free_radical The drug design tools are coming around just great! We have tested them internally and we have just some slight additions to make 13:04
DrumpeterIRC yay i made it :D 13:04
free_radical that being said, the first drug design puzzles will be released this fall 13:04
free_radical we will start off slow with some very simple tools, then we will slowly add new tools. 13:05
free_radical there are also tutorial levels associated with each tools 13:05
free_radical The puzzles will be a new category, but not a separate game 13:05
free_radical done 13:06
BletchleyPark please also train us better in H-bond making 13:06
@inkycatz sounds like a blog request to me! 13:06
@inkycatz Next up: 13:06
BletchleyPark done 13:06
@inkycatz "I am particularly curious if the top scoring puzzles with disulfides show much diversity. Or do the top 10 or 20 scores all converge on the same disulfide ensemble? Has there been any case recently where a non-native ensemble of disulfides has scored very well?" 13:06
DrumpeterIRC i came up with a question this morning. am I allowed to ask here or are we only doing posted questions? 13:07
@inkycatz we’ll do the posted questions first 13:07
@inkycatz it’s only fair to those who got in them in earlier :D 13:07
@inkycatz don’t worry I’ll open the floor for more when I run out 13:07
bkoep This is a good question, and I haven't looked at this specifically 13:08
DrumpeterIRC it's been awhile, so i wasn't sure. thanks. 13:08
bkoep In revisited puzzles, I think most of the best-scoring solutions tend to have the same disulfide arrangement 13:09
vmulligan There's definitely diversity in the disulfide bonding patterns that players come up with.  That being said, Foldit and Rosetta (the protein folding software that underlies Foldit) don't do a great job of scoring how good a particular disulfide bonding pattern is.  The scoring function is looking at the geometry of the disulfide (is the bond length between the sulphur atoms appropriate, are the bond angles good, etc.), not at the location of disulfide 13:09
vmulligan (*I was thinking more about total de novo puzzles.) 13:09
bkoep Yes, in design puzzles we certainly see everything 13:10
vmulligan Done! 13:11
@inkycatz “"Why there is a 3D structure in the revisit puzzles?/Where do the starting 3D structures for the revisit puzzles come from?  Why not a kind of starting flat de novo each time?” (Speaking of revisited…) 13:11
bkoep Right 13:12
bkoep The short answer is, of course, that the original puzzle started with that structure. We try to keep the starting point the same when we "revisit" an old puzzle. 13:13
frood2IRC perhaps the ‘game’ was simply simpler then? 13:14
bkoep Really, these starting structures come from random perturbations of the native structure. 13:14
@inkycatz That makes sense. 13:16
@inkycatz Ready for the next one? :) 13:16
bkoep Often, the starting structure is pretty close to the native structure in a revisited puzzle. From a benchmarking perspective, these puzzles are less informative about how broadly players are searching fold space, and tell us more about the score function close to the native structure. 13:16
DrumpeterIRC I figured they were from PDB entries or something 13:16
bkoep (Not so much, "Can the players find the native?"—rather, "Do structures closer to the native score better?") 13:17
bkoep (done) 13:17
@inkycatz "We sometimes get puzzles from the past that really are dimers, why not give those as a dimer? And I am also wondering if the unsolved puzzles we get are sometimes dimers, like 1138?" 13:18
bkoep As I mentioned above, it's important to us that we don't make too many changes to revisited puzzles. So if the original puzzle used a single chain, then the revisited puzzle will as well. 13:19
vmulligan Perhaps some of those puzzles were originally posted before the symmetry machiery (which allows dimers and trimers and whatnot) was in place. 13:19
vmulligan (Done.) 13:20
bkoep Unsolved cases, like 1138, can be more interesting 13:20
bkoep Often times we don't know whether a protein with unsolved structure forms oligomers 13:21
bkoep I think this has been brought up before, but predicted contacts can be especially ambiguous about oligomeric state 13:22
DrumpeterIRC yea, 1138 intrigues even my bio major friends lol 13:23
@Madde you mean a contact could really be between different chains? 13:24
vmulligan Yes.  That's always a possibility when coevolution-derived contact data. 13:24
DrumpeterIRC I've seen that happen quit often. 13:25
@inkycatz These next two were submitted in the same post so I’m going to bundle them together. After these two, we’ll open the floor for questions, starting with Drum (since you asked nicely and waited patiently :) 13:26
@inkycatz "When predicted contacts are generated initially, and they are generated for one subunit of an N-mer, how likely is it that 'odd' looking contacts belong to an interface between subunits, rather than an interface between different sections of the same subunit?" 13:26
@inkycatz ALSO 13:26
frood2IRC this may nor be the tine to ask - but can we have a known native as a contact puzzle? I think this wiyld help newbs 13:26
@inkycatz "Since the contact predictions you give us are based on co-evolution, it seems reasonable that interface contacts would get picked up by the prediction algorithm just as readily as intra-monomer contacts. Is there any way for us to tell them apart?" 13:26
@inkycatz We’ll get to that question-suggestion after Drum’s. :) 13:27
DrumpeterIRC sorry. I was reading 13:27
bkoep Right, as vmulligan said, we can't tell from sequence covariation analysis whether a contact is intra- or inter-chain. 13:28
DrumpeterIRC how would I put my solution into a pdb file? 13:28
vmulligan Ah -- yes, that's juts what we were talking about.  Yes, when we generate predicted contacts from coevolution data, we don't know whether they represent intra-subunit contacts, inter-subunit contacts, or other things that could be reflected in the coevolution data.  However, the most common contacts are intra-subunit, since even in an oligomer, there will be more intra-subunit contacts than inter-subunit contacts. 13:29
vmulligan The hope is that if we try to maximize the number of satisfied contacts when building models, the few that are unsatisfied will end up being the inter-subunit contacts. 13:29
vmulligan But no, we don't have a good way of knowing a priori which are which.  For an unknown protein, we often don't even know its oligomerization state in advance. 13:30
vmulligan (Done.) 13:31
BletchleyPark is there a link between teh type of sidechains and oligomerization ? 13:32
BletchleyPark probability ? 13:32
vmulligan Not really overall sequence composition.  Oligomers tend to have surface-exposed hydrophobic patches on the monomer that facilitate oligomerization. 13:32
@inkycatz cool, we’re in open question mode, unless bkoep had more to add to those… drum wants to know about PDB files, frood’s question above… and anything everyone else has is fair game 13:33
vmulligan But to identify a surface-exposed hydrophoibic patch, you already need to have an idea of the folded structure of the monomer. 13:33
vmulligan (Done.) 13:33
BletchleyPark that's a hint already :) 13:33
BletchleyPark thanks 13:33
@inkycatz (as long as it’s science, remember our topic here.. bug reports to feedback, more ‘feature suggestions’ to the forums please :) 13:33
@Susume2 in a family of homologs, is it common for some to be oligomers and some not? 13:34
vmulligan Not uncommon, though it depends on how closely related the homologues are. 13:34
vmulligan Often very close homologues share oligomerization state.  More distant relatives alter oliogomerization state faster than altering fold. 13:34
vmulligan (done.) 13:35
DrumpeterIRC is the irc foldit site going to be running anytime soon? 13:36
bkoep @drumpeter, right now there is no way for players to write Foldit solutions to a file in PDB format 13:36
jmbrownlee333 Is 1138 being worked on? Or is ir just some random ORF? It seems like it should be tractable. 13:36
bkoep @frood, I don't think we have a Beginner puzzle with predicted contacts. Maybe that would help? 13:38
vmulligan There's at least one paper on 1138, but it's analyzing the gene function, not the structure of the protein.  (Done.) 13:38
DrumpeterIRC maybe we can convert our solutions to PDB format in the future?   13:39
frood2IRC bkoep:  maybe - but players well past that are struggling 13:39
DrumpeterIRC I would love to read that paper :D 13:39
vmulligan Hold on -- let me find the paper. 13:40
frood2IRC if nothing else it would be interesting to see if vets can reah the native via a contact map 13:40
spvincent Why does a serine triangle have an HBNet score of 100% rather than 66.67%,  given that it has one donor and two acceptors (first raised by Susume in blog comments: I think the question may have slipped under the radar) 13:41
bkoep @frood, I would be interested to try that. Of course, the problem with using published structures is that we have no way of preventing players from referencing the native structure. 13:42
bkoep @spvincent, good question 13:42
frood2IRC agreed - but give them the chance huh? 13:43
@Madde maybe a solved but not yet published structure 13:43
frood2IRC put a note at the start….they want to understand - trust them….take it outside of the game score by all means 13:44
jmbrownlee333 It would also tell you if the filter bonus is too generous / or not enough. Even if the structure is published. 13:44
bkoep The scoring mechanism for HBNet still has some quirks. Right now, I think a polar atom only counts as "unsatisfied" if it makes no hydrogen bonds (?). 13:45
bkoep This may not be true for atom types other than Ser/Thr -OH group, I'd have to look into it. 13:46
@Susume2 so if one atom could make multiple bonds, and it makes one, it is marked satisfied? 13:46
vmulligan Sorry for the delay.  Here's the paper about the 1138 protein: 13:46
DrumpeterIRC can a light weight contact override a heavy weight contact? 13:46
bkoep For serine, anyway, the hydroxyl group is a much better H donor than acceptor. I think it is not uncommon to find buried serines that make just one hydrogen bond. 13:47
frood2IRC I would love to see a view that shows the unsatifieds in HBnet 13:47
frood2IRC like exposeds - if u see what I mean 13:48
bkoep @Madde, we would love a solved, yet-to-be-published structure with good predicted contacts. Unfortunately, those are extremely difficult to come by. 13:48
vmulligan @frood:  We're actually not great at computing unsatisfied hydrogen bonds in Rosetta.  We have ways, but they're not quite where they should be.  It's something actively being developed. 13:49
DrumpeterIRC try crystallography :D 13:49
frood2IRC kk - thx vm 13:49
BletchleyPark I find the real-world hbond puzzles very different in bonding behaviour than the sample puzzles we had, can this please be improved with different training puzzles and better explanations of what bonds to what in  real-life puzzles 13:50
vmulligan (Part of the problem is that it's a fundamentally non-pairwise decomposible problem.  If A and C have the potential to form a hydrogen bond, C needs to know about the existence of a hydrogen bond between A and B.  The algorithm that underlies the "shake" and "mutate" functions is called the packer, and it relies on pairwise decomposibility to try many combinations of side-chains very quickly.) 13:50
vmulligan (That was in response to @frood.  Done.) 13:51
frood2IRC thx VM - brai9n spasm continuing  LOL 13:52
@inkycatz Do we have any more questions? We’re about out of our official time. :) 13:54
frood2IRC thx BK and VM :) 13:55
frood2IRC and inky of course 13:55
BletchleyPark I just asked a question 13:55
DrumpeterIRC no worried. Ill stop time :D 13:55
@inkycatz hahah :) no problem, just wanted to get a last call in on questions 13:55
jeff101 the contact map on a solved protein could be a contest 13:56
jeff101 if we do really well on a puzzle and want to make a "trophy" for ourselves, could we ask you for the pdb file so we can use it to 3D print our structure? 13:57
BletchleyPark They will print it for you, Boots McGraw has one 13:58
@inkycatz Sorry BP - thought it was more of a suggestion but I’m sure the explainations can be addressed in future blog posts. Thanks for the idea. :) 13:59
bkoep @Bletchley, are you referring to the first HBNet puzzles we used, with fixed backbones? 13:59
BletchleyPark Please do because you're missing out on better designs this way 13:59
BletchleyPark @bkoep, the trial puzzles we had in devprev in comparison to the posted game puzzles 14:00
BletchleyPark I did pretty well on the trial puzzles which formed bonds real easily, bu the puzzle ones hardly make bonds 14:00
DrumpeterIRC we can get our solutions in a pdb file? 14:00
BletchleyPark in fact the combination that worked well on the trial puzzles barely work in the puzzle versions 14:01
bkoep @Bletchley, this could be a couple things 14:01
BletchleyPark so more explanation and training would be good, i'm sure im not the only one 14:01
@TimovdL Would it not be better to make 2 abeta puzzles with each 4 starting points. Now it looks like one starting point is superior to all others so we will not explore all in the same depth 14:02
jmbrownlee333 No Bletchely, you are not. 14:02
BletchleyPark @timo, most of them score similarly 14:03
BletchleyPark @jmbrownlee, thanks 14:03
vmulligan If one starting point is better than the others, that's fine -- we'd like you to focus on whatever works, and disregard whatever doesn't. 14:03
vmulligan We don't know in advance what's best.  By having one large puzzle rather than many small puzzles, we have a better chance of giving you something that will work.  (A puzzle with no good solutions wouldn't be much fun, or much use.) 14:04
@inkycatz That’s a good way of looking at it. 14:04
BletchleyPark are you seeking a sidechai sequence rather than configuration for 1137 ? 14:04
jeff101 @Drum, in 2014's CASP, the teams that picked their own solutions to submit received about 30 pdb files for each puzzle of the best structures their team made 14:04
@TimovdL So not really interested in more variation? 14:04
vmulligan In the early stages, we just wnat to get a hit, which doesn't necessarily mean pursuing every avenue to completion.  With our automated algorithms, we try to eliminate dead ends as quickly as possible, and I hope that players will do the same. 14:06
vmulligan @Bletchely: We want both. 14:07
BletchleyPark thank you, working on all models 14:07
vmulligan Ultimately, we use the sequence to make the protein, but we try to rule out designs not worth making based on the predicted structure. 14:07
vmulligan If a particular sequence yields huge voids in the core, or very poor shape complementarity to the Abeta, we'll focus on other designs. 14:08
DrumpeterIRC I was told i cant convert my solutions to pdb files. maybe a better word is submit. 14:08
jeff101 @Drum, maybe if you ask bkoep or others after a puzzle, they can e-mail you certain pdb files that you made during a puzzle 14:10
@TimovdL Why a 2 component design, do scientists think it is needed, or will they be part of a bigger structure? 14:10
DrumpeterIRC oh ok. thanks :D 14:11
@MikeCassidytoo From what I have seen in scoring HBond puzzles seem to want to be helices is that real or just a problem in scoring? 14:11
vmulligan @Timo:  We have wet-lab methods of recombining the two halves and screenig in a high-throughput manner.  So if we have 100 good designs, we can create 10,000 variants from the two halves.  It increases our shots on goal. 14:12
@TimovdL That is really nice to know 14:12
jeff101 thanks for having this Chat and going a bit overtime 14:16
@inkycatz Yes, thank you everyone for coming :) 14:16
@inkycatz And to bkoep and vmulligan (and free_radical who I think had to go) 14:17
vmulligan Cheers! 14:17

Generated by 2.13.1 by Marius Gedminas
- find it at!

User login
Download links:
  Windows    OSX    Linux  
(10.12 or later)

Are you new to Foldit? Click here.

Are you a student? Click here.

Are you an educator? Click here.
Social Media

Only search
Other Games: Mozak
Recommend Foldit
Top New Users

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