Developer Chat

@inkycatz Welcome to 2017, and the first foldit science chat of the year! 11:59
@inkycatz We have a lot of questions in the pile - thank you ALL for your contributions. :) 12:00
@inkycatz We have gone through the big list 12:00
@inkycatz And we have prepped some stuff ahead of time. 12:00
@inkycatz We also had to cut the list down a bit. 12:00
@inkycatz So if you don’t see your question, take heart. Yes, we saw it. We saw all of them. :) 12:00
@inkycatz Chances are if you don’t see your question: 12:00
@inkycatz It will be either something we’re now planning for a blog post later 12:00
@inkycatz Or more development vs science and moved to later 12:00
@inkycatz Or its something that maybe needs more looking into. 12:01
@inkycatz As usually happens, 1-2 side questions are fine but only if they’re quick ones, 12:01
@inkycatz please no multi-part “followups” today, or hold those until the end. 12:01
@inkycatz As a reminder, chats are logged and we post the log afterwards. 12:01
bkoep Hello all! 12:01
@inkycatz Welcome to bkoep, free_radical, and jflat 12:01
@inkycatz joining me today :D 12:01
free_radical :) 12:01
@jflat06 Hello! 12:02
@inkycatz Thanks all for coming - let’s get this started! 12:02
@inkycatz If rebuild always works with a fragment length of 3, what does a rebuild of length 2 do?  (This one’s for you, jflat) 12:02
@jflat06 In this case, rebuild is not actually inserting a fragment at all. It just places a cut, and then does a loop closure to close it again. 12:02
@inkycatz Awesome. 12:03
@inkycatz It sounds as if rebuild picks its fragments based on amino acid sequence, while remix picks fragments based on shape alone. Is this correct? 12:03
@jflat06 I'm not sure exactly why this functionalty was added originally 12:03
@inkycatz (oops thought you were done :) 12:03
@jflat06 but it was kept in in order to preserve existing functionality 12:03
@jflat06 Rebuild picks its fragments based on the labeled secondary structure 12:04
@jflat06 Remix picks the fragments based on the length and endpoints of the desired fragment. 12:04
bkoep Yes, Rebuild selects fragments based on AA sequence and secondary structure assignment; Remix selects fragments based on the position and orientation of the selection endpoints. 12:04
bkoep But that's not the whole story... 12:04
bkoep There is a bigger difference between the two tools 12:05
bkoep Rebuild selects a large number of 3-residue fragments, and builds a loop by stringing them together randomly. Of course, the endpoints of the new randomized loop won’t align to the original selection, so Rebuild often has to make some drastic changes in order to insert the new loop—this can result in some really bad backbone geometry. 12:05
bkoep Remix, on the other hand, looks at your selection and tries to find a fragment of the exact same length with endpoints that already match up almost perfectly. The new Remix loop may not be compatible with the amino acid sequence, but the backbone geometry is much better. 12:05
@jflat06 Yeah - since the fragment fits "naturally", it doesn't end up getting destroyed by trying to MAKE it fit 12:07
@jflat06 like it does in rebuild 12:07
@inkycatz So rebuild does pay attention to the current secondary structure but not remix? Or do they both pay attention to it? 12:07
frood2IRC I’m sorry to interrupt - but I think many would like this made very clear…..rebuild considers sidechains? remix does not? 12:07
bkoep Rebuild considers the identity of the sidechains (but not their shape, or "rotamer") 12:08
bkoep Remix could not care less about the sidechains 12:09
free_radical Brian, could you let us know what scientists prefer to use? Remix or rebuild in their designs? 12:09
free_radical sorry, @bkoep 12:09
frood2IRC kk - I don’t believe this is how many understood it - thx bkoep 12:09
LociOiling thanks, great info 12:10
bkoep @free_radical, we definitely prefer Remix for designs, because it gives better backbone geometry and we can always tailor the AA sequence to match the new loop 12:11
bkoep In prediction puzzles, Remix probably isn't so helpful, because the AA sequence is fixed 12:12
@jflat06 That's part of the reason that we didn't allow Remix on prediction puzzles for a while 12:12
@jflat06 However, we came to the conclusion that Rebuild probably wasn't doing any better, so we decided to allow it as wel 12:13
@jflat06 well* 12:13
@jflat06 It's also worth noting that while Remix does not consider AA sequence, if you are forming your loops from the correct portion of a prediction puzzle 12:14
@jflat06 then there's a reasonable chance there will be some fragments in there that are suitable or close 12:14
bkoep And to clarify about secondary structure, only Rebuild considers secondary structure assignment 12:15
Skippysk8sirc so if remix changes puzzle so it no longer passes filter, we have a bad fold? 12:15
@jflat06 Are you saying you have a fold that passes filters, and when you run Remix on it, it no longer passes them? 12:17
Skippysk8sirc yes -- lost over 300 points on one run 12:17
bkoep @Skippy, there could be a few things going on here 12:17
@jflat06 Which filter was being violated? 12:17
Skippysk8sirc I think ideal loops (you have examples from Gitwut & I on some issues to improve there) and residue filter 12:18
bkoep Yes, I think this falls within expected behavior, but I'll have to take a look at your specific examples 12:19
Skippysk8sirc there is a thread. it can be separate from this 12:19
bkoep First, there is no guarantee that Remix will return a loop that satisfies the Ideal Loop Filter 12:19
bkoep ...right now, the Filter is more stringent than the Remix database 12:19
@inkycatz Does remix always return the same set of results for a given selection and a given pose? In contrast, does rebuild return a random list of results? 12:20
bkoep But it's also important to note that the Ideal Loop Filter relies on secondary structure detection. If Remix causes a beta strand to swing away from a beta sheet so that it no longer forms hydrogen bonds, the filter will recognize that lone strand as a loop (probably a non-ideal one). 12:20
LociOiling the lone strand state 12:21
LociOiling (sorry) 12:21
frood2IRC I think SKippy’s point is more extreme - it has been proven posddible to have an ideal loop for the filter whilst having severe Bacbone issues within the loop. 12:21
bkoep So, if the Ideal Loop Filter is complaining that you "broke" a good loop, it might just be because you altered the secondary structure adjacent to that loop 12:21
bkoep @frood, yes, and some of this is actually related to the sequence 12:22
bkoep There are certain backbone configurations that allowed for some amino acids and not others 12:22
bkoep Proline, in particular, can adopt a backbone that is normally flagged as an "issue" for other residue types 12:23
@inkycatz Shifting gears a little…. 12:24
@jflat06 Also keep in mind that the ideal loop filter is checking some very specific things - the backbone torsion angles of the residues along the loop. There are other degrees of freedom that can be altered in Foldit that the filter does not check - such as the inter-residue bondlengths and bondangles. 12:24
bkoep And the Ideal Loop Filter only considers the Rama colors of your loop; it is perfectly possible that there is a poorly-closed cutpoint somewhere in the loop 12:24
bkoep @jflat06, exactly 12:24
frood2IRC kk 12:25
@inkycatz While free_radical is in the chat we had a question for him 12:25
@inkycatz Or a couple really 12:25
@inkycatz Any news about the TB and MD challenges (Puzzles 1258, 1261, 1291, 1293b, 1311, & 1312)? 12:25
@inkycatz Will there be sequels to these anytime soon? 12:25
@inkycatz Will we be able to use results from the original puzzles in the sequels? 12:25
@inkycatz Take it away, free_radical :) 12:26
free_radical The short answer is yes, there will be sequels to all of them. We currently are in the process of creating new blog posts with them 12:27
free_radical The colloboration has extended to include more scientists at Stanford to take Foldit players solutions into MD to test their stability 12:27
free_radical We also plan on letting players design the TB proteins for stability 12:27
free_radical and make does proteins in lab, in order to get crystal structures 12:28
free_radical Look for a blog post with this information soon! 12:28
@inkycatz My inbox waits happily for this! 12:28
free_radical As a final step, we plan on also designing small molecule drugs for it, with Foldit 12:28
free_radical The great news is that we have permission from Eli Lilly and IDRI to do all of this great work 12:28
@inkycatz That’s a good update. Thanks! 12:30
@inkycatz Next up: Have there been enough results from the design puzzles using the Blueprint tool to draw any conclusions? 12:30
bkoep We’ve been running some more analysis on the recent Blueprint puzzles, and the results are just starting to trickle in. 12:31
bkoep We've been plagued with some computer downtime here at UW 12:31
@jflat06 (also, the website downtime from last month set back our analysis a bit) 12:31
@jflat06 Though it's worth mentioning that our servers stayed up through a nasty power outage that affected the Seattle area recently! 12:32
@inkycatz That is pretty nice, I have to say. 12:32
@inkycatz A suggestion that came from the thread: make a selection and moving a few atoms rama map! I want to move a group of atoms. and sort them. 12:33
bkoep So, unfortunately, I can't say anything about how designs are looking since we introduced the Blueprint Panel 12:33
@inkycatz Soon, hopefully. 12:33
bkoep Yes, this is not the first time we've heard this request about moving selections of atoms on the Rama Map 12:33
bkoep From a biochemical perspective, I have to admit that the desire for this feature is puzzles me; but it is certainly possible, and it is on the to-do list for the Rama Map! 12:34
@inkycatz How closely do the models parallel actual protein dynamics? How many proteins developed by users have been formulated? 12:34
bkoep I'm not completely clear on what this question is getting at, but I can try 12:35
bkoep First, I’d like to clarify that Foldit is not optimized to model protein “dynamics” (that is, the thermal or kinetic motions of proteins) 12:36
bkoep Foldit is most appropriate for modeling the native state of a protein, or the thermodynamic energy minimum. In contrast, "molecular dynamics" tools are designed to model how proteins move in solution. 12:36
bkoep About "formulated" proteins: protein models created by Foldit players are used in different ways. 12:37
bkoep Models from structure prediction puzzles, like the famed MPMV protease puzzle or the more recent TB puzzles, are used by researchers to make sense of biological experiments. 12:37
bkoep Or these models might be used to form testable biochemical hypotheses 12:38
bkoep Protein designs by Foldit players have a much more interesting fate 12:38
bkoep Proteins designed by Foldit players are actually brought into the wet lab, where we express the proteins in E. coli and conduct different experiments to characterize their structure. 12:38
bkoep Since we started testing Foldit designs in 2013, we’ve attempted to express about 80 proteins designed by Foldit players. 12:39
NinjaGreg How many of those were "successful"? 12:40
bkoep In fact, we're about due for another blog post on Foldit designs in the wet lab; look out for that soon! 12:40
@inkycatz Are selected foldit puzzles presented to the community in advanced, near final states? 12:41
bkoep @NinjaGreg, well it depends on how you define "successful" 12:41
bkoep But by the most stringent definition, probably about a dozen 12:41
LociOiling not a bad batting average 12:43
spvincent It would be nice to know which these proteins are: give us a better a sense of what are considered interesting 12:43
@inkycatz (sounds like a good blog post or video :) 12:43
spvincent And let us know the results, even if "unsuccessfull". Like erterna 12:43
NinjaGreg Yes, it might help us know how to make better folds 12:46
Susume2 does stringent definition of success mean you have a solved crystal struture or NMR? 12:47
bkoep @Susume, unfortunately, no 12:47
Susume2 :-( 12:47
bkoep You're right, that would be the most stringent definition 12:47
tokens I would also be nice to know more about each denovo prediction puzzle when that puzzle is a foldit design (why is this particular design chosen, which puzzle does it come from...) 12:48
@inkycatz (we do have a few questions to get through still, folks, although these followups are great) 12:48
bkoep About the question on presenting puzzles in "advanced, near final states"... 12:49
bkoep I think we're missing part of that question here, and it's a little unclear what is intended 12:50
@inkycatz oops 12:50
@inkycatz Given the hyper-astronomical number of intermediate folding stages and final states, even if all the people on the planet were foldit players, there would be little probability of anyone achieving strong solutions. Are selected foldit puzzles presented to users in advanced, near final states? 12:50
bkoep But I think the best answer is that we prefer to let Foldit players approach a puzzle as a blank slate 12:50
@inkycatz (there you go, sorry about that wonky pasting) 12:51
bkoep Sometimes we provide computational models from other programs that Foldit players might be able to refine 12:51
@inkycatz Here’s the next one when you’re ready for that:  Are thermal perturbations incorporated into the design of the game? Perhaps by wiggle power or clashing importance? 12:53
bkoep We haven't incorporated anything like that explicitly. Wiggle is a simple energy minimizer, and does not make any attempt to model thermal perturbations. 12:55
bkoep However, Foldit players have developed their own scripts for “perturbing” a model in order to explore conformational space 12:55
bkoep There was a whole paper about the Blue Fuse recipe 12:55
@inkycatz Next up: I’ve noticed the top scoring monomer design puzzles (usually a linear-looking structure with 2 helices and 3 sheets) get recycled back as de novo puzzles: I've also seen these structures in Rosetta@home. These structures are clearly interesting but it would be good to know why. 12:56
bkoep Right! We use Rosetta@home to validate Foldit designs that we think look promising. We wrote a blog post about this process at some point... 12:58
bkoep Recently, we’ve been challenging Foldit players with the same task as Rosetta@home, by using Foldit designs for De-novo Freestyle puzzles. We’re curious to see how Foldit results compare to Rosetta@home results. 12:58
@inkycatz Is there a plan to expand the Blue Print and remix databases? 12:58
bkoep Maybe 12:59
bkoep The current BuildingBlocks represent canonical loops that are significantly over-represented in native proteins; so they tend to be very stable structures. 13:00
bkoep We may expand the BuildingBlocks selection somewhat. But this would mean including non-canonical loops, and might, in a sense, dilute the canonical BuildingBlocks. 13:00
@inkycatz And our last question although we have SO many (again thank you everyone): Will there ever be any new H-bond network puzzles? 13:00
bkoep A similar argument holds for the Remix database 13:01
bkoep Expanding the database can decrease the quality of the Remix loops 13:01
bkoep And yes, there will definitely be more H-bond network puzzles! 13:02
bkoep Coming soon, hopefully. It’s clear that the current filter needs to be tweaked a little, so that it better encourages networks on the interior of protein. But we do indeed have intentions to run more HBNet puzzles. 13:02
@jflat06 I think the biggest concern is really size - the Remix database is already quite large, and to get meaningful improvements in the fragments, we would have to expand it by several orders of magnitude. 13:03
@jflat06 (for the remix database) 13:03
@inkycatz This ends the formal part of our chat - look for a number of sweet, information filled blogs and other postings about things we just didn’t have official time to cover. :) Thanks again to the team for pausing their busy schedules to come chat and all of you for the material we could probably go on for hours about. 13:03
@inkycatz And for coming today. 13:04
bkoep Thanks everyone for your questions! I'm sorry we didn't get to all of them today! 13:04
@inkycatz The chat log WILL be posted soon, no worries. There’s a lot of info in there now to go through and we know you’re going to be eager to do so and prep all those followup questions :) 13:04
free_radical It was great chatting with everyone! 13:04
@jflat06 Thanks for coming! 13:05

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

Supported by: UW Center for Game Science, UW Department of Computer Science and Engineering, UW Baker Lab, DARPA, NSF, HHMI, Microsoft, and Adobe