Tue Jul 16 10:27:28 PDT 2002 t0156 16 July 2002 Kevin Karplus Not an obvious recognition target, possibly ab initio. try1-opt gets a few hairpins, but has let other beta strands peel away and roll up. Adding all the constraints needed to hold this together could be difficult. One fairly obvious connection to add is for the two strands predicted to be parallel: G62-I66 and W90-V95, since they have a single helix between them and G62-I66 is already starting to form a parallel sheet. The exposed edge of the sheet has O G62 O L64 N I66 O I66 which is NOT the normal pattern for parallel strands. The pairing of the two strands is also not obvious---perhaps SAG GVLVIDGAG STGWTGLIVHGAV 62 GVLVI 90 WTGLIV 24 July 2002 Jenny Draper Attempting to add above parallel stand pairing constraints. 24 July 2002 Kevin Karplus Looking at model 50 of T0156.t2k.undertaker-align.pdb suggests also G111-G138 A114-T140 and L136-V139 I134-F141 (a hairpin) 25 July 2002 Jenny Draper Added parallel strand pairing of W90-V95 and G62-I66, and a C-beta constraint for I112 F141 for try2. This didn't work out too well. I'm trying explicit H-bond guesses now, to put the two C-terminal hairpins into a sheet, and strands 1,2,3 into another sheet... 26 July 2002 Kevin Karplus Ran try5 for one iteration and noticed that strand 1 probably should be parallel to 3, not anti-parallel to 2. Jenny will run try6 with a new scoring function for these constraints. From learithe@cse.ucsc.edu Sun Aug 4 00:07:38 2002 MIME-Version: 1.0 X-Authentication-Warning: sundance.cse.ucsc.edu: learithe owned process doing -bs Date: Sun, 4 Aug 2002 00:07:37 -0700 (PDT) From: Jenny Draper To: Kevin Karplus cc: learithe@soe.ucsc.edu Subject: Re: T0156 In-Reply-To: <200208040424.g744OUF13688@bray.cse.ucsc.edu> Content-Type: TEXT/PLAIN; charset=US-ASCII X-SpamBouncer: 1.4 (10/07/01) X-SBPass: NoBounce X-SBClass: OK I did put together a few constraints for try6, and ran it. It looks terrible. A few of the earlier structures were looking better (satisfied the new constraints), but try6-scwrl-opt (the best scoring) blew the strands apart again. I haven't looked at it since, I've been working on the alignment test scripts. I can try a few more runs next week though. -Jenny On Sat, 3 Aug 2002, Kevin Karplus wrote: > Are you working on T0156? I haven't seen any changes since 27 > July---I thought you were planning to put together some constraints > for try6. > > The due date is 19 Aug, so it isn't SUPER urgent, but it does look like > an ab-initio prediction, which takes much longer. > 5 August 2002 Kevin Karplus I see 8 predicted strands: # length residues 1 11 Q34-C44 antiparallel 2 4 V63-I66 parallel 3 4 G92-V95 parallel 4 4 I112-L115 parallel 5 4 E133-L136 antiparallel 6 3 V139-F141 antiparallel 7 4 I146-S149 antiparallel 8 3 I153-V155 antiparallel I don't know about the first strand, but it looks like there are two other sheets. The specific fragments seem to be giving us the end of strand 1 parallel to strand 2---possibly forming a sheet 1^ 2^ 3^ 4^ (though, of course, the order of the strands could be almost any permutation). I moved all the constraints that had been put in the score function to "try6.constraints" and re-scored the decoys with no constraints. try6-opt-scwrl still scores the best, probably because it has worked hard at packing stuff in. Let's try another optimization run with no initial conformation and no constraints. It is not likely to do anything good, but I'd like a baseline with the current protocol for optimization. 5 August 2002 Kevin Karplus try7-opt-scwrl does score best with no constraints, but the beta sheets are not put together. We do get Hbonds I134-F141, L136-V139, but no other useful Hbonds (unless the 2ry predictions are badly wrong). T140 probably hbonds with the next strand which hairpins with the final strand. I suspect that the first three strands are 1^2^3^, but other permutations of the parallel sheet are possible. I'm too tired tonight to work out the most likely hbonds or CB pairs and put them in as constraints. 6 August 2002 Jenny Draper All H-bonds and pairings mentioned above are in try6.constraints... Including Hbonds I134-F141, L136-V139, parallel sheet 1^2^3^, and T140-Y148 to bring strand 7 close to strand 6... but try6-opt didn't have them... perhaps the contraints should be weighted more heavily? 10 Aug 2002 Kevin Karplus Putting try6.constraints back in, rasing the constraint weight, and rescoring makes try6-opt-scwrl score best. Let's do another run, starting from scratch, with no initial alignments but with alignment insertion, using the try6.constraints. 11 Aug 2002 Kevin Karplus T0156.try8-opt-scwrl is new best score, and looks somewhat better than previous ones, though still not as compact as I'd like. Let's try upping the constraint weight slightly and reoptimizing from here. I'll also add some constraints to try to improve the final hairpin and to attach the short strand at I112-L115 to the antiparallel sheet. Doing so brings some other decoys to the top-- T0156.try8+T0156-1pjbA-2track-...+T0156-1ja9A-2track-protein-STR-global-adpstyle1.pw.a2m.gz:1ja9A.18.60.pdb 11 Aug 2002 Kevin Karplus Looking at try9.2.60.pdb leads me to suspect that the RIGHT thing to do is to make I112-L115 PARALLEL to P38-V42 and I134-L136. We could add CB constraints for I39-I112-V132 through V42-L115-T135). Replacing the antiparallel constraints by these parallel constraints makes try9.1.60, try9.2.60, and try9.3.60 score best. Let's take several of the try9 intermediates as starting points (try9 hasn't finished yet) and start try10 with the new constraints. 12 aug 2002 Kevin Karplus Reran make with new template library, which eliminated 1kpjI as possible target. It looks like try10 stomped on undertaker-try9.log. Unfortunately, try10-opt-scwrl, the new lowest cost, has wound the strand I112-L115 into a helix. Trying again with same score function as try10, but starting from scratch. If that fails, we can try adding constraints to straighten the curled up strand. 12 aug 2002 Kevin Karplus Run try11-opt scores poorly. Both try10 and try11 mess up the strand I154-V156, putting it through another strand. Perhaps the constraints are messed up? 13 August 2002 Kevin Karplus None of the robetta models score well. Adding constraints to hold the I154-V156 strand more firmly in place and increasing the penalties for clashes and breaks, still leaves try10-opt-scwrl as the best scorer. Let's reoptimize with the new score function, starting from try10-opt-scwrl. 15 August 2002 Kevin Karplus Best score is now T0156.try12-opt-scwrl. I don't like the helix for strand I112-L115, but the rest is starting to look OK. I want parallel hbonds from I39 and T41 16 Aug 2002 Kevin Karplus try13-opt now scores best, but we still have that awful strand turned into a helix at I112-I115 and G92-V95 has slid up to collide with the strand at L64-I66. Perhaps we have some constraints set wrong? Breaking the chain at V99 and G111 may help OptSubtree move things around a bit. We can move G92-V99 2 Ang (G92 away from L64), and G111-R120 5 Ang away from T41. I created move-residues.perl as a special-purpose script to move these segments in try13-opt, creating decoys/T0156-moved.pdb, which I'm using as the starting point for try14 optimization run. 17 Aug 2002 Kevin Karplus New top score is try14-opt. It begins to get the single-sheet idea. Perhaps we need a run with break turned down and constraint turned up to get things into better positions, then turn break up to close the gaps. 17 Aug 2002 Kevin Karplus New top score is try15-opt, which still hasn't succeeded in flattening the strand. Maybe I should copy 110-117 from try7-opt, where it was straight, and let OptSubtree try to put it back into the right place. I tried two runs try16 and try17, but both seemed not to be applying OptSubTree soon enough, so I killed them both. 18 Aug 2002 Kevin Karplus try18-opt almost gets as good as try15-op, but it STILL has the strand coiled into a helix!! Perhasp teh problem is my insistence on the strand being parallel. Let's try making it antiparallel, and doing another run from the cut-and-paste start. Actually, I'm doing two runs, both using the new "no_initial_fill" option for OptConform. try19 is trying to get an antiparallel middle strand, and try20 is trying to get a parallel middle strand. We'll have to score both ways and see which one we want to tweak more. We may end up submitting two models, one with each direction for the strand. 11:42 try19 has finished. It has gotten the strand into roughly the right place, but has a lot of work to do fixing up gaps. I'll start another run from it, but with more generations and less weight on OptSubtree. With the anti-parallel score function, try19-opt is the best. 13:14 try20 has finished. With the PARALLEL score function, try20-opt-scwrl is best. It looks, to my mind a bit worse than try19-opt, but it should probably be given an opportunity to try to get fixed. Starting with try18, we're alternating runs with define-score.script set up for parallel and anti-parallel. The odd-numbered runs are anti-parallel, the even numbered ones are parallel. 14:56 try21-opt does not look much better than try19-opt, nor does it score much better. Maybe I should move L115-V132 away and E133-V157, and let OptSubtree try to repack them? We're running out of time on this target, so may need to give up soon. 16:54 try22-opt-scwrl is the new best scorer in the parallel-constrained set, but it still looks awful in the region of the strand. Let's rescore everything with NO constraints, and see what looks best. With all constraints off, the two best scores are for try14.17.80 and try21-opt. The try14.17.80 one has the strand curled into a helix, so I would rate it lower than try21 and try22. Currently, my rating for models to submit is try21-opt, try22-opt-scwrl, try14.17.80 Perhaps I should just submit these and stop beating my head against the wall on this target? With only a day to go, I don't see that I'll get any sudden breakthroughs.