Developer Chat

@inkycatz OK! 10:58
@inkycatz Everyone, it’s time for SCIENCE. 10:58
@inkycatz Welcome everyone to our science chat today. :) A few quick housekeeping items. 10:58
@inkycatz First of all, chats are logged, so if you have to go don’t worry we post the log afterwards 10:59
@inkycatz - for all your previous chat needs 10:59
@inkycatz and all your future chat needs 10:59
@inkycatz props to jflat for making those happen super quick post-chat 10:59
@inkycatz Today’s topic and question list is pretty full, which makes sense because we haven’t had a chat in a while. 10:59
@inkycatz That being said, let’s try not to get too overloaded on one topic with side questions. 10:59
bkoep Hi, everyone! 10:59
@inkycatz We always have the forums for follow up threads on things. :) 10:59
@inkycatz Let’s start with any updates and whatnot from the team. 11:00
@inkycatz Looks like we have several of them here today :D 11:00
betahelix Hi everybody! 11:00
bkoep Hey everyone, we're looking forward to taking your questions today 11:01
bkoep I want to mention that we are joined today by another scientist from the Baker Lab, sboyken 11:01
bkoep sboyken is the resident expert on hydrogen bond networks 11:02
@inkycatz Did you guys have anything in particular to share in terms of news, or should I get to the question list? 11:02
vmulligan Hello! 11:02
sboyken Hi Everyone! 11:02
spvincent hello 11:02
betahelix I also wanted to introduce blivens, who is our collaborator on the VEGF puzzles: 11:03
betahelix 11:03
blivens Hi everyone! 11:03
betahelix Perhaps we can start with an update on those puzzles? 11:03
@inkycatz That works for me! 11:03
blivens Sure. 11:03
blivens So we've been working on the VEGF receptor 1, which is a big 7-domain protein attached to the outside of the membrane 11:04
blivens We have electron density data for the protein, but it's pretty low resolution so we couldn't make a protein model ourselves 11:05
blivens We had three foldit puzzles focusing on the hardest two domains 11:06
blivens We got a lot of great solutions back from those, one of which fit quite well 11:06
blivens So we were able to fit those into the electron density, clean up a few things, and now we're starting to write it up for publication 11:07
@inkycatz Exciting! 11:07
spvincent neat! 11:07
blivens It was too big a protein for a proper electron density puzzle, but I was happy that we were able to get good results from players even without that data 11:08
@Susume2 can you say whose solution fit well? 11:09
LociOilingIRC 2nd that 11:09
blivens I'm trying to remember exactly whose model was used, but maybe one of the other mods would know? 11:11
@inkycatz Or perhaps we can even get a followup blog post for the news and look into that. 11:11
blivens I do know that it wasn't the model with the best score 11:11
betahelix Yes, as we'll need to get the player(s) permission first! 11:11
free_radical hi 11:12
betahelix Thank you so much for stopping by with this exciting update, blivens. 11:12
blivens Sure! 11:12
@inkycatz Yes, thank you! I’m looking forward to learning more :) 11:12
@inkycatz As I’m sure everyone else is now too 11:13
@inkycatz Should I go ahead and tackle this list of questions, team? :) 11:13
vmulligan Sure/ 11:13
vmulligan *Sure! 11:13
@inkycatz I see a vmulligan so you get the first question, since you spoke up first: Are you running simulations on our Marburg binders? If so, what are the results? 11:14
vmulligan I'm going to try putting the results from the 25-residue design puzzle through our forward folding tests, but haven't had a chance to do so yet.  However, we're not as good at predicting small peptide structures as we are at predicting protein structures, so false negatives are more likely.  There are some other tests that I might try as well. 11:16
frood2IRC sorry to interject…but when the useful solution for VEGF is found…perhaps someone at foldit could do a quick vid explaining why it was useful? 11:16
vmulligan We'll also have a 17-residue inhibitor puzzle that will hopefully go live later today -- we just need to test-play it a bit more. 11:16
@Susume2 0.0  17! 11:17
LociOilingIRC it's like golf, lower is better 11:18
@inkycatz ha! 11:18
frood2IRC will this involve a large frozen section? 11:18
spvincent Why would small peptides structures be less easy easy to predict than protein strucxtures? I'd have thought the opposite would be true 11:18
@inkycatz that leads into the great peptide question I have on my list, spvincent! 11:19
@inkycatz “Why do small peptides like 2 or 3 bridges? Doesn't 1 bridge already reduce the available conformations (entropy) enormously? Don't 2 bridges have a 50% chance of mis-folding, and 3 bridges 66%-84%?” 11:19
blivens @frood2IRC that sounds like a good idea! In brief, it's essential for making new blood cells. That's a good thing in fetuses, and a bad thing in cancer cells 11:19
vmulligan Peptides are harder because they require greater precision.  Our scoring funciton is an approximation of reality.  In a large protein, our mistakes kind of average out.  In a small peptide, we need to calculate each interaction energy EXACTLY. 11:20
frood2IRC so will there be any point in lower power wiggle options on this one? I know that sounds trite. 11:20
vmulligan Two disulfide staples do result in four possible disulfide bonding patterns, and three result in fifteen, so yes, there are more ways of forming incorrect disulfides with more disulfides.  This is why more than four or five are rarely seen, except in very large proteins.  But without disulfides, there are many, MANY more ways that a peptide could misfold, so the benefit of having two disulfides instead of one usually outweighs the detriment.  (And in 11:22
LociOilingIRC is is icro one orkin? 11:26
blivens The fact that lots of natural proteins have multiple disulfide bonds means that the misfolding doesn't happen too often. Probably having a nice hydrogen bond network makes the protein zip up correctly during folding, so the right disulfide bond forms and then locks down the correct conformation. 11:26
jeff101 what's so great about skipping around w/s-s bonds like 1-4 2-5 3-6 seen in some toxins? seems that would be harder to fold than 1-6 2-5 3-4 for example 11:26
frood2IRC so the next step is to run HBnet with bridges? 11:26
@inkycatz vmulligan, did you finish your answer? I think it just got cut off :) 11:28
vmulligan We think that the "leapfrog" disulfide pattern maximizes the stablility benefit from having the disulfides.  Some patterns are obviously not very useful -- for example, 1-2,3-4,5-6 just gives you three rings on a string, which doesn't favour a compact structure so much. 11:28
blivens Also you have proteins called disulfide isomerases whose job is to break incorrect disulfide bonds and allow them to reform 11:28
vmulligan But 1-3,2-4,3-6 tends to really lock a structure into place.  Of course, it is necessary for other considerations to play a part -- good hydrogen bonding networks, good shape-complimentarity of hydrophobic interactions, etc.  This helps to destabilize the alternative disulfide bonding patterns relative to the correct one. 11:29
@inkycatz (should I go to the next question? :) 11:29
vmulligan And in the lab, there are tricks that we can play with synthetic peptides to force the disulfide bonding pattern that we want.  But yes, blivens is correct -- provided the desired pattern is the lowest-energy one, PDIs (protein disulfide isomerases) can give an incorrectly disulfide-bonded protein a chance to find its correct pattern. 11:30
vmulligan Sure, inkycatz! 11:30
@inkycatz This one’s a little more generic, but still interesting, and I’m sure something a lot of folks wonder: 11:30
@inkycatz We do a number of revisiting puzzles, how helpful are these to the process, and how are the "revisiting" results compare to the originals so far? (Do you have any examples of where they're radically different or the same?) 11:30
jeff101 1-3, 2-4, 3-6 seems like it won't work. did you mean a different combination? 11:30
bkoep We don't expect to see any "groundbreaking" results from the revisited puzzles 11:32
vmulligan @Jeff101:  Sorry, I meant 1-4,2-5,3-6.  Bad typing. 11:32
frood2IRC was there a reason for 1113 only giving bridge bonus for 3? 11:32
bkoep We mostly use the revisited puzzles to make sure recent changes haven't broken the basics of Foldit 11:32
@inkycatz frood, we’ve got time for questions, but I’d like to get through some of the ones people have posted, thanks 11:33
frood2IRC sure-sorry 11:33
vmulligan One notable exception recently has been the Marburg 37-residue puzzle.  I revisited that one because the energy function setup the first time was guiding players to not-so-great solutions.  Tweaking that led to better results from players.  It's a eal challenge for us, trying to set up the scoring so that designs with the qualitative characteristics that we're looking for are the ones that score the best. 11:34
bkoep I haven't looked at many of them closely, but in general Foldit players seem to perform pretty consistently on revisited puzzles. 11:34
@inkycatz This one’s more of a results in general question, but: How much do the top-scoring solutions differ from each other, and how far are they from the "native”? 11:35
vmulligan Sounds like it's a question that's more about the structure prediction puzzles.  Do you want to answer that, bkoep? 11:36
bkoep Sure 11:37
bkoep There's actually a wide range in results 11:37
bkoep Sometimes, all of the top-scoring solutions look very similar (especially if the starting structure suggests a particular fold) 11:38
bkoep Other times, we see several players with similarly high scores, but vastly different folds 11:38
jeff101 do you see the nice cones you like to see? 11:39
jeff101 the ones that go to smaller rmsd as the score improves? 11:39
jmbrownlee333 how bout comparison to known native folds? 11:39
bkoep The "nice folding funnels" that I've shown in blog posts are more relevant to protein designs 11:40
bkoep Natural proteins rarely have such picturesque funnels that favor the native fold 11:40
bkoep We say that the folding landscape for these proteins is "rugged" 11:40
LociOilingIRC so they're likely to get stuck during folding? 11:41
vmulligan Natural proteins, it's worth remembering, are marginally stable.  They have to be made, do their thing, and then be broken down.  With an artificial protein, we have the luxury of "over-designing" for incredible stability and a huge energy gap between the desired conformation and alternative conformations. 11:41
bkoep And that makes structure prediction for these proteins much more difficult than for our ideal, designed proteins with nice funnels 11:41
jeff101 is it easy to post such dot plots after each puzzle? 11:41
vmulligan @LociOilingIRC:  No, natural proteins are better optimzed than artificial ones for good routes from unfolded to folded states.  But it's an inherent property of all proteins to have "rugged" folding free energy surfaces.  It's just a result of being made of atoms -- a small rotation about any bond can result in large changes in the interaction energies of atoms in the chain. 11:42
@Madde Do you use our Denovo results as starting points for Rosetta@home? 11:43
vmulligan @jeff101:  The "forward folding" prediction is expensive, so we can't perform it for every design that playres produce.  (We use large-scale distributed computing to do those predictions.  Each point on the plot is the result of a separate trajectory run on a different computer.) 11:43
jeff101 but what about showing dot plots of just the Foldit results from all players and teams 11:44
vmulligan @Madde:  Rosetta@home is the computing infrastructure that we use for our forward folding predictions, to generate those plots -- so yes, we test Foldit designs on Rosetta@home. 11:44
@inkycatz It’s good feedback, jeff101 and one I’m sure we can look into 11:44
vmulligan But not each and every design, since we don't have the computing power for that. 11:44
@inkycatz I’m sure there’s a good compromise between more info and “all” info. :) 11:45
bkoep @Madde, no, Rosetta@home builds its structure predictions from scratch, using small fragments of known protein structures 11:45
jeff101 surely you make some graphs for each puzzle, you could post those graphs for us to see 11:45
bkoep If it's okay with everyone, I'd like to steer the conversation toward hydrogen bond networks 11:45
@inkycatz Yes, the clock tells me we don’t have a lot of official time left, and I know we want to talk about that. :) 11:46
jeff101 I would like to ask sboyken: 11:46
bkoep We're really excited about incorporating hydrogen bond networks into Foldit designs, but we want to know what questions or suggestions you have about the Hydrogen Bond Network Filter? 11:46
sboyken Definitely, I'd love to talk h-bond networks and answer any questions 11:46
jeff101 11:47
jeff101 11:47
@MikeCassidytoo Does H-Bonds occur in structures besides hilices? 11:47
@inkycatz could you summarize vs links, jeff101? :) 11:47
@inkycatz (much appreciated) 11:47
@Susume2 could you identify a handful of natural dimers with mixed hydrophobic and polar interfaces, that we could look at to get a feel for how they look?   11:48
vmulligan @MikeCassidytoo:  Yes!  Sheets are held together by backbone hydrogen bonds, too, and loops usually have largely satisfied backbone hydrogen bonds.  All structures can be stabilized by side-chain or side-chain/backbone hydrogen bond networks. 11:48
@MikeCassidytoo Thanks vmulligan 11:49
sboyken @jeff101, that is a good question: jeff is asking about proline backbone N and whether it forms h-bonds (as an acceptor).  The short answer is no 11:49
frood2IRC the current seg info does not refer to HBnet…is that correct? 11:50
sboyken Thanks vmulligan! Yes, @MikeCassidy H-bonds occur in all protein structures, if not with other protein atoms, with the water (solvent) 11:50
jeff101 sboyken, if you could post a link for the long answer about proline backbone N, I would like to see it 11:51
jeff101 perhaps when Chat is over 11:51
jmbrownlee333 how about heteroatoms that may or may not be charged? I feel like histidines in aprticular have more options than foldit gives credit for, 11:52
bkoep @frood, correct, the Segment Information panel does not include any information from the HBNet filter 11:53
frood2IRC kk - ta 11:54
sboyken @jeff101, definitely (working on it) a more detailed answer to your question is that the backbone N in Pro does loes it's H because it is making 3 bonds when it is part of the polypeptide chain.  And the N does have a lone pair of electrons that could act as an h-bond acceptor; however, I believe this is rare (I'll post a better asnwer in reponse to your post 11:54
@inkycatz I think jmbrownlee’s question is a good one to wrap up the official chat on. Anyone want this? :) 11:55
vmulligan @jmbrownlee333:  Yes, histidines are a bit of a problem in Rosetta and in Foldit.  They have an isoelectric point pretty close to blood pH, so they can be neutral or positively charged.  Since proper determination of the charge state of a histidine would require a quantum mechanical calculation of the electric field in its vicinity, we can't reliably predict the charge state of a histidine in our real-time simulations.  For simplicity, we treat them 11:55
LociOilingIRC @vmulligan, you were cut off again 11:56
blivens @jeff101 Wikipedia does mention that proline "cannot act as a hydrogen bond donor, but can be a hydrogen bond acceptor", but I think it's very weak. Good question! I'll also look forward to the long answer. 11:57
vmulligan Proline can be a hydrogen bond acceptor via its carboxyl oxygen, not generally via its nitrogen. 11:57
LociOilingIRC we missed the last part of the answer wrt histidine 11:58
frood2IRC vmulligan: fantastic statement above…I don’t doubt it’s authenticity…..but I think the average newb here will be lost.   But made me smile :) 11:59
@inkycatz @vmulligan - it cut off after “For simplicity, we treat them" 11:59
@inkycatz just fyi in case you had more there :) 11:59
vmulligan The short answer regarding histidine is that it's hard to calculate its exact charge since the charge depends on its local environment. 11:59
@inkycatz ah! 11:59
vmulligan But for simplicity, we treat them all as neutral. 11:59
@inkycatz thanks! :D 11:59
@inkycatz I’d like to thank all our scientists and guests for dropping in today - your time is always appreciated! Our next science chat will be in September, and I would like to thank everyone for coming too. Please look forward to a return of Black Belt Folding in the next month or so, several of us are working hard to bring that back! 11:59
frood2IRC :) 12:00
vmulligan It's probably relatively safe, since at blood pH, most of the side chains will not be charged.  Just below neutral pH, though, the histidine side-chain can be positively charged. 12:00
vmulligan Take care, everyone!  Bye! 12:00
sboyken @jmbrownlee333, yes (thanks vmulligan!) His does have more possibilities than we "give it credit for", to build on vmulligan's answer, the # of hydrogen atoms can change in His depending on the pH and solvent and local environment.   12:00
sboyken These effects are more complicated than what ROsetta and Foldit can currently capture well 12:00
@inkycatz (As this is the end of the official chat, some of our scientists may stick around, but be mindful they may drop off to head back to work  :) 12:00
frood2IRC u’e clearly not seen MY blood ph…hehe 12:00
@Madde a short question: do the contact predictions of 1114 come from Gremlin? 12:00
betahelix Thank you, everyone! Keep up the great folding... 12:00
blivens Great chat! I had fun. 12:01
jeff101 thanks to all 12:01
LociOilingIRC thanks 12:01
frood2IRC thx guys…good science chat….helpful. :) 12:01
jeff101 any answers for Madde? 12:02
bkoep Everyone can look forward to a blog post with more details about hydrogen bond networks, including some good examples for positive reinforcement 12:02
bkoep ... coming soon! 12:02
@Susume2 examples, yay! 12:03
@inkycatz Examples are great :D 12:05
sboyken yes, I'm really excited about the hydrogen bond networks I've been seeing designed by y'all!  (I'm also happy to stick around for a little bit if anyone as any additional h-bond / network questions) 12:05
bkoep @Madde, yes, we use Gremlin for contact predictions 12:05
jeff101 is there a way to read filter scores from within LUA? 12:06
jmbrownlee333 sorry? whats Gremlin? did I miss something. 12:07
jeff101 Madde asked earlier about it 12:07
@Madde thanks, so we can expect some more of these in the future? 12:07
@Susume2 12:08
jmbrownlee333 thx. 12:08
jeff101 in Rosetta @ home, does it always start like an extended b-strand, or can other initial structures be used? 12:08
jeff101 can protein complexes with more than one monomer be done with Rosetta @ home? 12:09
bkoep @Madde, we got a lot of feedback from players who enjoyed contact map puzzles and wished to see more of them, so we will try to post more in the future 12:09
@Susume2 baker lab's use of coevolution data (in Gremlin) for contact prediction helped them do well in CASP 11 12:10
jmbrownlee333 Is 1114 close to being solved. Is there data of some sort, xtal,EM,NMR? 12:10
bkoep Unfortunately, these puzzles are difficult to come by, and the native structure is rarely known—so the puzzle results are usually less informative 12:11
bkoep @jmbrownlee, as far as I know, nobody has any structural data for the protein in Puzzle 1114 12:12
bkoep I've got to head out—thanks everyone for a great chat! 12:13
@inkycatz thanks for coming by! :) 12:13

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