Wed Jun 7 09:28:26 PDT 2006 T0322 Make started Wed Jun 7 09:29:20 PDT 2006 Running on lopez.cse.ucsc.edu Wed Jun 7 16:34:16 PDT 2006 Kevin Karplus The T06 and t04 alignments don't show much conservation, but the t2k one shows strong conservation. All agree on FY94, but there are 8 key residues in T2K: NGDQS55 HQF61 G62 G63 D70 TSA87 FY94 LIV104 The secondary structure predictions are fairly similar. The top 10 alignments are in excellent agreement, with only minor movements of the helices, and they are almost full length, so I think this will be a fairly easy comparative modeling problem, despite the lack blast hits other than 2cy9A. Wed Jun 7 21:02:48 PDT 2006 Kevin Karplus The try1-opt2 model looks pretty good, but this is clearly a dimer. I should build a dimer based on 2fs2[AB] and optimize that. Mon Jun 26 22:44:58 PDT 2006 Kevin Karplus 2fs2A/T0322-2fs2A.dimer-a2m created make-dimer.under written and run through undertaker to create dimer/decoys/dimer-try1-2fs2A.pdb (ROBETTA_TS5 is the best-scoring server with our unconstrained costfcn, after SAM_T06_server_TS1) Mon Jun 26 22:50:41 PDT 2006 Kevin Karplus In the initial dimer, it looks like there ought to be anti-parallel hbonds from R89-D93 of chain A to D93-R89 of chain B, with L90 and C92 hbonding. I'll add that as a constraint for the first optimization of the dimer. Mon Jun 26 23:04:42 PDT 2006 Kevin Karplus dimer/try1 started on camano. Tue Jun 27 00:00:07 PDT 2006 Kevin Karplus I forgot to tell dimer/try1 that it should optimize as a dimer! I'll start try2 with that addition. Tue Jun 27 00:02:18 PDT 2006 Kevin Karplus started dimer/try2 on cheep. Tue Jun 27 09:53:10 PDT 2006 Kevin Karplus The dimer/try2 run is much more successful than the dimer/try1 run, but there is still a break that needs to be closed. I'll do a polishing run without constraints, but with several of the packing terms turned up (as well as breaks and soft_clashes). Tue Jun 27 09:57:36 PDT 2006 Kevin Karplus dimer/try3 started on cheep Tue Jun 27 12:27:16 PDT 2006 Kevin Karplus dimer/ try3 makes small improvements on dimer/try2 Tue Jun 27 12:53:46 PDT 2006 Kevin Karplus Submitted InFilePrefix dimer/decoys/ ReadConformPDB T0322.try3-opt2.unpack.pdb chain A ReadConformPDB T0322.try2-opt2.unpack.pdb chain A ReadConformPDB T0322.try1-opt2.unpack.pdb chain A InFilePrefix decoys/ ReadConformPDB T0322.try1-opt2.pdb InFilePrefix ReadConformPDB T0322.undertaker-align.pdb model 1 # from 2fs2A I also submitted the dimers, before I realized that they would not be accepting dimer submissions for this target. Thu Aug 10 11:04:20 PDT 2006 Kevin Karplus This was chosen as a refinement target. I should do the refinment of TR322 in a dimer context. Note: The CASP model lacks residues 1-11, 154-157, so I may want to optimize just the subdomain. The tr322 model has terrible hbonds and terrible sidechains, but phobic_fit is good. Overall, it scores like a terrible model. Cleaning it up may be difficult without just throwing it out and starting over. Fri Aug 11 03:06:14 PDT 2006 Kevin Karplus The N-terminal end of the sheet is a little more like the template in tr322 than in our models, but I think that the N-terminal helix may be misplaced. We'll have to check in the dimerized models. Running gromacs and rosetta repacking on tr322 make it score much better, though still a long way from try1-opt2. tr322's C-terminus may interfere with dimerization. Fri Aug 11 04:39:41 PDT 2006 Kevin Karplus Made a dimer from tr322.gromacs0.repack-nonPC using make-refine-dimer.under called dimer/decoys/dimer-tr322-2fs2A.pdb Fri Aug 11 04:51:44 PDT 2006 Kevin Karplus dimer/mult4 started on cheep. Note: it does output a Template.atoms file, which will need to be gzipped, since cleanup has removed the old Template.atoms.gz files. Fri Aug 11 07:47:06 PDT 2006 Kevin Karplus dimer/try4-opt2 does not look bad, but I should probably add a sheet constraint between the monomers SheetConstraint V88 F94 F251 V245 hbond L90 10 to get the sheet to continue between the monomers. Fri Aug 11 07:53:54 PDT 2006 Kevin Karplus dimer /mult5 started on cheep. Fri Aug 11 11:01:20 PDT 2006 Kevin Karplus dimer/try5-opt2.unpack.gromacs0.repack-nonPC is the first dimer that Rosetta gives a negative energy to. The try5 costfcn moves try5 a little ahead of try4, but still doesn't like it much. W22, W85,and L57 are a bit foamy in try5, and there are some bad breaks still. We should probably polish again with breaks and packing terms turned up. dimer/mult6 started on cheep. Sat Aug 12 04:46:19 PDT 2006 Kevin Karplus dimer/try6 does better than try5, but is still not up to dimer/try1 much less try2 or try3. Rosetta likes try6-opt2.unpack.gromacs0.repack-nonPC best of any it has repacked (including our own). The inter-monomer hbonds seem to be forming fine. Sat Aug 12 04:56:05 PDT 2006 Kevin Karplus I started dimer/mult7 to do a polishing run. We should probably look to see if there are any differences with our own models, to see if there are further changes we want to make. Sat Aug 12 09:22:00 PDT 2006 Kevin Karplus On the dimer/try7 costfcn, try7-opt2 finally beats out our dimer/try1-opt2, and rosetta really likes try7-opt2.unpack.gromacs0.repack-nonPC (much more than try6). I think we may have reached the point of diminishing returns on this model, being as likely to make things worse as better. I don't see any features from our own models important enough to try to transfer. Sun Aug 13 17:34:34 PDT 2006 Kevin Karplus I'll do one more polishing run, just to ensure that we have indeed converged. dimer/mult8 started on cheep Sun Aug 13 20:06:49 PDT 2006 Kevin Karplus dimer/try8-opt2.gromacs0.repack-nonPC is new best for rosetta. try8-opt2 has passed try1-opt2 (though not quite caught up with try2 and try3) for the try8 costfcn. Breaks are still a bit big, so I'll turn up break cost for another polishing run. dimer/mult9 started on cheep, but I'm not sure it is such a good idea---the dimer seems to have a hole, exposing L64 and F94 to the solvent. Perhaps I should back up and repeat the optimizations from an earlier stage with much higher dry weights. Mon Aug 14 11:20:21 PDT 2006 Kevin Karplus dimer/try9-opt2.unpack.gromacs0.repack-nonPC is rosetta's new favorite (by a lot), and try9-opt2 slightly beats try8-opt2 on the try9 costfcn (though still not close to try2 and try3). The structure has very low clashes, but is a bit foamy, as evidenced in all the dry costs. I'm not so sure that the big hole areound L64 and F94 is such a problem, as that is also where many of the conserved residues (including conserved charged residues) cluster, so it may be the binding pocket. The hydrophobics may not be exposed when bound to a ligand. Mon Aug 14 11:46:37 PDT 2006 Kevin Karplus I started dimer/mult10 to try to pack try4 through try7 tighter (starting from the gromacs0 and gromacs0.repack-nonPC models, to allow more room for score improvement). I turned the dry weights up quite a bit and the clashes down. I'll have to repolish with clashes and breaks turned back up, if I like the resulting model. Mon Aug 14 14:11:48 PDT 2006 Kevin Karplus dimer/try10 did get denser packing, though the breaks are not too good. The try10 model was built mainly from try7-opt2.unpack.gromacs0 model. The try8 and try9 costfcns prefer the try7-opt2 model to try10-opt2, so if we polish try10, we'll have to be careful not to include the earlier try7 models. Mon Aug 14 14:20:42 PDT 2006 Kevin Karplus dimer/mult11 started on cheep to polish try10-opt2. Mon Aug 14 18:44:19 PDT 2006 Kevin Karplus dimer/try11 gets pretty good packing, but clashes need to be fixed still. I'll try polishing it with clashes and breaks turned up a bit. Mon Aug 14 18:52:14 PDT 2006 Kevin Karplus dimer/mult12 started on cheep to polish try11-opt2.unpack.gromacs0 and try11-opt2.unpack.gromacs0.repack-nonPC Tue Aug 15 07:59:58 PDT 2006 Kevin Karplus dimer/try12-opt2 does not score as well as try11-opt2 with the try12 costfcn. rosetta likes try12-opt2.unpack.gromacs0.repack-nonPC a bit better than the try11-opt2 version it was polished from, but still greatly prefers try9-opt2.unpack.gromacs0.repack-nonPC In dimer/, depending which costfcn we pick, generally either try9-opt2 or try11-opt2 comes out best. (The try3 costfcn manages to move try12 ahead of try11 and try9.) Hmm, it looks like I did not have "multimer 2" in the OptConform commands for most of the multimer optimizations I've done on tr322. For dimer/try13 I'll be sure to turn it on. Tue Aug 15 14:34:45 PDT 2006 Kevin Karplus dimer/try13 beats out all but try3 with the try13 costfcn, and try13-opt2.unpack.gromacs0.repack-nonPC beats out all but try9 with Rosetta's energy function. Tue Aug 15 14:52:07 PDT 2006 Kevin Karplus Trying to get a new best rosetta score, by polishing try13-opt2.unpack.gromacs0* models with a high-break-cost cost fcn. dimer/mult14 running on cheep. Tue Aug 15 17:48:34 PDT 2006 Kevin Karplus dimer/try14-opt2 scores better than all but try3 with try14.costfcn, but try13 still beats it at try3.costfcn and rosetta still prefers try13. The biggest differences between try13 and try14 seem to be sidechain, soft_clashes, breaks, and hbond_geom_beta_pair. Tue Aug 15 17:55:50 PDT 2006 Kevin Karplus I think I've reached the point of diminishing returns on this refinement target, so I'll submit InFilePrefix dimer/decoys/ ReadConformPDB T0322.try14-opt2.unpack.pdb chain A # fewest breaks ReadConformPDB T0322.try9-opt2.unpack.gromacs0.repack-nonPC.pdb chain A # rosetta favorite ReadConformPDB T0322.try13-opt2.unpack.pdb chain A # tightest packing ReadConformPDB T0322.try13-opt2.unpack.gromacs0.repack-nonPC.pdb chain A # rosetta 2nd ReadConformPDB T0322.try8-opt2.unpack.pdb chain A Refinement models submitted with comment: The tr322 model was initially optimized in three ways gromacs repacking sidechains (except PRO and CYS) by rosetta gromacs + repacking sidechains (except PRO and CYS) by rosetta The four tr322 models were evaluated with undertaker and tr322.gromacs0.repack-nonPC.pdb was chosen for further optimization. The model was superimposed on 2fs2A and 2fs2B to make a dimer and all subsequent optimizations were done in the dimer context. Refinement consisted of rounds of optimization with undertaker (with slightly different cost functions each time) optimization with gromacs repacking of sidechains (except PRO and CYS) by rosetta Refinement model 1 is try14-opt2, the best-scoring model with most of the undertaker cost functions used. It has the smallest chain breaks. Refinement model 2 is try9-opt2.unpack.gromacs0.repack-nonPC, the model that rosetta likes best of all the ones it repacked. Refinement model 2 is try13-opt2, the model that scores best with some undertaker cost functions, and second best with most. It has the densest local packing. Refinement model 4 is try13-opt2.unpack.gromacs0.repack-nonPC, the model that rosetta likes second best of all the ones it repacked. Refinement model 5 is try8-opt2, an earlier version of model 2. Although many of the undertaker runs started with multiple initial model and allowed crossover, the following history traces the main contributions to the models. try14-opt2 < try13-opt2.unpack.gromacs0 try13-opt2 < try12-opt2 < try11-opt2.unpack.gromacs0 try11-opt2 < try10-opt2 < try7-opt2.unpack.gromacs0 try9-opt2 < try8-opt2.unpack.gromacs0 try8-opt2 < try7-opt2.unpack.gromacs0.repack-nonPC try7-opt2 < try6-opt2.unpack.gromacs0 try6-opt2 < try5-opt2 < dimer-tr322-2fs2A ------------------------------------------------------------