Tue Jul 23 10:31:24 PDT 2002 t0161 23 July 2002 Kevin Karplus The T2K script found NO homologs for target T0161! This is going to make the 2ry predictions and the scoring in general very weak. The best scores are indeed very weak, but I get multiple hits on quite different c.55 domains, so there is a chance that we can get this rather difficult fold recognition. None of the alignments look particularly great though, so we'll probably have to treat this as mainly ab initio. There seems to be no consensus of CAFASP servers either. 23 July 2002 Kevin Karplus The try1-opt-scwrl model has found a bit of beta sheet, though I'm not really sure I believe it. The rest of the model is still crap. Tue Aug 20 16:34:54 PDT 2002 Modernized score function. robetta1 through robetta4 all score much better than try1-opt. Added try-alignments target to Makefile and the resulting alignments to undertaker.script. Rerunning undertaker with the larger library of alignments and with OptSubtree and JiggleSubtree turned on. 21 Aug 2002 Kevin Karplus try2-opt-scwrl now scores best, beating the robetta models, but it has just made the whole thing a 4-helix bundle, ignoring the secondary-structure prediction. Actually, L17-D74 look fairly reasonable, but the rest is rather dubious. Let's turn pred_alpha2 up and try again, though I don't have that much confidence in predicted alpha values when there is only one sequen ce in the alignment fed to the neural net. We can also skip the TryaAllAlign, and turn up InsertAlignment slightly, to make this more of an ab initio run. Thu Aug 22 16:42:08 PDT 2002 kevin Karplus try2-opt-scwrl slightly beats try3-opt-scwrl. Try3-opt-scwrl is ALSO an all-helical bundle. All the CAFASP 2ry predictors agree on strands around 125 and 149, and half put a strand around 138, half put a strand around 114, and half out a strand around 18. I'm not yet willing to accept an all-helical protein. Let's try adding parallel strand constraints for Y146-W154, F123-L126, Q102-N107---maybe with the folowing register: 102> QVLEL 123> FYVLL 146> YVRVSFEY With these constraints (and sufficient weight on them), robetta1 scores best, followed by robetta3. Let's reoptimize starting from nothing. Fri Aug 23 10:47:36 PDT 2002 Best score with try4 constraints is try4-opt-scwrl. The helices are beginning to bundle, and the parallel sheet is almost forming, but the whole structure is very loose. It looks like Y146 might pair best with L119 or P120. Let's try again with a new set of constraints (try5.constraints) 102> QVLEL 146> YVRVSFEY 119> LPAPFYVLL With these constraints, T0161.try3+T0161-1khhA-2track-...+T0161-1lpuA-vit-adpstyle1.pw.a2m.gz:1lpuA.19.40.pdb scores best, though it has no strands. Let's add straightness constraints also for Q102-L106, V147-Y153, F123-L127. With these added, the best score is still for T0161.try3+T0161-1khhA-2track-...+T0161-1lpuA-vit-adpstyle1.pw.a2m.gz:1lpuA.19.40.pdb Let's do another run (from random conformation) with the try5 constraints. Fri Aug 23 15:52:32 PDT 2002 Try5-opt-scwrl scores best with the try5 constraints, but is not great. I'll turn this over to Jenny now. Perhaps it would be worthwhile to try a run starting from try1-opt, try2-opt, ..., try5-opt (or opt-scwrl, since the difference is small). This could be done with or without constraints, though I fear that without some constraints undertaker will favor the helix bundles too much. Scoring without constraints now DOES favor try2 and try3 over try4 and try5. Wed Aug 28 Rachel Karchin Let's try above suggestion, we start with try1-opt ... try5-opt and use constraints. I'll bring down constraint minimums for parallel strands to 0, so they can come as close as they like without penalty.i Aug 28 Jenny Draper 7:30pm It appears that try6 was accidentally run without constraints. The no-constraint try6 turned the protein back into a 4-helix bundle again (surprise,surprise), and no try6 models score decently with the constraints. Aug 29 Jenny Draper 12:10am Running try7, using all the same settings as the intended try6; will check in the morning, when (hopefully) my brain is back. Any input or ideas are more than welcome, since I'm just starting on this one! 29 Aug 2002 09:33 Kevin Karplus It looks like the try7-opt-scwrl and try7-opt are the best scores now. (try7-try5.40.30 is just another copy of try7-opt, they score differently because of the differences in rounding for the 0.001 Angstrom positions, which may casues slight differences when patching up breaks when reading in PDB files, I think). The sheet seems to be trying to form, but not quite succeeding. 29 Aug 2002 10am Jenny Draper I'm going to edit the constraints to be hbonds instead of CB, then do a run with higher-weighted constraints. Any other ideas? Does the three-strand sheet found here seem reasonable? 29 Aug 2002 11am Jenny Draper I can't seem to get undertaker to run. Sending out an email for help... 29 Aug 2002 11am Jenny Draper I can't seem to get undertaker to run using condor. Sending out an email for help; running undertaker on squeek & roar instead... 29 Aug 2002 12:30pm Jenny Draper Hmmm. I meant for try9 to be a try8, only starting from try7 (try8 started from try5); but instead it seems to have started from a bunch of alignments. I'm not sure why this happened? try8-opt-scwrl has put the sheet together; however, I'm not happy with the long helices sticking out into space, or with the exposed very hydrophobic face of helix A68-G82. try9-opt-scwrl has done something radically different, but still managed to pull a bit of sheet together. It's very loose, and has lots of small helices. I'm not sure what to think of it? Date: Thu, 29 Aug 2002 12:36:02 -0700 (PDT) From: Jenny Draper Subject: t0161 Hi everyone, I'd like some eyes on t0161 for me, to help me find a direction; I'm not sure where I want to head with this target. Right now I have about 4 different models I could try to optimize: try5-opt-swcrl, try7-opt-scwrl, try8-opt-scwrl, try9-opt-scwrl try9 was a run that tried a bunch of alignments; I'm not sure how to tell which alignment try9-opt was using? please, whoever has a moment, help me find a direction to go in for the next undertaker run or two (a few different directions would be OK). -Jenny Date: Thu, 29 Aug 2002 12:59:06 -0700 From: Kevin Karplus To: learithe@cats.ucsc.edu CC: rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu, karplus@soe.ucsc.edu Subject: Re: t0161 According to undertaker-try9.log, the try9 run was using try8.constraints, but though try7-opt was read in, it was replaced by a random conformation in the ConformFromSeq command. Something seems to have been wrong in the try9 run---none of the specific Fragments seem to have been read in. The try8 run had no trouble though, so I wonder if this was related to the wierdness of the cluster this morning. ------------------------------------------------------------ 13:04 Kevin Karplus Since try9 did not run the way it was expected to (it started from random conformation and dind almost entirely generic insertions), I'm running a try10 optimization, with the try8.constraints still, starting from try7-opt, try8-opt and try9-opt. try5-opt is very loose, but the helix packing for S28-K40 and L45-H65 looks pretty good. try7-opt has similar helix packing, but is more compact. try8-opt seems to be struggling a bit to get a parallel connection between L104-G105 and F151-Y153. try9-opt is very loose and has put together the strands requested in try8.constraints only by chopping things up. I think that we need to do a scoring run without constraints. Without constraints, try7-try5.13.30 scores best. It looks quite good, aside from packing helix P42-H65 with its hydrophobic face out and its charged face in robetta4.pdb scores next best, but it has would the predicted strands into helices. Let's try a no-constraint optimization from try7-try5.13.30, try7-try5.40.30, try7-opt-scwrl, robetta3 (yes, 3 not 4---robetta 3 has one of the strands straight): try11. Date: Thu, 29 Aug 2002 13:44:53 -0700 From: Kevin Karplus To: karplus@soe.ucsc.edu, rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, learithe@cats.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu Subject: t0161 I've looked at the models Jenny suggested, and made comments in the README file. In brief, I don't think that the constraints are helping much, so I started another run without constraints (try11). Right now the best-looking model I've seen is try7-try5-13.30, and it does seem to be trying to form a strand-helix-strand conformation. If the unconstrained optimization starting from there doesn't blow it up, we may want to do an optimization with constraints guessed from the best run. Maybe parallel for 4 or 5 starting at F123 with V149? papFYVLLp pyvrVSFEYwh We could play with the final strand alignment papFYVLLp pyVRVSFeywh papFYVLLp pyvRVSFEywh papFYVLLp pyvrVSFEYwh Perhaps it would be worth doing 3 sets of CB constraints for these three choices, and optimizing the best few models with each set of constraints. Jenny, do you want to try that? ------------------------------------------------------------ 29 Aug 2002 14:37 Kevin Karplus try10-opt (which used the try8.constraints), produces a rather pretty x-shaped conformation. It is not very convincing, but it could be submitted as an alternative model. Scoring without constraints currently favors try11.2.100, which looks pretty good except for the lack of connections for strand P122-L127. As soon as try11 finishes, I'll do another run with some strand constraints---probably papFYVLLp pyvrVSFEYwh In try11-opt, the available Hbonds are on V147, V149, F151, Y153 Date: Thu, 29 Aug 2002 14:47:50 -0700 (PDT) From: Jenny Draper To: Kevin Karplus Subject: Re: t0161 i noticed you're planing on starting a try12. I wanted to try the three alternative alignments you have suggested; I can call them tries 13-15. Which model or two do you suggest i start from? -Jenny -------------------------------------------------- Date: Thu, 29 Aug 2002 15:04:30 -0700 From: Kevin Karplus To: learithe@cats.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: (message from Jenny Draper on Thu, 29 Aug 2002 14:47:50 -0700 (PDT)) Subject: Re: t0161 OK, I've started try12 (from try11-opt), you can start tries 13-whatever. Take a look at try11-opt, as it looks quite promising. Date: Thu, 29 Aug 2002 15:01:22 -0700 (PDT) From: Jenny Draper To: Kevin Karplus Subject: Re: t0161 It looks like you're just going on this. I wrote CB constraints for > papFYVLLp > pyVRVSFeywh (try13.constraints) and > papFYVLLp > pyvRVSFEywh (try14.constraints) let me know if you want me to do anything here. -Jenny ------------------------------------------------------------ 12 Aug 2002 3:30pm Jenny Draper try-11 opt does look very promising! Very pretty. From karplus@bray.cse.ucsc.edu Thu Aug 29 15:42:36 2002 Date: Thu, 29 Aug 2002 15:41:36 -0700 From: Kevin Karplus To: karplus@soe.ucsc.edu, rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, learithe@cats.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu Subject: t0162 Now that t0161 is making some progress (we have something that could be submitted, in a pinch), I looked a bit at t0162. Date: Thu, 29 Aug 2002 15:42:13 -0700 (PDT) From: Jenny Draper To: Kevin Karplus Subject: t0161: try12 It looks to me like try12 is going off into the wierd direction try9 did. Am I wrong, or should it be re-started? -Jenny Date: Thu, 29 Aug 2002 16:23:31 -0700 From: Kevin Karplus To: learithe@cats.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: (message from Jenny Draper on Thu, 29 Aug 2002 15:42:13 -0700 (PDT)) Subject: Re: t0161: try12 try12 was doing what was requested of it. I had set the constraint weight much too high. I turned the constraint weight back down. Would you please rerun it (as try15 I guess), as soon as you have started tries 13 and 14. I thought you were running them already, and was just coming to look at them! WE only have a few hours left, and I DON't want to stay up to midnight if I can help it. Date: Thu, 29 Aug 2002 16:25:43 -0700 From: Kevin Karplus To: learithe@cats.ucsc.edu Subject: Re: t0161: try12 You might want to start from try11.try7-13.0 and try11-try7-13.1 which seem to be doing pretty well. ------------------------------------------------------------ From karplus@bray.cse.ucsc.edu Thu Aug 29 18:42:28 2002 Date: Thu, 29 Aug 2002 18:41:29 -0700 From: Kevin Karplus To: karplus@soe.ucsc.edu, rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, learithe@cats.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: t0161 We need to make decisions by 8:30 tonight about what the current best models are for t0161. Jenny, I'd like you to make a list by 8pm of the most likely models. Anyone who wants to comment on them can send me e-mail. I'll try to make decisions by 8:30. (Jenny, if you happen to continue working, and get a better model, you know how to replace a submission.) ------------------------------------------------------------ Date: Thu, 29 Aug 2002 20:08:30 -0700 (PDT) From: Jenny Draper To: Kevin Karplus Cc: rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu Subject: Re: t0161 Sorry for the late email; I've been having all sorts of email trouble; this message was bounced twice, from both learn & soe! My list of most likely models: T0161.try14-opt-scwrl.pdb (which is a LOT like try 13 & 15 & try7-try5.13.30) Current best score w/try12-15 constraints Third best score w/o constraints T0161.try11-opt Which I like more than try14 Best score w/o constraints Undertaker seems to get stuck in the try7/13/14/15 conformation; most of our try-opts look like they're headed in this direction; try11 seems to be the most different. The current score-decoys.rdb file is w/o constraints; to see one w/constraints, look at score-decoys-try12.rdb I think I'll be going to sleep early tonight, not stay up working... I hope that's OK. I'm really tired! -Jenny --------------------------------------------------------------- From learithe@cats.ucsc.edu Thu Aug 29 20:15:22 2002 MIME-Version: 1.0 Date: Thu, 29 Aug 2002 20:02:12 -0700 (PDT) From: Jenny Draper X-Sender: learithe@learn.ic.ucsc.edu Reply-To: Jenny Draper To: Kevin Karplus cc: rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu Subject: Re: t0161 In-Reply-To: <200208300141.g7U1fT407844@bray.cse.ucsc.edu> Content-Type: TEXT/PLAIN; charset=US-ASCII X-UCSC-CATS-MailScanner: Found to be clean X-SpamBouncer: 1.4 (10/07/01) X-SBPass: NoBounce X-SBClass: OK My list of most likely models: T0161.try14-opt-scwrl.pdb (which is a LOT like try 13 & 15 & try7-try5.13.30) Current best score w/try12-15 constraints Third best score w/o constraints T0161.try11-opt Which I like more than try14 Best score w/o constraints Undertaker seems to get stuck in the try7/13/14/15 conformation; most of our try-opts look like they're headed in this direction; try11 seems to be the most different. ------------------------------------------------------------ 29 Aug 2002 20:45 Kevin Karplus I agree with Jenny. model1 try11-opt model2 try14-opt-scwrl