Developer Chat

Scientist Chat

--- Log opened Thu Feb 23 12:54:08 2012
beta_helix Hello everyone! We will be starting the Scientist Chat shortly, Aaron will be here to talk about the Flu Puzzles and then we'll talk about CASP ROLL. 12:58
frood66 morning! *stands on toes 12:58
u-AQ awesome! 12:58
brianhixon morrning! 12:58
@auntdeen2 Attention all players...  the developer science chat will start shortly - please use the puzzle channels for all chat that is not related 12:58
DoodelThePlatypus Nice that it's morning there, it's  close to bed time here in Stockholm. 12:58
@tealight-irc ahh yes teh scientist chat..had actually forgotten about it lol 12:59
@auntdeen2 hi beta :-) 12:59
*** mode/#global [+o beta_helix] by Judecca 12:59
@auntdeen2 and good morning, tea 12:59
phi161 hi all 12:59
BletchleyPark hi beta 12:59
frood66 present and correct 12:59
mottiger back 12:59
mottiger sorry frood that i left, had a little issue here in the flat 13:00
@tealight-irc *passes evryone their coffee and muffin* 13:00
frood66 n mott 13:00
frood66 np 13:00
DoodelThePlatypus I have a quick question, don''t know if it's appropriate  here. When I start up FoldIt, I get the error: Unable to write output. 13:00
Aaron Hi Everyone 13:00
DoodelThePlatypus Ialso my puzzles were never saved 13:00
@auntdeen2 hi Aaron :-) 13:00
DoodelThePlatypus I deinstalled and installed again. 13:00
BletchleyPark hello Aaron 13:00
@marie_s in windows 7 dont install in programm files 13:00
Flimsor Too late 13:00
@beta_helix This is Aaron from the Baker Lab, he's been analyzing all your Flu results! 13:00
DoodelThePlatypus Oh where to then 13:00
spvincent Hi Aaaron 13:00
@marie_s c:\foldit 13:00
DoodelThePlatypus thanks 13:01
@CFC Hi Aaron :) 13:01
Aaron Yes, I'm the graduate student in Baker Lab who has been coming up with and analyzing the flu puzzles 13:01
lamoille hi Aaron 13:01
frood66 hey 13:01
@mimi2 hiya 13:01
spvincent So what can  you tell us about them? 13:01
karstenw hellooo 13:01
mottiger hey aaron, do you know what hilvert contributed to the project? 13:01
Aaron with the help of beta_helix 13:01
spmm2 can you tell us a bit more about fitness landscapes - like what sort of fitness? 13:02
frood66 baptism by fire  lol 13:02
Aaron So the way the analysis works is I get back the top scoring structures from beta_helix, but they are only numbered 13:03
Aaron so i don't know who did what 13:03
Aaron But I generally look at the too100 scoring models 13:03
Aaron which can be across many different teams 13:03
@beta_helix (this is the top100 models per puzzle) 13:03
Aaron whats interesting, is that many of you converge on the same solutions, and it's those convergences that I find useful 13:04
Aaron so yes, fitness landscapes! 13:04
Aaron fitness for protein binders is kind of two components 13:04
spvincent We're always being encouraged to  crreate more divergent solutions as a rule. 13:05
Aaron the first the ability of the protein to fold or it's stability 13:05
Aaron the second is the ability of the protein to bind to the target 13:05
Aaron so fitness = stability + binding 13:05
Aaron @spvincent: yes 13:06
Aaron we want to try and find as many convergent clusters as possible 13:06
frood66 are the scripts causing the convergence? 13:06
spvincent Or perhaps the result of group play? 13:06
Aaron we are basically looking for as many clusters of solutions as possible 13:06
@beta_helix @mottiger we are unfamiliar with hilvert's contributions to this specific project 13:06
@auntdeen2 or does the fact that many of us converge possibly have to do with our software & tools? 13:06
Aaron because these clusters are probably local or global minima in the overall fitness landscapes 13:07
@beta_helix what is interesting is that many different Foldit Groups all converge on the same solutions. 13:07
mottiger just saw in one article a comment from him, so i thought he is somehow involved 13:07
@tealight-irc "great minds think alike" 13:08
frood66 but those same groups use (generally )the same scripts beta 13:08
karstenw is that a good thing, bad thing, or just an observation? 13:08
f1pokerspeed Does anyone know how I can add recipes to my cookbook from the website? It won't let me 13:08
Aaron when I see these clusters, it means that many people find a particular solution, and THAT solution might be one(of possibly many) actual real solutionsp 13:08
@CFC with the mutate designer puzzles, I'm guessing there's a tighter smaller range of convergence - the freedom lies in moving the 'ligand' component 13:08
@auntdeen2 All players - global is being used right now for a developer science chat… please use the puzzle channels for all chat that is not related 13:08
@tealight-irc log into webiste to be able to get recipes 13:09
f1pokerspeed thanks 13:09
Aaron as a puzzle designer, we try to balance convergence with diversity 13:09
Aaron if everybody converges on the same solution, that really only gives us one thing to try 13:09
BletchleyPark Does the rule better score = better model apply ? 13:09
Aaron and if everybody has completely different solutions then that is too many things to try 13:09
TimovdL How many clusters of solutions did you find? 13:09
mottiger i'm a bit confused, are we talking about the flu or design puzzles in general or about foldit in general? 13:10
Aaron but perhaps we've been too restrictive in the puzzles we've given you 13:10
@beta_helix Right now we are strictly speaking about past Flu Puzzles. 13:10
@CFC But I am curious if there were specific schains at specific locations, that spawned clusters... say a Phe or Tyr? 13:11
phi161 too restrictive how? 13:11
Aaron @CFC: YES! 13:11
Aaron You've noticed 13:11
Aaron so many of these sidechains are inspired by antibodies that have been shown to bind the flu 13:12
@beta_helix @TimovdL it depends on the Flu Puzzle, for example the last one (where we just gave you a helix) came back with 1 cluster, but they were all very interesting (for the exact reason CFC just brought up) 13:12
Aaron we try to use nature as inspiration 13:12
spvincent How different were the Foldit results compared to those produced by rosett@home, assuming  you did that? 13:12
@beta_helix @phi161 perhaps we put in too many constraints, or in most Flu Puzzles the backbone is completely locked. 13:12
Aaron @spvincent: I found that the foldit players come up with much more reasonable solutions than by purely computational means 13:13
mottiger we cannot judge if this is positive or negative if you put so many constraints, this is up to you 13:13
Aaron both methods have there advantages and disadvantages 13:13
TimovdL In nature wouldn't the backbone be influenced by the blocker? 13:13
Aaron We tried doing helix design and docking on our computer clusters(the last flu puzzle) but were unhappy with the results 13:14
mottiger s 13:15
spmm2 do you try different binding sites or are they optimised already? 13:15
@beta_helix @BP that seems generally true, 80% of the time the highest scoring models pass Aaron's criteria for what "might be useful" 13:15
BletchleyPark thanks beta, have  you had success yet with a working model ? 13:16
Aaron @spmm2:  we have been trying to go after this same binding site aggressively because it is a location that if we bind to it, we think we can actually inhibit flu virus 13:16
Aaron so there are other places to bind, but we KNOW there is something interesting about this location 13:16
karstenw you can keep constraints intact if you give us models, like casp. this way you can control the number of clusters you get 13:16
spmm2 ah ha thanks 13:16
mottiger do you synthesise some of the solutions or just use them digital? 13:17
Aaron also, that particular binding spot is highly conserved across many different flu viruses 13:17
Aaron so if it works for one it might work for others 13:17
Aaron in terms of the last flu puzzle 13:17
@MikeCassidytoo It would be good if there was an manual way to gently move parts of the puzzle - maybe the arrow keys 13:17
Aaron I think it was probably the most useful yest 13:17
Aaron yet 13:17
spmm2 we had a bit more flexibility on that one 13:18
Aaron we know that we like a helix in that position but it's really only me in the lab trying to figure out how to fit it 13:18
Aaron your top100 puzzles were totally awesome in coming up with ways to interact that I hadn't thought of 13:18
Aaron so we are going to take a few of the top scoring designs, and try to build a protein around it and test if it sticks to the flu 13:19
frood66 good with a hammer here :) 13:19
@marie_s have you test if they will make an helix with our solutions? 13:19
Aaron So, do we synthesize the solutions?  Well we use them as inspiration on things we do synthesize 13:20
@auntdeen2 when was that last flu puzzle played?  during the time that "wiggle" was bugged?  if so, you may want to present it to us again 13:20
@auntdeen2 we might be even more awesome ;-) 13:20
mottiger ah i remmeber the last one, sorry it is some time ago, it was just this little helix, and you think we brought up good solutions but just the helix isnt usefull so you need some decoration on it 13:21
Aaron and also, as we get better as using and designing puzzles we are trying to streamline moving from foldit players solutions to actually synthesizing them 13:21
frood66 absolutely 13:21
@beta_helix @auntdeen, it was posted in January: http://fold.it/portal/node/991577 13:21
urlbot Long URL from beta_helix has been shortened to http://tinyurl.com/7xdkvlx 13:21
spmm2 do you actually look at results in foldit - how do you examine the results? 13:21
@auntdeen2 then really, beta - it may be worth another round 13:21
@beta_helix in PyMOL 13:21
Aaron @marie_s: we are going to use a few of the top solutions combined with our computational methods to try and make some new binders, yes 13:21
Aaron BUT we are also going to give you a puzzle and let you try and do the same thing 13:22
Aaron which is build the backbone/scaffold of the whole protein around the helix 13:22
BletchleyPark can the models be quicker with less constraints ? 13:22
BletchleyPark or by sorting the constraints in a different way ? 13:23
Aaron @BletchlyPark:  What constraints do you not like? 13:23
Aaron We are basically trying to focus your efforts on a particular area 13:23
frood66 well the mission sounds interesting 13:23
BletchleyPark I do not know the internals of how constraints are processed, but I can imagine that they are rules 13:23
Aaron but the next puzzle will have lots of freedom in terms of how to design the scaffold 13:23
Aaron there are, and it is something that me and beta_helix talk about quite a bit 13:24
@beta_helix We don't want you to waste your time trying to fit the helix on the other side of the protein for example, because even if that scores better it is utterly useless... 13:24
BletchleyPark I have some background in raytracing, and one may apply rules there to speed things up regionally. The same methodology might apply to constraints here 13:24
@beta_helix but we also don't want to constrain you too much, either. 13:24
mottiger is it possible to make the whole protein 'smaller' so that you don't see all of the flu protein, i had the feeling that this flu puzzle was really hard for old machines 13:24
@beta_helix we tried to trim it as much as possible, and lock the sidechains that don't interact 13:25
spmm2 maybe an electron density type shape to show us where you want it be be? 13:25
Aaron so yes 13:25
Aaron ha 13:25
karstenw thats what i was thinking,  move constraints a bit and give us 5 models. this keeps within your boundaries, but gives us some wiggle room 13:25
Aaron I have suggested that and we are working on implementing that idea 13:25
Aaron but really it's an regional area that we want you to be 13:26
@beta_helix (we need to bug the developers for that one ;-) jflat06, you here? :-) ) 13:26
Aaron the problem is a different the structure prediction because we just want you to be in a particular region but you can do anything you want in that region 13:27
Aaron unlink structure prediction there is no one solution 13:27
Aaron If we gave you an 2d image of some designs that already worked, would that be useful as inspiration? 13:28
TimovdL Don't you want also binding in the surrounding area's? 13:28
karstenw but the constraint itself can be varied, leave eveything else alone 13:28
spmm2 pictures are always good for me :) 13:28
Aaron @TimolvdL:  There is a particular loop on the target that is used by virus to infect cells, so we try to keep you near that loop 13:29
@beta_helix would an image as a guide be sufficient for you? then we wouldn't have to enforce any constraints (or a lot fewer) 13:29
BletchleyPark give it a try beta 13:29
@auntdeen2 definitely worth trying 13:29
spmm2 yes - personally I would stay in that area because it was useful even if people were getting high scores binding in the wrong place 13:30
@MikeCassidytoo I think that unless the score was tied to the image we would follow the score 13:30
mottiger in ED puzzles you already have this align to something feature, can't this be implemented to realign the designed protein to that region that you want it to have? 13:30
TimovdL I the primary target is that loop, but maybe a bigger protein can give an even tighter coupling 13:30
TimovdL I mean 13:30
karstenw align to guide wont give you guys too much diversity? 13:31
BletchleyPark @timo: pins of a powerplug need to connect, but having two ends of the entire plug may yield more stability ? 13:31
@beta_helix @spmm the issue is if everyone else gets a high score by binding the wrong place, we might miss your solutions :-( 13:31
@beta_helix this is why we always put in some minimal constraints 13:32
BletchleyPark then keep those at least 13:32
@marie_s and recipes will lost our best 13:32
TimovdL Add bonus points for binding at least the correct place 13:32
@beta_helix sounds good, really loose constraints (that you might not even notice ;-) 13:32
frood66 sounds sneaky 13:33
karstenw i like that. loose constraint 13:33
@MikeCassidytoo Those of us with no bio knowledge can only follow the score 13:33
karstenw right mike 13:33
@beta_helix @timovdL it's basically the opposite, if you violate the constraints you get a big penalty, but if you are in the correct zone everything is fine 13:33
@CFC :) I remember the bond angle puzzles with bonuses... They were an interesting exercise 13:33
TimovdL same effect 13:33
spmm2 well lets try with mimimal constraints and you can repost if you want to tweak 13:33
@beta_helix yep... 13:34
@marie_s no way to make other place not attractive? 13:34
frood66 sounds good 13:34
mottiger or try different methods 13:34
@beta_helix @mike that way you can follow the score and know you are at least in the correct place! 13:34
spmm2 x marks the spot :) 13:34
@beta_helix @marie well, if you try the other place and your score gets horrible, isn't that unattractive enough? or do you mean visually? 13:35
Aaron @everybody: the overall goal of these puzzles to try to make small proteins that can inhibit the flu, something that has never really done before 13:35
Aaron it's is very cutting edge 13:35
@MikeCassidytoo Yes beta the penalty idea would work 13:35
@beta_helix I think on that note, are there any last questions for Aaron? 13:35
frood66 we have limited computing power - seems sensible to use it wisely 13:35
BletchleyPark 2aaron, would it also bind to the unpublished one that is in the news lately ? 13:35
Aaron @frood66:  ya'll have more than you think 13:36
frood66 haha - we wish lol 13:36
karstenw more power! lol 13:36
alwen <-folding on netbook 13:36
Toozinger westin!!! 13:36
mottiger installed foldit today on my lab computer and it was amazingly fast 13:36
Aaron @BletchleyPark:  The designs we are doing are against all group I viruses, which include H5, so you never know.... 13:36
@CFC we can only do what we do at the 'frontline' with what you give us, and we thank you for it 13:37
frood66 that's sneaky mott! 13:37
Aaron @BletchleyPark:  The region that you keep see is highly conserved across most all of the viruses, that's why you keep seeing it ;) 13:37
BletchleyPark thanks Aaron, it will be a high profile discovery if you find it 13:38
@beta_helix Ok gang, it's time to let Aaron get back to work! 13:38
Aaron Thanks everybody, I'm gonna head back into the laboratory! 13:38
BletchleyPark thanks Aaron 13:38
@auntdeen2 thanks, Aaron :-) 13:38
@marie_s thanks 13:38
karstenw bye aaron 13:38
lamoille thanks Aaron for your time 13:38
frood66 thx Aaron - appreciate yr time 13:38
@tealight-irc thanks for chatting with us aaron :) 13:39
mottiger enjoy aaron 13:39
@MikeCassidytoo bye aaron 13:39
BletchleyPark CASP R12.. 13:39
spmm2 I understand a lot more now thanks 13:39
Aaron Yep, I hope you enjoy the next puzzle, you'll really do a fantastic job..... 13:39
@beta_helix So one of the promises that we had made in the recent dev. chats was to not post any non-CASP ROLL puzzles until we got the client issues sorted out. 13:39
@tealight-irc 0.o 13:40
frood66 Hmmm 13:40
@tealight-irc how many are we getting? 13:40
@tealight-irc lol 13:40
@beta_helix Hopefully we can post that new Flu Puzzle today! 13:40
@auntdeen2 they are looking sorted, beta 13:40
frood66 u r kidding? 13:40
lamoille nice 13:40
@auntdeen2 but..... 13:40
tokens yeah, non-Casp is good 13:40
spmm2 the dev preview is pretty good be nice to get it to main 13:40
@auntdeen2 on quick caution - we didn't realize there was a problem in the last client until endgame 13:41
karstenw but stagger expiration day pls 13:41
@tealight-irc i noticed a tremendous improvement with client yesterday, was back to "normal" 13:41
mottiger same here tealight 13:41
frood66 e have (effectively) 21 current puzzles and u want to add? 13:41
mottiger much faster, huh? 13:41
@tealight-irc yes much faster and more consistant 13:41
@auntdeen2 yes - much faster 13:41
@CFC ...but the old rebuild would be a very welcome addition to the 'weaponset'.... 13:41
@auntdeen2 second CFC 13:42
@tealight-irc thirded 13:42
spmm2 they can keep working on the fixes but dev preview imo can go to main now 13:42
mottiger cannot say anything about the rebuild thing 13:42
@beta_helix so it would be ok to have 2-3 CASP ROLL puzzles up and 1 design puzzle at the same time (making sure the expiration dates are staggered... it's a bit harder with the CASP ROLL puzzles) 13:42
@auntdeen2 spm - it did 13:42
frood66 ready to hang up my boots here 13:42
karstenw keep in mind, the more puzzles you send. the less time is spent on each puzzle. 13:42
@beta_helix @karstenw of course 13:43
BletchleyPark @beta, CASP roll 12 seems to be 499 long ? 13:43
@auntdeen2 beta - the large casp puzzles are an issue 13:43
@tealight-irc with staggered dates it shouldnt be too bad 13:43
@beta_helix that is the issue: do we only post the CASP ROLL puzzles until CASP ROLL is over, or do you have time for design puzzles as well? 13:43
@auntdeen2 even before the 499 comes out 13:43
tokens I will look forward to a design puzzle for a change 13:43
spmm2 lets have the flu puzzle 13:43
frood66 499 - no way 13:43
spvincent It's the combination of too many puzzles and a relatively short time to work on each that is a problem. 13:43
@auntdeen2 agree, let's have it 13:43
@beta_helix don't worry, we will not be doing R0012 (it is 499 residues long and no de-novo, there are many templates for it) 13:44
@auntdeen2 and large & slow 13:44
@mimi2 I don't have the time for what we are getting already unfortunately 13:44
@beta_helix ok, we'll come back to design puzzle in a second, let's talk about CASP ROLL for a bit: 13:44
frood66 we are already overloaded inmy view 13:44
@tealight-irc can we have the flu one with a long time frame? 13:44
infjamc E.g. 30 days? 13:44
@tealight-irc yeah 13:44
@CFC and CASP10 isnt here yet frood... :) 13:45
mottiger multistage flu design like the casp puzzles? 13:45
frood66 either i do my best or not at all 13:45
tokens 30 days is too much I think 13:45
spmm2 it is selecting the template and then feeling anxious about picking the wrong one I don't like :( 13:45
@beta_helix one way to lighten the puzzle load would be to do away with the TOP Foldit predictions round. Currently, they give us 3 weeks for each target and we basically do: 1 week on 10 server models, 1 week on rosetta models, 1 week on the TOP predictions from the first puzzle. 13:45
@auntdeen2 what are you getting the most out of, beta? 13:45
@CFC there must be an easy way to cut that to 2 rounds? 13:46
@beta_helix As I mentioned in this post: 13:46
@beta_helix http://fold.it/portal/node/991921#comment-15968 13:46
urlbot Long URL from beta_helix has been shortened to http://tinyurl.com/6u73554 13:46
frood66 something has to give here 13:46
@beta_helix the important thing for these targets that have no homologs/templates is to try to get as much diversity as possible. 13:46
Voclav IMAGE: http://fold.it/portal/files/chatimg/irc_404663_1330033610.png 13:46
@auntdeen2 will your foldit submissions suffer if you do away with the top predictions round? 13:46
@auntdeen2 All players - global is being used right now for a developer science chat… please use the puzzle channels for all chat that is not related 13:47
@auntdeen2 thanks players 13:47
@beta_helix The goal for the TOP predictions round was to take an early version of a high scoring prediction and see if there is a better topology that scores even better. 13:47
frood66 as a self confessed addict i play - but not if I cannot play 13:47
TimovdL Did that work? 13:48
mottiger was this idea a success beta? 13:48
@auntdeen2 has that worked, beta?  significantly? 13:48
spvincent And how does it seem to be working so far, beta? 13:48
@beta_helix @CFC we could post all 15 server predictions in one puzzle but that might be way to overwhelming, and the Rosetta ones usually score better already so the other 10 might be ignored 13:48
@auntdeen2 why not do that and award no global points? 13:49
@auntdeen2 so that everygthing will get looked at? 13:49
@mimi2 timescales are such that unless you have a very powerful system you can only really look at one or two options 13:49
@tealight-irc many people wouldnt do it if there were no points in it for them 13:49
@Madde in sense of more diversity wouldn't it be better to get rid of the TOP prediction round and instead add a round de-novo? 13:49
@beta_helix generally, a better score is found from improving those top predictions, but everyone converges on that same model from the last round. 13:49
frood66 and newbies? 13:49
@auntdeen2 true, tea - but at this stage, many can't even look at one of the current 3 13:50
TimovdL Beta, I have understood that there are even more models. Maybe in stead of the top predictions, the best 3 of them? 13:50
TimovdL Of those other models 13:50
karstenw just stagger puzzles a bit more and introduce new puzzles a tiny bit slower and we can handle it i think, just the way it is. 13:51
@auntdeen2 kars - it's the CASP deadlines 13:51
@auntdeen2 not foldit 13:51
frood66 i can only be blunt - the current situation is alrady too heavy 13:51
@beta_helix @Madde & spmm's reply to my comment, that is why I always provide an extended chain in the alignment tool for every CASP ROLL puzzle... but freestyle scores rarely get as good as starting from server models 13:51
karstenw ah, then , idk :D 13:51
frood66 I have not the time to follow through on thing I would like 13:52
@CFC posting 2 targets a week recently has caused the chaos, but we can only react 13:52
@beta_helix @karstenw I asked the organizers to push their deadlines and they were able to do that for R0009, @frood66 that is why we are having this chat :-) 13:52
karstenw ty beta, more time is good 13:52
spvincent IHow about 10 days on the first 2 stages of the CASP Roll and skip the final stage of refining players solutions then if that doesn't seem to be generating anything new? 13:53
*** mode/#global [+o thomirc8] by ChanServ 13:53
@mimi2 An idea of what we can expect  would be helpful It was a shock today to find new puzzles with end dates earlier than the current puzzle 13:53
frood66 @beta - u think I don't know that? 13:53
@beta_helix So we could do away with the "top predictions/all-hand-ish round" or one way to decrease the amount of puzzles would be to combine that round with the 5 Rosetta Server predictions. 13:53
@tealight-irc mimi, yes i was surprised last night when i encountered them lol.. 13:54
BletchleyPark everyone will go with the rosetta ones.. 13:54
spvincent That sounds  like a possible solution. 13:54
@beta_helix @mimi I can post a schedule of future CASP ROLL puzzles... I try to give more time on larger targets 13:54
@beta_helix @BP that is exactly my fear 13:55
@tealight-irc that would be useful beta :) 13:55
karstenw i wouldn't combine them 13:55
frood66 qa bit of organisaton would be lovelly 13:55
@MikeCassidytoo It sounds like I should not be jumping on the alignment giving other people better scores but staying with my low scorinng alignement 13:55
@beta_helix general question: how much combination are you trying? Is partial threading being used often or do you mostly focus on 1 model at a time? 13:55
@marie_s one model or 2 for me 13:56
@MikeCassidytoo I do no partial threading 13:56
BletchleyPark I hardly use partial, i always try all models 13:56
mottiger no partial threading here 13:56
spmm2 how many more CASP Roll puzzles/targets are there? 13:56
frood66 partial - no way 13:56
@marie_s no partial, too hard 13:56
@mimi2 can't get it to work for me 13:56
@beta_helix http://predictioncenter.org/casprol/targetlist.cgi 13:56
urlbot Long URL from beta_helix has been shortened to http://tinyurl.com/86p6xu9 13:56
@MikeCassidytoo I do mess with different alignments 13:56
@tealight-irc im no good at partial threading so run thru the various models and pick on i like...granted i seem to pick the ones tha dont go far lol' 13:56
@CFC spmm, it's a rolling process, could be lots, could be some 13:57
frood66 partial is a waste of time 13:57
TimovdL One model, I just evaluate all and choose one or 2, no partial threading, not time enough 13:57
spmm2 I try every template no partial 13:57
@tealight-irc it seems many dont use partial...... 13:57
@CFC the fun will be 2 ROLL targets a week, during the madness of CASP10 13:57
@mimi2 I think only infjamc knows how to do partial 13:57
@beta_helix I always try to give you the most time on the first puzzle (since that one has 10 different starts) but the next 2 rounds only have 5 starts each, so that is another reason why those are shorter 13:57
frood66 try getting through the 10 on 519 first 13:57
frood66 who will do all 10? 13:58
BletchleyPark I do 13:58
@beta_helix right, that's why that one has such a long deadline 13:58
spmm2 me too 13:58
bobcat I've looked at all 10 13:58
TimovdL Just a startup 13:58
frood66 only 7 days 13:58
@tealight-irc i look at all, but only play with maybe 2 or 3 13:58
frood66 we have 6 days for 5 13:58
TimovdL Then a spreadsheet to choose the most promissing 13:59
@MikeCassidytoo I usually pick the one with the lowest starting score if I can find it 13:59
@beta_helix generally, would it be better to have a lot longer time (~2 weeks) on the first puzzle (with 10 servers) and then post the Rosetta server puzzle for just 1 week at the very end? 13:59
@Madde I'd need a few more computers to try all 10 predictions 13:59
@tealight-irc lol same 13:59
@marie_s you think the first step is useful for the second? 13:59
frood66 roetta being 5? 13:59
@beta_helix yeah... I have to admit I didn't think CASP ROLL would have this many targets either... the first 3 targets they gave us months to work on! 14:00
@tealight-irc yes i think it may be better beta peeps would not feel so pressured 14:00
@auntdeen2 no time for that, beta :-( 14:00
@auntdeen2 I look at all ten - pick one & go - not really evaluate 14:00
@auntdeen2 yes 14:00
Susume2 2 weeks for 10 starts soudns good, more time to try different stuff 14:00
@beta_helix yes (rosetta=5) @marie would you want them up at the same time? 14:00
@auntdeen2 beta - why 5 predictions from Zhang & quark? 14:00
spmm2 as we all probably elimate several templates quickly - maybe less time on the first ten 14:00
@auntdeen2 someone earlier asked why not just 3 from each....? 14:00
@tealight-irc maybe they are trying to overwhelm so that computers do better than hoomans.....;) 14:00
Susume2 often there is a quark-zhang pair that are very similar, could leave out one of those 14:01
@beta_helix The top ranked servers from CASP9 for the Free Modeling category was the ZhangServer & Quark (which is also from Zhang) 14:01
mottiger evil thoughts you have tealight :D 14:01
frood66 beta: I do all 10 - this takes days for me - it is not fun and I do not have time to do what I would like woth any as a result 14:01
karstenw 2 weeks? seems aweful long to me. burnout. 12 days is plenty 14:01
@tealight-irc i am evil mott..havent you heard :P 14:01
@beta_helix @tealight... I never thought of that! :-P 14:01
@CFC perhaps there's a 'tipping-point' approaching - rather than offering less options beta, we need to recognise that we maybe cant play all puzzles? 14:02
@auntdeen2 but beta - do we need all 5 from each? 14:02
@tealight-irc well it is a competiton of sorts...and those sorts of tactics wouldnt be unheard of lol' 14:02
@beta_helix how about 10-12 days for the quark/zhang servers (depending on the length of the protein and how similar each model is to one another) and then 5-7 days for the 5 Rosetta models? 14:03
karstenw yes beta 14:03
BletchleyPark 7 days rosetta please, we also have private lives. 14:03
@beta_helix @auntdeen I don't think you want me deciding which server models should be filtered out! Isn't it better for you all to decide that? 14:03
spmm2 so why not an elimination round first - to weed out templates that we just wont use 14:03
@marie_s why not rosetta first? 14:03
@auntdeen2 lol BP 14:03
karstenw yes, 7 14:03
@beta_helix @BP 7 days, got it. 14:04
frood66 is there really no way to concentrate our efforts without compromising our abilities? 14:04
@auntdeen2 beta - in a perfect world where we all have infinite computing power..... 14:04
@beta_helix @marie the fear with Rosetta first is that since they are already minimized with the same energy function used in Foldit, they already score better and so that might influence players when working on the Quark/Zhang models 14:05
@Madde couldn't an algorithm filter a prediction out that is too similar to another one? 14:05
@beta_helix @spmm do you mean a quick round where Foldit players decide which Server Predictions they want to work on? 14:06
spmm2 yes I think so 14:06
@beta_helix @Madde certainly we could do that, but if they are that similar does it really hurt for you to have those (and just ignore them?) 14:06
mottiger but how do you decide so fast? 14:06
PIECEofTOAST this beta has some pretty bad hitboxes, or is a clash tenchnically when they are too close 14:06
@beta_helix @mottiger that seems to be the issue, it's another time problem 14:07
Susume2 it is not quick in foldit to recognize the near-twins, beta, so we spend time on both while choosing 14:07
@MikeCassidytoo I think the first round even with ten should be the shortest; we are suppose to be using are intuition to decide what to play; too much time means we are trying to play them all 14:07
@Madde we could use the slot for another server's prediction 14:07
@tealight-irc i think it really comes down to player instinct and discretion...the model that gets left out can be the "winner" 14:07
@tealight-irc i agree with mike 14:07
@beta_helix @susume & Madde ok...that's not a problem at all. I can easily filter out server models that are "too" similar to one another 14:07
frood66 but the instruction IS to play them all mike 14:07
@auntdeen2 beta - that would be great 14:08
@MikeCassidytoo I LOOK at them all at most I try 2 maybe 3 14:08
spmm2 yes the extra time is to improve the score for points - templates get eliminated quickly 14:08
@tealight-irc that may be frood but i actualy dont think that foldit should leave any models out..its up to us to "feel" what needs working on 14:08
@marie_s I dont follow instruction like this, i am a player 14:08
frood66 that sounds good beta 14:09
@tealight-irc i just play what i like..i suck..but it keeps it fun lol 14:09
infjamc Would it be possible to limit each player to a smaller number of models, say 2 out of 10? Of course, scoring would be tricky because you would need a separate ranking for each model 14:09
TimovdL I would suggest only very simular weeded out 14:09
@beta_helix at keeping it fun is very important... we don't want anyone (you guys, us) burning out before CASP10 even starts! 14:09
frood66 tealight: we can omly do that if we have the time and computing power 14:09
infjamc The idea of this approach is to ensure that everyone has enough time to explore the model thoroughly while ensuring that every model is looked at 14:09
BletchleyPark @amc, i do not want that 14:09
spmm2 @frood - I do the same start routine on every template which takes several hours that eliminates quite a few 14:09
@beta_helix @TimovdL agreed, just the ones that are obviously toplogically similar 14:10
TimovdL amc neither do I 14:10
@MikeCassidytoo I eliminate by looking at them   14:10
rav3n_pl Hello folders:) 14:11
@tealight-irc hi rav 14:11
@beta_helix My other question (with respect to the Top Predictions round) was to bring back the dreaded RMSD filter, to ensure that any improvements to another player's model would end up with a different topology... 14:11
frood66 @spmm: as do I - but that is my point - large protein - how many hrs - and I'm an addict! 14:11
spmm2 well you have been playing for years mike not so easy for us noobs 14:11
@beta_helix Hi rav, we're talking about CASP ROLL puzzles 14:11
@tealight-irc yes mike i think it should be the players making the decision which to play 14:11
@MikeCassidytoo RMSD filter???? 14:11
@beta_helix yes: your score won't count unless your model is structurally different enough from the one you loaded in... 14:12
rav3n_pl maybe exploration if we are not so creative? 14:12
@MikeCassidytoo spmm2 your scores are always better than mine 14:12
@beta_helix RMSD = root mean square deviation 14:12
@auntdeen2 beta - if foldit can live woithout the top predictions round - we surely can 14:12
Datacute Can the exploration map be available for non-exploration puzzles, so we can see what is untried by others? 14:12
mottiger feel free to implement such stuff, we deal with everything 14:12
@auntdeen2 without, even 14:12
mottiger even broken wiggle functions and the proteins are top :D 14:12
@marie_s one of the problem is that recipes make micro moves with many points 14:13
@tealight-irc ruddy messy score depriver = RMSD :P 14:13
karstenw are top predictions  round helping, has this been answered already? 14:13
spmm2 but if we are using the same template - timo sp and I all independently started and convereged on the same template  no sharing at all 14:14
TimovdL Another question: can you publish what solutions are sent in with a small explanation? 14:14
spmm2 who get the points? 14:14
@beta_helix they help at determining which model everyone has converged to 14:14
frood66 soory to sound like a difficult chap - but it sounds like we need two game - one fore two cores - and one for more 14:14
@Madde beta, is there an official threshold of how different a model has to be from it's parent server prediction to get credit? 14:14
@beta_helix @Datacute the Exploration Map was turned off when we had too big an influx of players and it took up too much bandwidth... we should be able to bring it back now. 14:15
@mimi2 @ frood - and one for less as well, some people have OLD systems 14:15
Datacute Ah -- I was part of that influx 14:15
@tealight-irc yep i still paly mostly n a single core lol 14:15
@tealight-irc *play 14:15
@marie_s madde, you remmber the contrinst at the end of casp9 ; it was a nighmare 14:15
BletchleyPark @frood  a client runs on one core, the other core is used to feed the graphics card and chat 14:16
rav3n_pl Do it on dev and add data counter somewhere 14:16
@beta_helix @Madde that is a great question... to be honest, there have been so few successes in the Free Modeling Category that it rarely comes up! 14:16
@beta_helix @marie that's why I did not want to bring that back :-) 14:16
@mimi2 it does make a lot of difference  with multistarts as to how many clients you can run 14:16
@auntdeen2 major, mimi, agree 14:16
@tealight-irc at the end of the day sometimes you just cant do everything..and have to simply pick one and go with that 14:18
spmm2 especially on the bigger puzzles 14:18
@beta_helix Ok... maybe we should have another CASP ROLL chat in a couple weeks (by then we'll know how many more targets they have released, hopefully it's not 2 new targets every week!) but for now we can do away with the TOP predictions round? 14:18
rav3n_pl On 3 days rounds I use only one oil f multi starts -that one which looks best for me 14:18
@auntdeen2 yes 14:18
@tealight-irc worth a shot beta 14:19
@marie_s ok 14:19
@marie_s I never paly it anyway 14:19
BletchleyPark @beta I'd say yes 14:19
spmm2 but that method works well and could be used for other puzzles later - better than all hands 14:19
frood66 BP - noyt worth commrnting on 14:19
@Madde yes, kill the TOP predictions round 14:19
frood66 jeeze - can'y type 14:19
karstenw yes, get rid of top predictions, so we can breathe. can always add it on slow weeks, lol. 14:19
TimovdL Let it deend on the number of coming puzzles 14:20
@beta_helix and starting with R0010 (which is only 80 residues!) I'll post the first round (10 servers unless they are filtered) for 10 days? 14:20
@tealight-irc yep 14:20
krulon What about everyone worked on  one server prediction for just one day. 14:20
rav3n_pl Lol, no 14:20
@beta_helix @krulon not everyone has 1 day free! 14:20
frood66 now there is a thought 14:20
@CFC impractical krulon, and will cause riots... :) 14:20
smilingoneIRC one day isn't enough 14:21
@auntdeen2 and many of us invest much time in hand work that may not pay off until day 3 14:21
TimovdL some people have jobs 14:21
@tealight-irc krulon. if people want to do that..it should be their choice..not foisted on everyone :) 14:21
rav3n_pl 1day for 10pts? 14:21
karstenw *grabs pitchfork* ;) 14:21
spmm2 yes I like that idea hard to mange maybe two or three days for two or three templates - if we were working as a team we would be trhing to eliminate the duds and then 14:21
frood66 but  at least us poor chaps with small computing power are on a relatively even field 14:21
krulon After all were completed everyone could work on the highest scoring ones. 14:21
spmm2 pushing diversity on a selected set of templates 14:22
@beta_helix the shortest puzzles we did were the exploration rounds, we would never post anything shorter than that (hopefully!) 14:22
karstenw nope, gaming rig will kick everyone's butt in one day 14:22
@tealight-irc oh frood..i play on a single core...you keep going on how hard things are for you..but your ranking belies that, you are on a more level field than you acknowledge 14:22
frood66 and I miss sooo much 14:22
@tealight-irc thats life 14:22
spmm2 and as CFC pointed out you don't have to play every puzzle it is your choice 14:22
@beta_helix If that sounds like a plan (for CASP ROLL) we just need to discuss the Flu Puzzle real fast 14:22
@tealight-irc you can only do what you can do 14:23
infjamc See, this is exactly why I came up with the "limit each player to n templates" idea. I know that several of you don't like it, but it does make things easier for those with less computing power 14:23
@tealight-irc but people can make that choice for themselves infj 14:23
infjamc If the trade-off is between width and depth 14:23
BletchleyPark it is their choice amc. 14:23
@beta_helix @infjamc the problem is with the implementation of that... 14:23
karstenw yes, you do have to play every puzzle. this is a competitive sport :) 14:23
@tealight-irc why take away other peoples choices?.. 14:23
spmm2 agreed infjamc we don't work as a team that way unfortunately 14:23
@auntdeen2 but inf - that would be unfair to people who have teams to work with 14:23
@beta_helix @karstenw we could make the TOP prediction round worse 0 points :-P 14:23
@tealight-irc no kartsenw..yu dont have too...you choose too..big difference 14:23
@auntdeen2 I mean that don't have teams 14:24
@marie_s no karstenw, poitns and score and rank are not so important 14:24
frood66 this is a competitive game - so far 14:24
infjamc I agree that it's better to give players more choices, but my concern with that is that some models might receive disproportionally less attention 14:24
TimovdL It is like giving 80% of the people no chance at all it they are unlucky 14:24
infjamc ...as things now stand 14:24
spmm2 what is important is all of us beating the CASP competition 14:24
frood66 remove the competition then much of what is saud is fair 14:24
westrel3 remember  KISS 14:24
@tealight-irc well that comes down to teh people playing infj 14:24
@beta_helix @infjamc that is the hope with at least giving you 1 day/model (10 days for the first puzzle) 14:24
infjamc For example, models that are closer to the native but also need to dig out of a deep local minimum 14:24
infjamc Actually, here's a 2nd idea... 14:25
spmm2 anyway beta probably needs to go to a meeting :) flu 14:25
infjamc Why not split a puzzle into several "sub-puzzles"? 14:25
infjamc Say, instead of having one puzzle with 10 templates, have two with 5 each 14:25
@beta_helix yes, and I did want to ask: is it ok to post a Flu Puzzle with all this CASP ROLL craziness? 14:25
frood66 why not remove teh scoreboards? 14:25
spmm2 ok by me beta 14:25
@tealight-irc personally i dont think options should be limited...people make their own choices...longer time sframes and better staggering are of more use 14:26
@auntdeen2 yes, beta - if long running 14:26
@tealight-irc fine with me beta 14:26
@beta_helix so there will be 3 CASP ROLL puzzles up, 1 beginner puzzle (which none of you should be wasting your time on ;-)) and the Flu Puzzle 14:26
@tealight-irc if you can make it run long enough 14:26
spmm2 yes infjamc - ten at once is hard to get my head around 14:26
rav3n_pl Beta: make it small as possible... 14:26
@marie_s frood , you can competet with pleasurein one puzzle, but not want to do all puzzles, many players play like this 14:26
smilingoneIRC Beta...some want choice... post it for those that want to play with the flu... personally, I won't be touching it since my plate is full with casp 14:26
frood66 I have no issue with that marie - that is not what I said 14:27
@beta_helix @infjamc so 1 Quark puzzle (5 models) and 1 Zhang puzzle (5 models) removing any models that are too similar and then the final Rosetta puzzle (5 models) 14:27
@marie_s give us the flu 14:27
spmm2 yes beta 14:27
frood66 ok - so max 15 models plus 1 14:27
karstenw many days for flu and i'll squeeze it in when my othe puzzles stall 14:27
rav3n_pl <- 3 more days off the grid 14:27
lamoille agreed kars 14:28
@beta_helix @marie it's very rare to have someone ask "give us the flu" ;-) 14:28
spmm2 design puzzles can stall pretty quickly as well 14:28
@auntdeen2 beta - how many days each then? 14:28
@tealight-irc ahh but beta this one is one we cant catch :P 14:29
@auntdeen2 if you split to one puzzle for quark & one for zhang?  did I understand that correctly? 14:29
@beta_helix that is the issue (we have 3 weeks, but have to wait 4 days for the server models to come back... and then we need at least 1 day to process/submit everything, that really leaves a bit over 2 weeks) 14:29
TimovdL We want to kill the flu 14:29
spmm2 not too long for early ones - last few days is just getting points on the pose usually 14:29
karstenw don't split them.  can't compare all ten at once. it will actually take longer if you split them 14:30
@auntdeen2 but krs - then 5 models per puzzle 14:30
@beta_helix why don't I examine R0010 and R0011 and see how similar the Quark/Zhang models are... maybe we can lower that 10 models down to 5-7 already! 14:30
@auntdeen2 that could be much easier 14:30
@tealight-irc worth a shot 14:30
frood66 yes beta 14:30
@auntdeen2 good beta 14:30
infjamc karstenw: Part of the idea of splitting the 10 models into two groups of two is to find the best quark model and the best zhang model... 14:31
@beta_helix luckily they are both nice and short: 80 & 81 residues :-) 14:31
karstenw yes beta  first see what you can filter out 14:31
@thomirc8 would an exploration puzzle make sense for CASP ROLL? 14:31
@mimi2 yes - better to remove similar and keep models together for comparison 14:31
TimovdL That will speed up things 14:31
spmm2 agreed infjamc better to split them 14:31
tokens About the flu puzzle: Some people might find it more fun to work on a design puzzle, so I think it's a good idea to have that along with the casp puzzles 14:31
TimovdL 80  is tiny 14:31
@tealight-irc nothing wrong with tiny.. 14:32
spmm2 how many cystines :) 14:32
@beta_helix @tokens there should be a Flu puzzle up later today... speaking of which, I should go get that ready! 14:32
tokens :) 14:32
@auntdeen2 thanks, beta :-) 14:32
mottiger since there is no deadline for flu puzzles for any competetion why not just release it and make it a long runner ? 14:32
frood66 80 is gorgeous 14:32
BletchleyPark regarding splitting: you will end up with one puzzle of fairly useless results and another with useful results, yet both will take time 14:32
@beta_helix @spmm hahahaha, yeah, we'll have to see! 14:32
karstenw infjamc, you are right, but you are expanding the project by doing so. time is not on our side 14:32
infjamc 80. Good, that means 8 times the processing speed relative to 160 14:32
TimovdL tx for your time 14:32
lamoille thx beta 14:32
infjamc karstenw: I don't see splitting the puzzle as a problem if the due dates are the same. But again, that's just me 14:33
@beta_helix @infljamc yeah, so hopefully we don't need to split them up... but we'll look into that idea again in a couple weeks. 14:33
rav3n_pl Ahh looks like I miss nice small puzzles... 14:33
spmm2 exactly and there will be better exploration of the quark models 14:33
@CFC the reduction of options... from FoldCentrals point of view, with CASP10 approaching, is that a good idea? Or will encourage convergence? 14:33
@beta_helix hopefully! 14:33
karstenw you running the puzzle twice. once as zhang and once as quark 14:33
BletchleyPark and often only one of them is best 14:34
@beta_helix we are only reducing the final round which was just leading to convergence already 14:34
karstenw exactly 14:34
BletchleyPark so one waisted puzzle time 14:34
@thomirc8 also, will the diversity tools make it into the client anytime soon? 14:34
@beta_helix that is what we are trying to get rid of: wasted puzzle time 14:34
infjamc BP: That's true, but part of me wonders if that's due to the score function 14:34
@beta_helix ok gang, I really have to go. 14:34
spmm2 @BP the quark models are in anyway - do you just not use them? 14:34
BletchleyPark @amc what difference does splpitting then make ? 14:35
karstenw lol, bye beta 14:35
BletchleyPark bye beta 14:35
frood66 thx for yr time beta :) 14:35
@beta_helix Thanks for this great chat... hopefully we can set another one up in a couple weeks! 14:35
spmm2 thanks beta 14:35
@tealight-irc cya beta! 14:35
@tealight-irc ..now go get the flu :D 14:35
rav3n_pl Days... 14:35
@beta_helix I'll try to post up the CASP ROLL puzzle schedule in the Forum soon 14:35
rav3n_pl ;) 14:35
BletchleyPark @spmm I use all of them. 14:35
infjamc BP: Fewer models per puzzle = those with lesser computing power can focus on one of the sub-puzzles 14:35
@MikeCassidytoo bye beta 14:35
spmm2 hi rav having fun :P 14:35
karstenw bletch, you can't compare starts if you split them 14:36
@beta_helix bye everyone, thanks again for all your hard work! 14:36
rav3n_pl I'm only on chat via my phone... 14:36
BletchleyPark @karsten that is why I am not in favor of slitting 14:36
BletchleyPark splitting 14:36
f1pokerspeed Dores anyone have some suggestions for essential scripts that I should have? 14:36
infjamc karstenw: I'm not familiar with the technical details, but I think that problem is fixable 14:36
infjamc After all, there are already puzzles that allow you to load solutions from a previous round 14:36
@marie_s blue fuse, quakeR, DRW, loop rebuid 14:37
karstenw oh, you meant what's the difference the other way, lol 14:37
infjamc So it should theoretically be possible to create two concurrent puzzles that are interconnected 14:37
@marie_s local quake 14:37
infjamc Namely, you can load the solutions of version A in version B, and vice versa 14:37

Generated by irclog2html.py 2.10.0 by Marius Gedminas
- find it at mg.pov.lt!

User login
Download links:
  Windows    OSX    Linux  
Windows
(7/8/10)
OSX
(10.12 or later)
Linux
(64-bit)

Are you new to Foldit? Click here.

Are you a student? Click here.

Are you an educator? Click here.
Social Media


Search
Only search fold.it
Other Games: Mozak
Recommend Foldit
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, Boehringer Ingelheim, RosettaCommons