Developer Chat

inkycatz It’s time for us to get started with our Science Chat for March. :) 13:59
inkycatz jflat06 - are you back from lunch and ready to log? 13:59
bkoep Hi everyone! 14:00
BletchleyPark hi bkoep, inky 14:00
inkycatz Let me go check and make sure jflat is over there. One sec! 14:00
inkycatz ok we’re good to roll 14:01
jflat06IRC I'm here, was just finishing up the last of my lunch 14:01
inkycatz Ha! Okay. :) As you all probably are aware, chats are logged, so if you have to go don’t worry we post the log afterwards. 14:01
inkycatz Question priority goes to those who posted ahead of time. 14:01
inkycatz We have a lot of questions to get through - I know there will be side questions so if you can please hold them to the end that will help us get through folks who posted questions ahead of time. 14:02
inkycatz Let’s dive right in then. :) 14:02
inkycatz We’ll start with some Rama stuff today, since that’s the hot new thing 14:03
inkycatz The Rama plot is used for all kind of amino acids. Are there significant differences when you look at the Rama plots of the individual amino acids? Is it better to be able to see the Rama plot of a specific acid? 14:03
bkoep Yes, the Rama Map 14:03
bkoep I'd like to start off with a little backstory for the Rama Map, if that's okay 14:03
inkycatz Go for it! 14:03
bkoep There are two motives behind the Rama Map: 14:03
bkoep 1) There has been a longstanding desire to give Foldit players more “hands-on” control of protein conformation. Tools like Wiggle or Rebuild allow players to shape proteins only indirectly. The manual Pull tool is difficult to control in 3D space, and behaves non-intuitively because the dragging movements have to be transformed into torsion space. 14:04
bkoep (Because bond angles are fixed, a protein cannot bend like your arm bends at the elbow; bonds can only rotate, like your forearm might turn a doorknob. To “drag" a residue from point A to point B requires the cooperative rotation of many different bonds, which only like certain rotations.) 14:04
bkoep The Rama Map allows players to manipulate these rotations directly. I personally think it’s a more logical way to think about protein structure, and I hope it will help newcomers understand the concept of bond rotation. 14:04
bkoep This is also why I like the viewport at the top of the Rama Map—even if it’s not useful to veteran players, I think it clearly illustrates the way that proteins are able to move. 14:05
bkoep 2) Loops are a problem in Foldit designs. We know that certain “ideal loops" occur more frequently in nature than you’d expect, and follow certain ABEGO patterns. The Baker lab has had a lot of success using these ABEGO patterns to design proteins. 14:06
bkoep For a while now, we’ve wanted to incorporate these ABEGO patterns into Foldit. It’s difficult to explain the ABEGO concept without a Ramachandran plot, which was also on the to-do list, so we incorporated the Ideal Loops gallery into the Rama Map. 14:06
bkoep It seemed natural that, if a player wants to put a residue in a certain ABEGO region, he or she should simply be able to drag the point on the Rama Map into the desired ABEGO space. 14:07
bkoep Now, getting to Timo's question... 14:08
bkoep Different amino acids do have different Ramachandran plots, and it is useful to look at the different Rama plots for different amino acids 14:10
bkoep In the Foldit Rama Map, you can do this by selecting a residue on the map (or by selecting a residue in Selection Mode) 14:11
bkoep If I select a TYR, for example, the Rama Map will change to show the specific plot for TYROSINE, and will display "TYROSINE" in the bottom right corner of the plot 14:12
inkycatz Ok, next up: Y'all gave us the basic information for the Rama map, but can we know how we should or could apply it and how to use it? Y'all also gave us the possible native states for the loops, but how would we replicate them? We're also able to click on a plot and drag it around. What's the purpose of this? It seems like it's another way to pull the protein. Is there a preference between plots being scattered and plots being clustered? 14:13
spvincent What does ABEGO stand for? I must have missed the definition 14:14
bkoep Ah, good question @spvincent 14:14
bkoep Each letter represents a different region of the Ramachandran plot 14:15
bkoep I think something like: 14:15
bkoep A - alpha helix 14:15
bkoep B - beta strand 14:15
bkoep E - extended (?) 14:15
bkoep G - glycine-preference 14:16
bkoep O - (a rare special case that we don't often encounter) 14:16
bkoep To @drumpeter's question, I'm actually more curious to see how Foldit players apply the Rama Map. 14:17
spvincent sounds slightly contrived :) 14:17
bkoep You could imagine building a protein by hand, walking along each residue and setting it's specific (phi, psi) position on the Rama Map. 14:18
bkoep Or you could use it to spot-check specific regions of a mostly-folded protein. Unfortunately, you will probably have to use cut-points to limit large lever-arm effects 14:19
jflat06IRC I don't know if I would say it's contrived - these regions are defined from statistical averages across many naturally occuring proteins. bkoep can correct me if I'm wrong. 14:20
bkoep @jflat06, that's correct, the regions correspond to clearly-delineated populations of points in native Ramachandran plots 14:21
spvincent I meant the acronym itself 14:21
bkoep To the last question, there is no real preference for how scattered or clustered your points are on the Rama Map, so long as they mostly lie in the colored regions 14:22
inkycatz Next on the list today: R plots, like ED, seem to be most useful when used with crystallography. For us non- scientists, can you help us understand when these tools are more useful than other modes of studying a protein? 14:22
bkoep I'll assume that "R plots" is short for Ramachandran plots? 14:25
inkycatz I think so yeah 14:26
inkycatz Skippy? Is that right? 14:26
frood2IRC I think u can take that as read 14:26
inkycatz Yeah. 14:27
bkoep Hmm, I'm not really sure what this question is getting at 14:27
bkoep There are limited ways to study protein structure 14:27
bkoep X-ray crystallography I think provides the richest source of data, so it's often the preferred method of studying a protein's structure 14:28
inkycatz Susume asks: What is the auto structures tool based on - is it phi and psi angles? If not, could it be made to use phi and psi angles so it would show us what the rama plot shows (which residues are in a good SS shape for that aa and which are not)? 14:29
bkoep But in general, we use whatever we can get. Crystallography is also very resource intensive, and something of a lottery. If we can't get crystallography data, we'll use whatever else we can obtain 14:29
bkoep Yes, the "Idealize Secondary Structure" tool only affects (phi, psi) torsions—which is what we plot on the Rama Map. 14:31
LociOilingIRC are we talking about idealize or auto structures? 14:32
Susume2 auto struct is in the actions menu and applies sheet/helix/loop to your backbone based on ... something 14:33
LociOilingIRC right, is that what your question was about? 14:33
Susume2 yes 14:33
bkoep Ah, sorry, I had forgotten about auto struct and misinterpreted your question 14:34
jmbrownlee333 I would say without data, all model building is just idle speculation. And ED is the main kind of data, that 'proves' our models. My 2 cents. 14:34
jeff101 bkoep, were you talking above about button 2 or 5 in the selection interface when you said "Idealize Secondary Structure" tool? 14:34
bkoep Let's see, I think auto struct uses the DSSP algorithm, which relies on hydrogen bond patterns rather than (phi, psi) torsions 14:35
Susume2 ty 14:36
inkycatz Does the mutate tool take into account the probability of concurrence of specific AAs in general (probability in nature) and/or for specific SS (preference for helix etc)? (There are also more mutate questions but maybe we can revisit those at the end.) 14:36
bkoep So, @Susume, you would like auto struct to show something about the backbone score? Or something more specific? 14:37
Susume2 I was thinking it could be the poor man's rama map in non-design puzzles if it were based on phi, psi - but it soudns like you will consider adding rama map tool to other puzzles eventually 14:38
LociOilingIRC (auto structure is still broken, converts last segment to ligand) 14:39
jeff101 in the sel interface, button 2 says "Idealize peptide bonds!" while button 5 says "Idealize the secondary structure!" 14:39
bkoep I see. Yes, there's no reason the Rama Map needs to be limited to design puzzles, so we'll probably make it available in all puzzles in the future 14:40
alcor29 What about using "Idealize SS " tool to show the ramas directly, since it is already using phi, psi? 14:40
bkoep Thanks for the reminder @LociOiling! That's still on our radar! 14:40
frood2IRC will making the rama map available to all puzzles make all puzzles slower to play? 14:40
LociOilingIRC :-) 14:41
bkoep @frood, no, the Rama Map should not slow down general play 14:41
Susume2 and I think we can use abego coloring in non-design puzzles now to see what is in a good rama zone 14:41
inkycatz (Just a reminder we do still have a good number of questions left :) 14:41
bkoep Right, Bruno's question 14:42
bkoep Yes, there is a part of the Foldit score that considers amino acid preferences for a provided backbone. However, this is not always reflective of what we want in designs. 14:42
bkoep A good example is alanine, which is normally strongly favored in an alpha helix. We don’t like a lot of alanine in designs because they contribute to a “slippery” protein core, rather than a rigid well-packed core, so we artificially penalize alanine. 14:44
inkycatz Did you want to do the other mutate questions now or just come back to those? 14:45
inkycatz (Up to you, really.) 14:45
bkoep @jeff101, I meant button 5 above, which says "Idealize the secondary structure!" 14:46
bkoep Sure, Bruno's other questions? 14:46
inkycatz Question 1: is there any known generic "path" in nature or is it completely random and independent from context? thus DNA codes for a "nature selected" protein on a deterministic way independent from the context. Mutations are simply random and the natural selection does the job of retaining the useful ones. 14:46
inkycatz Question 2: For anti bodies, how does the nature work? does it try mutating existing antibodies until it finds a mutation that works? (then we would find a lot of "waste" in blood). Or does it "recognize" a virus protein (as we human do on the models given to us), identifying possible weaknesses, THEN sythetizing a "deterministic" antobody, this with limited waste "mutations"? 14:47
inkycatz there you go 14:47
bkoep Question 1 is getting at (I think) some complex biology 14:48
bkoep I think most people would agree that mutation in proteins is context dependent, in that a mutation has to play nice with all of the other residues currently in that protein 14:50
bkoep Question 2 is easier to answer 14:52
bkoep Antibodies are pretty much produced randomly, in large numbers, with a lot of what you call "waste mutations." If a particular antibody mutant is successful, then the body has ways of amplifying that particular mutant. 14:54
inkycatz Next a triple pack of beta barrel questions from Virgo. :) 14:55
inkycatz How often does a beta-barrel appear in nature? 14:55
inkycatz How many sheets does the tertiary structure require as a minimum? 14:55
inkycatz Should we focus on creating this tertiary structure in Foldit often? 14:55
bkoep Yes, Virgo's questions are in response to the blog post here: 14:56
bkoep 14:56
bkoep Beta barrels are very common in nature! However, many of them are membrane proteins, which we do not typically model in Foldit. 14:58
bkoep Although there are plenty of examples of water-soluble beta barrels 14:59
bkoep As for minimum size, I don't think I've seen any beta barrels with fewer than 8 strands. 15:00
bkoep As discussed in the blog post, there are still some peculiarities of beta barrels that we don't fully understand 15:00
bkoep My intuition is that it would be difficult to design a high-scoring beta barrel in Foldit 15:02
bkoep Although, I have seen some very nice beta barrel designs from Foldit players... 15:02
inkycatz One last Q, then we’ll get to some updates. :) “Why do I get voids shown on the outside of the protein when all local segs are hydrophiles?“ 15:02
bkoep It might be that Foldit players can teach us something about beta barrel design! 15:02
inkycatz Maybe! :D 15:02
bkoep Yes, voids... 15:03
jeff101 how do you tell if a beta barrel design is very nice if it has a poor score? 15:04
bkoep Voids can actually be computationally expensive to calculate. We use some approximations and short cuts to speed up the calculation, but this results in some funny behavior, like voids on the surface. 15:04
bkoep @jeff101, it's no simple feat to close 8 strands into a closed barrel, while maintaining good backbone scores 15:05
jeff101 what do you look for to tell it is "very nice" ? 15:06
bkoep An 8 stranded beta barrel also includes 7 loops, which can be difficult to close 15:07
bkoep @jeff101, I'm mostly thinking of general characteristics of stable proteins 15:07
bkoep A well-packed hydrophobic core can also be difficult to design in a beta barrel 15:08
inkycatz And now the updates: Marburg binders, VEGF, UMich, and the abeta binder contest results. Feel free to tackle those in any order, btw. 15:08
bkoep Yes, I'm sorry I have no updates on VEGF 15:09
bkoep There's a chance that Marburg/Ebola puzzles will pick up again soon, so look out for that 15:09
bkoep We're polishing up the results from the UMich challenge 15:10
bkoep I think I mentioned before that Foldit players produced some excellent models, but there were problems with residues that didn't fit into the density cloud 15:11
bkoep Those residues caused some problems, and in future puzzles we'd like to allow players to trim unwanted residues 15:12
frood2IRC we were instructed that some did not 15:12
bkoep @frood, yes I don't mean that they should have been crammed into the density somehow 15:13
frood2IRC kk 15:13
bkoep When we compared the models to the raw x-ray data, those extra residues interfered, whereas the models would have fit the data much better if we had simply left the residues out 15:14
jeff101 like removing their AA's from the peptide chain or removing their density from the ED cloud? 15:15
bkoep @jeff101, I mean removing the residues from the peptide chain 15:16
jeff101 do they move so much they seem invisible in the ED cloud? 15:16
@MikeCassidytoo .. 15:17
inkycatz (We all still here?) 15:19
BletchleyPark we are 15:19
jflat06IRC @jeff, I think that's essentially correct - the data is gathered from many, many copies of the same protein. In regions where there is high variance (say, a loop that moves around a bit), the data will be less clear, almost like it's out of focus. 15:19
jflat06IRC imagine that you're taking a thousand pictures of a person who is moving just their left arm, and then averaging those pictures. Their left arm will be blurry while the rest of them will be clear. 15:20
bkoep Sorry! Yes, @jflat06 is exactly right 15:21
frood2IRC is that looking at them from the front or the back? 15:22
frood2IRC sorry - couldn’t help it 15:22
bkoep Well if it's like crystallography, we'd have to shoot them with x-rays from every angle... 15:23
frood2IRC :) 15:23
bkoep Anyway, I have to run! 15:23
inkycatz And with that, we’re able to wrap up our chat today! 15:23
inkycatz Thank you bkoep and jflat :) 15:23
jeff101 thanks for coming 15:23
inkycatz especially staying extra today 15:23
BletchleyPark 15:23
inkycatz and thank you ALL for coming! 15:23
frood2IRC thx to all 15:24
@MikeCassidytoo Good talk, thanks 15:24
bkoep Thanks all for the great questions! 15:24

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, Boehringer Ingelheim, RosettaCommons