Developer Chat

@katfish Good morning, everybody! 10:59
ComputerMage Have anyone ever saw the native protein that has lone sheet that is not connected to other sheets? 10:59
TimovdL Just had dinner, evening for Europe 10:59
Susume2 good morning katfish 10:59
ComputerMage Good Morning Kat, jflat! 10:59
auntdeen2 hi katfish and all :-) 10:59
@katfish Then good evening to you, Timo. :) 10:59
@katfish Everybody ready for our devchat? 10:59
ComputerMage hopefully ;) 10:59
TimovdL cant wait 11:00
ComputerMage what will be the topic? 11:00
@katfish Well thanks for being here! We'll get started in just a second. 11:00
Susume2 I added a cople questions to the devchat post a few minutes ago; not sure if I was too late 11:00
@katfish Alrighty today we've got Seth, Jeff, Tamir, and me (Katfish) here 11:00
ComputerMage Hey Seth, Tamir! 11:03
@tamirh Hello everyone 11:04
TimovdL hi tamir 11:04
@SethCooper hello 11:04
@katfish It seems like most of the responses to our announcement about the devchat are filter-related 11:05
@katfish So let's start out with that 11:05
alwen Filters: super slow, even slower than trying to fold on netbook 11:05
auntdeen2 and computer resource intense 11:06
TimovdL Okay, one of the issues is that the pose scoring and evo scoring are not remembering if filters are on or off 11:06
@tamirh yes, we're working on modifying the algorithms for some of the more cpu intensive filters to have them run in realtime 11:06
ComputerMage brow actually did some modification to DRW which looks quite good and quite fast 11:06
brow42 also I saw that the filters seem to be what is spamming the log.txt 11:07
ComputerMage that disables filters for wiggling and then reenables them 11:07
Madde hmm, not for me - when I run one filter puzzle it's unbelievable slow but it still only uses 25% of my CPU 11:07
brow42 my modified script acts on the *assumption* that wiggle doesn't use the filter information...it may not be true. And the restore recent best problem timo mentioned still exists. 11:08
@tamirh if you're curious about what is taking the most time, the 'core exists', 'polar core', and 'layer design' filter all rely on determining if each residue is in the 'core', 'boundary', or 'surface'. to do this it calculated the surface area of the protein which takes quite a while 11:08
@SethCooper i think that the option for turning filters off will only be a temporary fix 11:08
@SethCooper really what we want is filters that run fast enough that it's not needed 11:09
@SethCooper so, we're looking at ways to make the layer design filter faster 11:09
TimovdL The glycine problem might be solved if on design puzzles bonding with a glycine is not counted, (still bonding but no score for that) 11:09
ComputerMage I think it can be done on distanses from the center 11:09
auntdeen2 question - if filters are turned on, why is something that is not mutating so slow also - like a blue fuse? 11:09
@tamirh TimovdL, I saw those feedbacks. We're looking into those. 11:09
alwen Turning filters off doesn't help us when turning them back on drops you back below the starting score 11:10
ComputerMage just calc the farthest point from the center and then to use them as a reference 11:10
brow42 the filters seem to be executing every mini-iteration of wiggle, so wiggle is unusable. 11:10
ComputerMage I did something in lua script 11:10
brow42 (that's the problem with blue fuze) 11:10
ComputerMage farhtest point = farthest segments ;) 11:10
@jflat06 the problem is that some proteins have multiple core regions, so distance from center isn't reasonable 11:10
ComputerMage something = something like that 11:10
vakobo there is AreConditionsMet lua function but no function about penalties 11:10
auntdeen2 (thanks, brow) 11:10
@SethCooper auntdeen, filters get recomputed every time the protein changes 11:10
ComputerMage I see. 11:10
@SethCooper even if it's not mutating, what is "core" or not might change 11:12
auntdeen2 Seth - so every movement prods the filters, then - thanks 11:12
auntdeen2 we are all really happy that you are working on speeding them up 11:13
TimovdL To evo scores should only be on scores with filters on. 11:13
brow42 unfortunately, many people have abandoned this round of puzzles. 11:13
auntdeen2 agree, Timo - but the very best / credit best bug makes it difficult, too 11:14
homme-proteine because ? 11:14
ComputerMage It's painful even on speeded up i7 boxes ;~( 11:14
auntdeen2 homme - you have walked into a dev chat 11:14
TimovdL Those ones should be splitted in 2 filter on/filter off 11:14
TimovdL Same for recent best 11:14
brow42 Also please be sure that scores are fully recalculated before the filter-on/off bit is set 11:14
auntdeen2 devs - why is credit best higher than very best? and if not saved, often lost? 11:14
ComputerMage Good question! I was pd-off quite oftem because of that or when my computer rebooted by BSOD :) 11:14
@tamirh auntdeen2, we're looking into it 11:14
brow42 It's hard to catch, but there has been some weirdness about having the wrong score. 11:14
auntdeen2 tamirh - good - that has caused some to abandon the mutate puzzles atm 11:17
alwen I have seen that complaint in global and in group 11:17
TimovdL Last time I had to restart all scores were lost on a track except the numbered ones so I could get my score back 11:17
@tamirh yeah, hopefully in the long run as seth said we won't need the 'disable filters' button but we'll get it working as soon as possible 11:18
Susume2 is brow correct that wiggle ignores the filters? and will that continue to be true when they are re-optimized? 11:18
auntdeen2 but - tamirh - it isn't when you have the disable checked 11:18
auntdeen2 the difference is without the redline 11:18
auntdeen2 your credit best is better than your very best - if you do not save the credit best, you lose it 11:18
@tamirh Susume2, yes and no. the values wiggle is trying to minimize is pre-bonus/penalty. the global minimize will not allow scores that are lower than previous scores though, and that does take into account the modified score by the filters 11:20
brow42 ah, ty. 11:20
@tamirh auntdeen2, ahh ok. gotcha. juggling a couple similar feedbacks in the tracker (= 11:21
auntdeen2 :-) 11:21
auntdeen2 it may be related to toggling the disable on & off 11:22
@tamirh i probably missed a question while typing some answers, was there anyone who asked something that hasn't been addressed? 11:22
@tamirh auntdeen2, i would assume so since that's what has changed recently, but it could be something else 11:23
auntdeen2 … really hope not :-( 11:23
TimovdL I have some questions not related to the filters 11:23
@tamirh TimovdL, go ahead 11:23
auntdeen2 I mean, hope it isn't something else 11:23
TimovdL Do the handffolding puzzles have any scientific purpose? 11:23
vakobo tamirh: ther is AreConditionsMet function butnothing about penalties 11:23
vakobo *there 11:23
ComputerMage Timo: Good catch! I like mix with rough hand folding and then script it to the maximum 11:23
@tamirh One of the major issues we find when a lot of people are all using similar scripts is that many solutions converge on very similar results. What we get when we add in handfolding rounds is a higher likelyhood that there are many divergent solutions before they are refined by scripts 11:25
@tamirh (let me know if that didn't make sense) 11:25
ComputerMage Tamir: Will we have API for ability to use Move and Cut functions? 11:25
TimovdL But does it really give that? 11:26
auntdeen2 Timo - don't know if any pure science function - but I have relearned - and learned so much from the hand folding 11:26
@tamirh vakobo, I believe there's a feedback for that, we'll be adding some more LUA API for checking values of filters 11:26
@tamirh auntdeen2, that's also a good point. a lot of people that are pros here now started out with just handfolding (since scripts hadn't been introduced yet). 11:26
auntdeen2 yes - much of that has been lost 11:26
brow42 I like hand folding, but I don't know how to use the rebuild tool by hand except in the most tedious way, so I stop handfolding pretty early on. 11:26
vakobo tnx 11:26
alwen The electron density puzzles really show how a script can run up a high score but not match the density (and the native) 11:26
auntdeen2 and doing the handfolding has given me some new ideas for scripts :-) 11:26
ComputerMage Tamir: Move and Cut functions in API. Will it be introduced? 11:26
@tamirh TimovdL, we havn't hand many hand-folding puzzles yet so we're still analyzing how much they affect it 11:26
@tamirh ComputerMage... maybe? 11:26
TimovdL In handfolding puzzles I do less changes because I cant cleanup easy 11:26
@tamirh TimovdL, what do you mean by cleanup? 11:26
ComputerMage I think that it's good idea to match tools that hand folding has and API functions that covers it 11:29
alwen Like doing a blue fuse after a rebuild 11:29
auntdeen2 some very good folders make radical changes by hand - then use drw scripts to refine the changes 11:29
marie_s fuse by hand is very easy 11:29
@tamirh ComputerMage, yes in general. i haven't looked into issues that may arise from them being added. perhaps jeff or seth could elaborate more 11:29
eromana in 690 dimer the Layer Design Approved Residues should be highlited in the design mode left click amino weel, maybe a white border arround the amino acid icon 11:30
ComputerMage Ok. Thanks! 11:30
markm457 tamirh - do you look at all tracks for diversity or only credit best solution? 11:30
alwen Yes, I can do it by hand, but running the script is very esy 11:30
TimovdL Remove clashing and bad backbones (normaly by running a banding script and fuze or DRW) 11:30
@tamirh eromana, I think that's another feedback (that has been buried since it has been posted) :p 11:30
auntdeen2 if you are continuing the hand building - is it possible to add mutate to the gui script functions? that is boring, boring, boring to do all by hand 11:31
auntdeen2 … and I'm the queen of loving the boring stuff :-P 11:31
marie_s mutate all work well 11:31
TimovdL Just let V2 scripts with disabled wiggle and shake to work on handfolding ones would help a lot 11:31
@tamirh auntdeen2, yes could you create (or reply to so it pushes it to the top) a feedback about that. we need to add some functionality to the GUI scripts so that hand-folding is easier to do. 11:31
@SethCooper we should add mutate all to the gui scripts 11:31
auntdeen2 works okay, marie - but would be sooo nice to do a shake wiggle in between 11:31
ComputerMage marie: Not really. I had many situations when mutate all created disallowed mutations 11:31
ComputerMage disalowed by the filters 11:31
auntdeen2 thanks tamirh & Seth 11:31
marie_s the filters dont work well, script will not solve that 11:31
ComputerMage that's why timo added 1-by-1 mutate in his DRW script 11:31
ComputerMage I mutated manually ;) 11:31
eromana does an unmutate button make sense 11:31
ComputerMage via mutate all button, so that wasn't issue of scripting ;) 11:31
TimovdL Yes but that was before the layer filter changed, not needed anymore 11:31
Susume2 can mutate function be changed to respect the filters? it still chooses invalid sidechains 11:31
TimovdL If so, it should be choosable susume 11:35
ComputerMage Ok. Good to know! 11:35
Susume2 good idea timo 11:36
@tamirh Susume2, we had the filters do that before but many people really hated that. The issue is that those residues are not invalid, it's just if you have too many of them it's probably not good... 11:36
TimovdL Yet another question. Why are already solved proteins used as puzzles? It is so easy to find the natives for those. 11:36
ComputerMage tamir: But MutateAll should account that preluminary info when it mutating 11:37
@tamirh with some added LUA interface and a better GUI for mutating residues, it may make the issue go away. If you can request which residues are expected 11:37
Susume2 ty tamir 11:37
brow42 tamirh, it is very different situation between the random mutator picking invalid residues and a human/script saying "Make This THIS" 11:37
brow42 although the code may be the same :( 11:37
alwen Like everything ending up arginine 11:37
@jflat06 @TimovdL, we use solved puzzles because it is actually quite hard to come up with blind cases on a consistent basis 11:40
TimovdL But invalid now might become valid after a wiggle 11:40
ComputerMage I have a question about weird behaviour of freezed segments during the wigglng 11:40
auntdeen2 "GUI script functions" feedback filed :-) 11:40
@tamirh another point that jeff brought up was that if you have bands, the "only allow score increases" is disabled to allow the bands to modify the protein 11:40
ComputerMage I suspect that they shouldn't move but they just fly away dueing the wiggling ;E~~~~ ;) 11:40
brow42 I've notice the pull tool behaves incredibly differently if a band is attached. 11:41
markm457 any-dev - do you look at all tracks for diversity or only credit best solution? 11:41
brow42 which is good, since the pull tool rarely goes where I'm pulling it. 11:41
@tamirh markm457, it depends on the puzzle 11:41
TimovdL The pull tool should really do what you want not jumping in a differnt direction 11:42
Susume2 freeze keeps things from bending but not from moving, CM (despite what tutorial implies) 11:42
ComputerMage Sus: That is bad :( In that case i should band them in space with 0 length and maximum strength then 11:43
brow42 this is a FAQ, I think a clear answer about how to be sure you see the solutions we want you to see (because they are distinct, separate attempts) should be put in the FAQ. Not everybody uses tracks, for example. 11:43
alwen You can put a band on a frozen segment, but if you try to pull it it says it is locked 11:43
Madde does the Core Existence filter take the whole polymer into account (i.e. the interactions between them) or just the workable monomer? 11:43
@tamirh Madde, it doesn't just take the monomer into account (it did at first but that bug was fixed) 11:46
Susume2 now that it takes the polymer into account, are you still getting monomers with viable cores? 11:46
Madde on 690 my monomer doesn't have a core but the dimer should have one, I'm not sure the filter noticed that 11:48
@tamirh Susume2, yes. we're seeing much better monomers in the proteins you all are designing. We also have a score modifier (not a filter) that will upweight the monomer itself in the score, which will ensure the monomer folds well as well. We haven't seen the need to turn that on yet. 11:49
brow42 it is difficult to get a core with 2 distinct layers of construction (2 sheets, or helix sheet). It seems I need a completely buried structure and 3 layers. This uses up a lot of residues. 11:49
alwen And it's so slow even with a low number of residues 11:49
Susume2 i've noticed that too brow 11:49
alwen I am still unable to send a screenshot to chat (this feedback:) 11:49
@tamirh Madde, the 'core' it cares about is for the entire protein. With the filters we've seen that monomers have tended to have a core even though we haven't added a direct check against it. If a lot of solutions end up with very poor monomers again we will can turn on the score modifier that upweights the monomer 11:49
eromana in 690 Approved Residues why is HIS repeated 11:49
alwen https://fold.it/portal/node/994029 11:49
alwen HIS has two forms, you can see them in Stick +PolarH view 11:53
@tamirh eromana, there are two kinds of HIS and when they're listed it just uses the 3-letter designation which is why they're there twice 11:53
TimovdL And sometimes changing to an approved one does not satisfy the filter 11:53
brow42 I wasn't going to ask since this is more a bio question, but why is HIS penalized but all the other hydrophillics allowed? 11:54
@tamirh TimovdL, interesting. if you have a save file where this happens could you post a feedback and share that solution with yourself then tell us the name of the share? 11:54
@tamirh brow42, that would be a question better answered by beta_helix or on of the other bakerlab scientists 11:56
TimovdL What I saw was that then another segment became unsatisfied 11:56
brow42 Timo, that is so aggravating! 11:56
@tamirh TimovdL, ahh. yes mutating can affect the surface area of the protein and change what is considered the core/surface 11:56
TimovdL But when I see one and the accepted ones change (that happens also) I will post it 11:57
@tamirh TimovdL, unfortunately that behavior is to be expected some times. The list of "accepted" residues isn't a hard list of "know" good residues. It's just an estimate and when you actually do the mutate it recomputes what's "accepted" based on the changed surface area. 12:00
TimovdL Then that would be the cause, it is the layer design filter that is hardest to satisfy 12:01
eromana in 690, in the amino acid wheel the white high lited amino acid should be the present amino rather than what it was 12:01
Susume2 maybe instead of "invalid", which sounds like a static category, the filter could say something like "unlikely residues" 12:01
@tamirh eromana, yes that's a very old bug that never got fixed (= 12:02
@tamirh Susume2, that's a good idea. it would make it more clear what's actually happening 12:02
@katfish Ok, it's time to wrap things up here in the next few minutes 12:02
@katfish Any last burning questions? 12:02
brow42 next scientist chat is when? :) 12:02
ComputerMage I have one 12:03
marie_s scientist chat is when? 12:03
TimovdL The valid list should be called maybe valid ones 12:03
ComputerMage if I have polymer that mimicking a ion chanel 12:03
@katfish Very soon, brow42. I'm talking to the scientists about organizing one. In the meantime, here's a blog entry from David Baker that might interest you: https://fold.it/portal/node/994822 12:03
brow42 ty kat 12:04
@katfish np 12:04
ComputerMage does internal part of chanel calculated properly? 12:04
@tamirh ComputerMage, that's a question probably best left for the scientist chat 12:04
brow42 oh I have a related question! 12:05
ComputerMage ok 12:05
@katfish Ok guys, if you have science q's hold them off for the scientist chat :) 12:05
alwen "Where are the cheat codes?" 12:05
brow42 tamirh, I thought hiding/packing had to compute the access to water already. Is that not true? Why is the filter so much slower then? 12:05
@tamirh alwen, up up down down left right left right b a 12:06
alwen lol 12:06
@katfish Ok everybody, it's time for us to sign off 12:06
marie_s bye and thanks 12:06
@tamirh brow42, it uses nearest neighbors and not the full surface area 12:07
TimovdL Bye and tx also 12:07
brow42 I see, ty. 12:07
@katfish Thanks for coming out! We'll have a transcript posted up on the site shortly. I'll have a link in the comments of the devchat announcement. 12:07
auntdeen2 bye and thanks 12:07
eromana interesting thanks 12:07
@SethCooper bye, thanks everyone. 12:07
@tamirh thanks for all the great questions everyone. back to work now! 12:08
auntdeen2 er… we are supposedly back to fun :-P 12:08
@tamirh back to fun for you guys. back to work for us (= 12:09
auntdeen2 lol 12:09
auntdeen2 thanks, tam 12:09
brow42 auntdeen, the comment on my latest recipe might clarify the wiggle issue for you 12:09
ComputerMage thanks guys and gal ! ;) 12:09
auntdeen2 thanks brow - I will look 12:09

Generated by irclog2html.py 2.12.1 by Marius Gedminas
- find it at mg.pov.lt!

Get Started: Download
  Windows    OSX    Linux  
Windows
(7/8/10)
OSX
(10.7 or later)
Linux
(64-bit)

Are you new to Foldit? Click here.

Are you a student? Click here.

Are you an educator? Click here.
Search
Only search fold.it
Recommend Foldit
User login
Soloists
Evolvers
Groups
Topics
Top New Users
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