Developer Chat

@inkycatz All right, everyone hop on board the chat train. Let's keep our hands and feet inside the vehicle at all times, or if we can't do that, at least make sure we have some safety helmets. :D Let's DO this. 13:59
@inkycatz Welcome to our last chat for 2014! Thank you all for coming :) We don't have our schedule set for 2015 yet but when we do, watch the news! 13:59
@inkycatz Some cutting and pastig of generic housekeeping follows, please don't fall asleep on me. 13:59
@inkycatz As we know, developer chats can get pretty busy and popular, and we want to ensure that as many questions get through as possible, especially since it's our last chat of the year. If things get too brisk, we have a moderated option available to us if needed. You may have seen us brilliantly testing that earlier. Twice. 13:59
@inkycatz For what it's worth, I don't see this as likely today but wanted to make sure you all weren't caught off guard in case that happens. :) Special shoutouts to Mike Cassidy, Susume, mimi and silverberg for helping me out with those testing bits earlier. 14:00
@inkycatz All chats are logged! If you miss something, or have to leave, check http://fold.it/portal/chats afterwards. Thanks to jflat's magic logging systems, those logs get posted pretty quickly. 14:00
@inkycatz Before we dive in, I just wanted to mention we will be adding some new moderators in the near future to help our current group. Please contact me privately aka not now if you have suggestions of outstanding community members we should consider in this process, or find this of something that would be of interest to you. 14:00
@inkycatz And now.... 14:00
@inkycatz I'd also like to introduce curium and kothiwsk who wanted to take a moment to mention some exciting upcoming stuff. So yeah, I'm mixing a little science in today, but I think you all don't mind. 14:00
@inkycatz You guys there? You're up! 14:00
curium Hi guys 14:00
kothiwsk Hey everyone 14:01
curium Kothiwsk and I started developing Foldit last year 14:01
curium we are working on creating a mode for drug design, with the idea that players can design drugs using foldit 14:01
curium We are going to set up puzzles that target third world diseases with the hope that you guys can design drugs for the diseases 14:02
curium for each puzzle, we are going to either make the drug you design or order it 14:02
curium and then test them biochemically and get back the results to you 14:02
curium we have a couple of videos if you want to see the progress 14:03
spvincent Peptide-like drugs or any chemical entity you like? 14:03
BletchleyParkirc transmembrane functionality ? 14:03
curium small molecules 14:03
spmm2 sounds great  yes please to videos 14:03
curium like saquinivar for hiv 14:03
curium https://en.wikipedia.org/wiki/Saquinavir 14:03
curium video one: http://youtu.be/CEaGj1knm9o 14:04
curium video two: http://youtu.be/pDm27XiWHT8 14:04
curium we are hoping to give you all the tools you would need to design the small molecules 14:04
curium like a scoring function that will tell you things about the small molecule 14:05
curium and then provide tutorials on how to do all of it 14:05
BletchleyParkirc Faster code ? 14:06
spvincent sounds like an interesting new direction for design puzzles 14:06
spmm2 can't see any filter conditions :D looks great 14:07
kothiwsk Its work in progress 14:07
curium the code should be fast 14:07
curium the algorithms that we are using have been optimized for drug design 14:08
curium and are the underlying basis for players manipulation 14:08
jeff101 will there be new LUA code we can use to make Recipes for such puzzles? 14:08
curium yes, we will provide that interface 14:08
spmm2 when can we start testing? 14:08
@mimi2 will this run in a seperate stream or be interspersed in the usual puzzzles 14:09
curium we are  hoping for a late spring date, but we are optimistic that it will be sooner 14:09
BletchleyParkirc are you implementing openCL ? I keep hammering on that topic. 14:09
curium they are going to be intersperesed 14:09
curium the problem with openCL is that Rosetta itself does not take advantage of openCL 14:10
BletchleyParkirc A good opportunity to optimize rosetta 14:10
curium however, some of the drug design code can use gpus 14:11
BletchleyParkirc that is a good start 14:11
curium the underlying software being used for drug design is the BCL 14:11
curium you can find more information about it here: http://meilerlab.org/ 14:12
@jflat06 there's also a sociological problem in that if it does provide a significant advantage in score, people with high end dedicated graphics cards will have a strong advantage. unless you can show that the advantage in score produces a corresponding advantage in scientific accuracy, this is actually a bad thing for the project as a whole. 14:12
BletchleyParkirc please use openCL, not CUDA or multithreading to keep it CPU/GPU independent 14:12
kothiwsk No we use OpenCL 14:12
BletchleyParkirc I can see that in sidechain selection a GPU accellerated version benefits science 14:12
BletchleyParkirc very good ! 14:12
BletchleyParkirc and in fact in general because you can calculate better results faster 14:13
curium we will strive to make the algorithms work on a variety of computers 14:13
curium this is a topic that kothwisk and I talk about at length 14:13
BletchleyParkirc Pretty much every GPU can be used for openCL these days, so not just expensive cards 14:13
kothiwsk I guess this is a question for the whole developer community to answer 14:14
BletchleyParkirc The cost of electricity is a multiple of the cost of a GPU on an annual basis 14:14
@jflat06 assuming people have a dedicated GPU, which many do not. And even in the case they do, there are very large differences in the range of graphics cards 14:14
kothiwsk as Jeff pointed out may give some people advantage 14:14
BletchleyParkirc most dohave a dedicated GPU 14:14
curium do you have more questions about the drug design project or the science of it? 14:15
jeff101 is there something like CASP for drug design? 14:15
BletchleyParkirc Does it also improve the size of the models we can work on ? 14:15
kothiwsk there is none as much as I know 14:15
curium there is CSAR, which is used for docking small molecules, but it is not drug design 14:15
curium you can find information about it here: http://www.csardock.org/ 14:16
curium Rosetta performs well at it 14:16
@Susume2 will the drug design puzzles be primarily docking design? 14:16
kothiwsk No they will be design puzzles 14:16
kothiwsk starting from a given small fragment/drug part 14:16
@jflat06 right, i think Susume means that you will interface your design with another protein. 14:17
curium You can perform docking with it, using wiggle funcitonality 14:17
curium sorry 14:17
curium wiggle/shake 14:17
curium of the small molecule 14:17
curium I believe in the first video I show some of that 14:17
@mimi2 I'm sure there will be more questions when we've had a chance to look at them 14:18
spmm2 yes a whole swag of new tools and you appear to be working on a ligand which I assume is the drug part 14:18
@inkycatz Can we count on you guys to come on back and have a full fledged science chat with us when this is near ready? :) 14:19
curium the idea is that you, the players, can design drugs for a given protein target 14:19
@mimi2 will you be doing any background advice on design 14:19
curium There will be background advice 14:20
spmm2 put me down for early testing and wiki help 14:20
curium we are hoping to get medicinal chemists to help you guys 14:20
curium in fact, we have some volunteers here that are more than happy to work with you guys 14:20
@inkycatz That's really awesome to hear. 14:20
curium for each puzzle that we release, there will be an infographic on the protein 14:20
curium it will tell you who is effected by the disease 14:20
curium the small molecules we know that help combat it 14:21
curium and any other information that will get you started 14:21
Aotearoa i like it 14:21
Aotearoa Ki Ora Curium 14:21
Brow42IRC The design filters encourage us to make very hydrophobic cores in multimer design.  Won't these interfaces lack specificity if we don't also include some hydrophillic residues? 14:21
curium for Brow42IRC question, this is the hard thing about drug design. the small molecule might be in a binding pocket that is hydrophobic 14:22
curium but in order to make a small molecule, you will need to make that small molecule hydrogen bond with the backbone 14:22
curium so, hydrophic/hydrophilic residues will play a huge role in your designs 14:22
curium and what you design around them 14:23
jeff101 must all of the small molecules be attached? like some proteins have internal waters 14:23
@inkycatz Sounds like a topic we could chat about all day, really. Let's take jeff101's question and get you guys back for a chat again soon. :) 14:23
curium we will provide tools to highlight these things. For example, in the video, you can display hydrogen bond acceptors and donors. The goal would be to match an element that complements those grid points. This is called QSAR design 14:23
curium for now, we will only have 1 small molecule per protein, but you can break that small molecule up, recombine it, and apply a whole set of things to the fragment 14:24
vmulligan @jeff101: Some proteins have internal waters, but internal waters are incredibly hard to design and score.  There is a huge entropic penalty to burying and ordering a water that would otherwise be disordered in solution.  Someday, maybe we'll be able to handle that well, but currently, our best bet is to avoid buried waters. 14:24
curium just like you would do to a protein during folding/design 14:24
curium and what @vmulligan said :D 14:25
vmulligan (Hi!  Sorry I'm late.) 14:25
@inkycatz (Always glad you can make it!) 14:25
@inkycatz curium and kothiwsk, thanks for dropping in. :) Can we get those links again to the videos? 14:26
curium sure, here they are 14:26
curium video 1: http://youtu.be/CEaGj1knm9o 14:26
curium video 2: http://youtu.be/pDm27XiWHT8 14:26
curium thanks for inviting us! 14:26
@inkycatz No problem, you'll be hearing more from me soon as I schedule a future chat on this. :) We probably could talk a LOT more, and should! 14:27
@inkycatz 2015 everyone, gonna be good times here. 14:27
kothiwsk Sure! sounds good 14:27
@inkycatz So let's wake jflat up and see if we can tackle some of those thread questions we got. :) 14:28
@inkycatz First question is about 'rotating segments' and a suggestion from spmm (http://fold.it/portal/node/1998734#comment-29190). Thoughts? 14:28
Brow42IRC A note for developers: I think many LUA scripts will break when run on ligands that aren't peptides. (just looking at the video now) 14:28
@inkycatz (good heads up, brow!) 14:28
@jflat06 While I agree that could be nice, sometimes we have to limit the functionality available to players in order to preserve the competitive integrity of the game. 14:29
@inkycatz Next, do we have current plans to put Foldit in the cloud at some point in the future? 14:29
@jflat06 This is one situation where I think providing hooks into directly setting torsion angles would make it too easy to copy from a known set of torsion angles. 14:29
Alreadytaken can you say that in simple words jflat? 14:29
@inkycatz (just priming the question list, take your time on that first one ;) 14:29
@jflat06 Sure - it makes it too easy to cheat by copying :) 14:29
spmm2 ah ok fair enough thanks jflat 14:30
@jflat06 we have already had issues with that in the past, where we gave too much power in the tools 14:30
@jflat06 so we're extra careful of situations like that now 14:30
@jflat06 as to Foldit on the cloud: this is unlikely 14:30
Alreadytaken could it be possible to get the tools and get excluded from the scoring system? 14:31
@jflat06 there are two major things preventing it. one is that foldit is computationally expensive, and we cannot afford to host all of that computing power. 14:31
@jflat06 the second is that a significant amount of work would need to be done to develop a system for offloading work to remote machines. 14:31
vmulligan And much of that work would have to be on the Rosetta side (the protein folding software that underlies the Foldit interface). 14:32
@jflat06 @Already - that's a good question. We have actually considered running "open" puzzles, where players would be able to save out their solutions and use whatever tools they want on them. 14:32
@jflat06 and potentially have access to more powerful tools in-client. 14:32
vmulligan The third thing, too, is that quite a lot of data would have to go back and forth between a cloud server and a player's client in real time.  That might not be feasible -- the game would likely get too slow, and would require too much bandwidth. 14:33
@jflat06 But these puzzles and features are entirely hypothetical at this point. 14:33
Alreadytaken Could automatic 'cloud" storage be enabled? 14:33
@jflat06 cloud storage is probably an easier problem. 14:33
@jflat06 you may actually be able to configure that already through 3rd party tools 14:33
@mimi2 I think that would be a good idea jf when the result is more important than the puzzle 14:34
BletchleyParkirc There already is such a thing , it is called share with 'myself' and share with 'group' 14:34
Alreadytaken yes but 3d party tools are a p in the a 14:34
Alreadytaken and I keep forgetting to do that... 14:34
@jflat06 @BP, I think he may want things like preserving options.txt and macro files across machines. 14:34
Alreadytaken yes everything 14:35
@jflat06 or just saves (I don't know) 14:35
BletchleyParkirc e-mailing it ? 14:35
Alreadytaken so that the pc doesn't matter 14:35
BletchleyParkirc memory stick 14:35
Alreadytaken floppy disks 14:35
@jflat06 alright, shall we move on? 14:35
@inkycatz carrier pigeon 14:35
BletchleyParkirc if you choose so 14:35
Alreadytaken =] 14:36
@inkycatz Any comments on the grand wishlist thread sitting here at http://fold.it/portal/node/998395 ? 14:36
@inkycatz I know you've read it. :) 14:36
@jflat06 there are a lot of really good suggestions in there, which I'd be a fan of implementing. 14:36
@inkycatz :) 14:36
@inkycatz I won't ask you to play favorites! 14:37
@inkycatz Just, keep 'em coming folks, we do read them. 14:37
@inkycatz Kea has brought up a request for some translation support efforts. Where are we at on that? 14:37
spvincent Where do performance improvements stand in the "to do" list? The current trimer puzzle is ridiculously slow. 14:38
@jflat06 unfortunately, it is difficult to find time for a lot of these quality of life changes when we're trying to keep the software/server running, while also developing the tools to continue advancing the science side of things. 14:38
@jflat06 @spvincent - i think a lot of the performance problems stem from some of the filters being slow 14:39
vmulligan The "mix-and-match" suggestion (being able to blend bits of two different solutions to a puzzle) is similar to something that we've thought about implementing in Rosetta, but haven't, yet.  Some of the others might also be Rosetta changes that could carry through to Foldit eventually. 14:39
@jflat06 i can tell you we are definitely working on improving the performance of specific filters 14:39
@jflat06 namely the fragment filter 14:40
@mimi2 HURRAH 14:40
spvincent seconded 14:40
Alreadytaken Could it be possible to have an overview of the functions that are being worked on, on the website? 14:40
retiredmichael thirded 14:40
frood2IRC :) 14:40
@inkycatz Ah you brought me to the next thing Already - letting jflat talk about the stuff he's working on. 14:40
@inkycatz :) 14:40
@jflat06 and yes, what vmulligan said - we actually have a deprecated function in code that allows you to pull bits from different solutions into your current solution 14:41
Brow42IRC Disabling slow filters has helped a lot, but when you want a recipe to use rebuild to optimize a particular filter, then you have to turn them (all) on, and it's not practical. 14:41
@jflat06 and we plan to include that one at some point. 14:41
spmm2 fourthed and -99999, cutpoint negative score? 14:41
@jflat06 @Brow one of my intended changes is that you should eventually be able to just click a 'disable filter' checkbox for each filter 14:42
@jflat06 in the filter menu 14:42
spmm2 that would be a huge help jflat 14:42
@jflat06 @spmm, while probably pretty far off, I would like to design an interface for showing you your 'modified' score function 14:43
Brow42IRC There's currently a discrepancy between the energy subscores and the displayed filtered score that we haven't been able to figure out. I think spvincent has a feedback on it. 14:43
@jflat06 which would let you just see the score without considering things like cutpoints (and potentially letting you customize what sort of score it shows) 14:43
@jflat06 @Brow - I haven't seen that one 14:44
spmm2 thanks - folding in the dark with wide open cutpoints at the moment 14:44
@jflat06 as for the fragment filter: 14:45
@TimovdL Finally catched up with what is said before, room for another question/remark? 14:45
@jflat06 my current plan is to pull some rosetta-side performance improvements into the foldit codebase, and then add a feature that lets us tell the fragment filter to stop calculating after it sees 3-4 bad fragments. 14:45
@jflat06 so if you are in the early stages of folding, it will just sense that there are a lot of bad fragments, and stop early instead of calculating every single one. 14:46
@jflat06 this should make the early-game fragment filter performance MUCH better 14:46
@jflat06 @Timo - sure! 14:46
spvincent 14:47
jeff101 any news about CASP? 14:47
@TimovdL Puzzle 1018 is really to big to handle with all the filters, fragment improvement should help but this one is even on my fast computer doing very slow wiggles 14:48
@inkycatz I believe the automated assessments are up and folks are still sorting through that. 14:48
@Susume2 http://predictioncenter.org/casp11/results.cgi 14:49
@inkycatz Thanks Susume! :D 14:49
@inkycatz I know you've been looking. 14:49
@TimovdL A wiggle on 2 segments takes about 1 minute for 1 iteration 14:49
@jflat06 @Timo - if the performance problem lies in rosetta, it might be difficult to address 14:51
@jflat06 but I'm fairly sure a lot of it comes down to our filters 14:51
@jflat06 which we are working to improve 14:51
@jflat06 how does it run with filters disabled? 14:51
@TimovdL I think it is the combi of filters and trimer and the number of residues 14:51
BletchleyParkirc Why is it difficult to adress performance issues n Rosetta ? Isn't the source available ? 14:52
@jflat06 because there is a limit to how fast you can make the required computations run 14:52
BletchleyParkirc openCL 14:52
@TimovdL (Wait 2 years till I retire, then I will volontaire to help optimize things) 14:53
@jflat06 again, i don't view that is a solution in foldit's case - nor is it something implemented in Rosetta 14:53
BletchleyParkirc I volunteered years ago 14:53
vmulligan Most supercomputers don't have GPUs.  OpenCL development for Rosetta has not been a priority due to the clusters that it typically runs on. 14:53
BletchleyParkirc then it should be implemented in Rosetta 14:53
@jflat06 you can try talking directly to rosetta folks about OpenCL development in Rosetta 14:53
BletchleyParkirc I strongly disagree with the supercomputer statement 14:53
BletchleyParkirc I will Jeff, thanks 14:54
BletchleyParkirc Tihange2 amd Titan are prime examples 14:54
vmulligan The higher priority has been better CPU-based parallelism.  There have been some recent changes there that may carry through to Foldit in the next year or so. 14:54
@TimovdL There is a huge wishlist for more Lua functions any progress on that? 14:54
@jflat06 @Timo - not currently. Our current focus is on improving performance and tools in design, as we feel this area is very important. 14:55
Brow42IRC functions that tell us if an atom is a donor/acceptor/both, and which pairs are bonded, would help a lot on ligand puzzles. 14:56
Brow42IRC (without looping over all pairs) 14:56
@TimovdL Also a function that tells us which are donor/acceptors 14:56
vmulligan That shouldn't be that hard.  What do you think, jflat06? 14:57
jeff101 somehow Foldit knows these things to make pictures, why not give us access to them too? 14:57
Brow42IRC again without looping please :D 14:57
spvincent In the early part of the game, when the proposed change to the way the Fragment filter works would be most useful, it's easy enogh to Disable Slow filters. 14:57
@jflat06 If the correct rosetta hooks and calls are available, designing a Lua interface probably isn't too hard. 14:58
jeff101 Foldit knows which things are h-bonded, why not give us a list of them? 14:58
@TimovdL I agree with vincent, later it destroys good scoring solutions 14:58
@jflat06 @Timo, are we talking performance, or giving it a poor score? 14:59
Brow42IRC remix never returns anything remotely like my designed shape so I do not use it. 14:59
@jflat06 @Brow - I agree that the situations it can be used effectively in are fairly limited 14:59
spmm2 did the standalone foldit client get updated for Yosemite? 15:00
@jflat06 we will be introducing improvements to the fragment filter that may make the results from Remix more useful 15:00
@jflat06 @spmm2 - if you're talking about the old standalone, no. 15:01
@TimovdL Yet another topic, idealize was said to improve the corresponding score, my feeling is it doesnt 15:01
@jflat06 that code is several years old and hasn't been maintained. I've made some efforts to completely rewrite the application hierarchy so that it would always be up to date, but there's a ton of work left to be done on that front. 15:02
@jflat06 @Timo - do you mean that you run idealize and your ideality score goes down? 15:02
@TimovdL It does not fix bad scores 15:02
Brow42IRC I have also used idealize and then seen penalties in the Tab score display right after. 15:03
@jflat06 The main issue you will probably see is that prolines still act strangely with ideality 15:03
@jflat06 prolines are a special case in a lot of the code, and this is one that we haven't gotten entirely consistent. 15:04
@jflat06 for the rest of residues, under normal circumstances, idealize should get the score penalty to be minimal 15:04
@jflat06 are you seeing situations with non-prolines where idealize doesn't fix a large ideality penalty? 15:05
Aotearoa i noted prolines once as being the most interesting,signalling kind of amino, at times, by just changing them all around a protein, i could rank up some good points and watch the protein change shape 15:05
vmulligan @Aotearoa: Prolines are conformationally very constrained.  They do a lot to help to determine the shape of loops of natural proteins, and influence overall fold quite a bit. 15:06
@TimovdL I noticed something else strange. Running acid tweeker normally does not find as much points as running it with some segments froozen, looks like the wiggle tool distributes too much in the non froozen situation 15:07
@jflat06 as far as I'm aware, there aren't really any guarantees about whether a set of DOFS should always minimize further than a subset of them. 15:10
@jflat06 meaning that what you're seeing isn't entirely unexpected 15:10
vmulligan On the contrary, minimizing a small set of DOFs can result in finding a deeper energy well than minimizing all DOFs. 15:11
@TimovdL DOF is? 15:12
vmulligan (That's why it's better to start with low-powered Wiggle rather than high-powered Wiggle). 15:12
@jflat06 Degree-of-freedom 15:12
vmulligan Each bond that you can rotate about is a degree of freedom of the system -- a DOF for short. 15:12
@jflat06 each little thing that can rotate or stretch on the protein is a degree of freedom 15:12
@jflat06 haha 15:12
vmulligan Jinx.  :P 15:12
@TimovdL Now I can understand why it works 15:13
@jflat06 Are there any other questions/topics people want to go over? 15:13
vmulligan I need to run.  All the best, folks -- happy folding! 15:14
spmm2 I just wanted to add: Thanks to all the devs and testers (and scientists) for the improvements you have made to the game this year and I hope you all have a great break, and thanks for answering our questions :) 15:14
@inkycatz Thanks for coming by vmulligan! 15:14
@inkycatz Thanks jflat :) 15:14
jeff101 thanks for having chats like this 15:14
@inkycatz And thanks to all of you! 15:14
BletchleyParkirc thanks 15:14
@inkycatz We'll be planning 2015's chats soon :) 15:14
@TimovdL Sorry for my late arrival but tx for the answers, I learned some more. 15:15
@jflat06 thanks everyone 15:15
spmm2 seasonal puzzle schedules for when we have some time off 15:15
@jflat06 @spmm - good reminder! 15:16
@inkycatz Yes! :) 15:16
@inkycatz Look for the chatlog soon for all those who couldn't make it, or if you want to read through all the tasty info again. :) 15:17
@jflat06 Alright - thanks for joining us for the chat everyone. Happy Folding :) 15:17

Generated by irclog2html.py 2.13.1 by Marius Gedminas
- find it at mg.pov.lt!

Get Started: Download
  Windows    OSX    Linux  
Windows
(Vista/7/8)
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, Microsoft, Adobe, RosettaCommons