Tue May 23 09:04:44 PDT 2006 T0298 Make started Tue May 23 09:06:15 PDT 2006 Running on lopez.cse.ucsc.edu Tue May 23 09:09:50 PDT 2006 Kevin Karplus No strong hits with blast of PDB. Closest is 1ys4A with evalue of 0.038, but this is only a little piece (49% id over 43 residues). At 336 residues long, T0298 may be a 2-domain protein. Tue May 23 11:37:54 PDT 2006 Kevin Karplus The t06 multiple alignment finds 60 pdb templates, so this is likely to be an easy fold recognition target. We get strong hits to the two-domain c.2.1.3,d.81.1.1 proteins (like 1gd1O, 1dssG, 1gadO, ...). Somewhat surprisingly, these did *not* come up in the simple BLAST search, so we'd better check to make sure that they aren't contaminmation. The t2k multiple alignment has only 2 PDB sequences in it: 1ys4A and 2cvoA, but the top-scoring sequence in t2k-w0.5-scores is 1mb4A, which has the same c.2.1.3,d.81.1.1 as all the top hits with any of the HMMs, and with very good E-values (4.2e-30). Since the target seems pretty clearly to be an Aspartate-Semialdehyde Dehydrogenase like many of the templates, I'm confused about why blast didn't see it, getting an E-value of only 0.038 for the best hit (to 1ys4A, which is an aspartate-semialdehyde dehydrogenase). Tue May 23 14:41:04 PDT 2006 Kevin Karplus The t06 and t04 template-library searches also agree on the same set of templates. Tue May 23 14:52:20 PDT 2006 Kevin Karplus The top templates in T0298.best-scores are all the same pair of domains, at least for about the first 50-60 hits. Tue May 23 20:08:38 PDT 2006 Kevin Karplus On try1, undertaker seems to be favoring alignment to 1t4bA, which scored best in the T0298.best-scores file, though it did not come up at all the BLAST search. Tue May 23 21:35:48 PDT 2006 Kevin Karplus Aside from the region from L163 to Q200, the try1-opt2 model looks pretty good. Some of the rr prediction looks ok, other parts look misleading, so I think that they should be left out for try2. Actually, the region from L163-Q200 may not be so bad---most of it is copied from 1t4bA. We may want to look for clusters of conserved polar residues (like S161, D314, N315) and add constraints for their polar atoms based on templates. Wed May 24 07:10:01 PDT 2006 Kevin Karplus try2-opt2 has greatly reduced breaks, though still a few bad ones (before T281, V104, P123 and V274). The break at V104 may be fixable by changing the sheet constraint for to align P107-M109 parallel to L125-S127 Hbond L125 (backing the edge strand up 2). This would make the loop with V104 have more room and would match the secondary structure predictions better. It would also improve the rr predictions. A couple of other breaks my be reduced by adding a SheetConstraint for E272-V274 parallel to E291-Y293 (hbond T292) backing the strand up by 1. Thu May 25 15:44:48 PDT 2006 Kevin Karplus Do try3 with the SheetConstraints mentioned above. Sat Jun 10 04:52:26 PDT 2006 Kevin Karplus try3 did not move the backbone significantly from try2. There's a little difference around S39-E51, but not a major change. The alignments were were trying to get with SheetConstraint P107 M109 L125 S127 hbond L125 2 SheetConstraint E272 V274 E291 Y293 hbond T292 2 were basically unchanged: SheetConstraint (T0298)L126 (T0298)S128 (T0298)P106 (T0298)V108 hbond (T0298)S127 1 SheetConstraint (T0298)I271 (T0298)G275 (T0298)E291 (T0298)G295 hbond (T0298)W273 1 Sat Jun 10 05:43:54 PDT 2006 Kevin Karplus I did a VAST alignment of try2-opt2 with the medium-redundancy set of PDB files, and got the best hit to 1brmC (=1brmA). There are gaps next to the strands that I think are mis-aligned, so it should be fairly easy to move the alignments to better fit what I think the sheets should be. Sat Jun 10 06:49:47 PDT 2006 Kevin Karplus I edited the 1brmA/T0298-1brmA-vast.a2m alignment to make 1brmA/T0298-1brmA-hand.a2m, fixing up the strands I wanted moved, and also fixing up a few other areas by looking for discrepancies between the sequence-based alignments and the structure-based one, and looking to see which alignment seemed better to me. Sat Jun 10 06:59:02 PDT 2006 Kevin Karplus I'm starting try4 on lopez with a cleaned-up cost function, but using only the new hand alignment to start. try5 will be the same costfcn as try4, but with all alignments to start (also running on lopez). Sat Jun 10 07:39:40 PDT 2006 Kevin Karplus try4 quickly adds other alignments---the initial one is not very successful. try5 does not seem to be using the hand alignment either. I may want to add something to undertaker to *force* a particular InsertAlignment operation, so that I can do a TryAllAlign like in try5, then force in the alignment I want. Sat Jun 10 10:50:11 PDT 2006 Kevin Karplus try4 and try5 did not even do as well as try1, try2, and try3 with the try5 costfcn. Since I don't feel like trying to move things manually, think I will just polish the existing models for the preliminary submission. We can look at this target again at a target meeting and see if we want to try tweaking the edge strands. Started as try6 on lopez. Sun Jun 11 11:50:15 PDT 2006 Kevin Karplus try6 is the best scoring with the try6 costfcn (which has no constraints), and rosetta likes it best for repacking. It is based on try2, which is quite similar. Next best is try1, then try3. The try6 model is pretty clearly based on the first alignment (to 1t4bA), but has closed most of the gaps pretty cleanly. The remaining bad gaps are T0298.try6-opt2.pdb.gz breaks before (T0298)V104 with cost 2.29406 T0298.try6-opt2.pdb.gz breaks before (T0298)P123 with cost 2.03517 I'll make one more polishing attempt, with breaks and clashes turned way up. (try7 on shaw) Sun Jun 11 19:55:41 PDT 2006 Kevin Karplus try7-opt2 has pretty much closed the breaks. The worst reamaining on is T0298.try7-opt2.pdb.gz breaks before (T0298)P123 with cost 0.678411 Rosetta still likes repacking try6-opt2 better. I'll submit try7-opt2 try1-opt2 try3-opt2 align1 (1t4bA) align2 (1ys4A) We could revisit the issue of exactly how the strands align, but I'm not sure it is worth the effort. Sun Jun 18 10:04:34 PDT 2006 Kevin Karplus I scored the servers and our models with the unconstrained costfcn a couple of days ago, and try7-opt2 scored best, with our server as the top server model. The top servers other than ours are RAPTOR_TS1-scwrl, HHpred2_TS1-scwrl, HHpred3_TS1-scwrl, SPARKS2_TS1-scwrl, SP4_TS1-scwrl, RAPTOR-ACE_TS1-scwrl, SP3_TS1-scwrl, ... Somewhat surprisingly, the ROBETTA server models are in less agreement with our cost functions than usual. I'll do a polishing run with several of the top server models, to see if there is anything we should pick up from them (like a different strand alignment, for example). The cost function, try8.costfcn, is a little less demanding on breaks and clashes than try7.costfcn, but otherwise similar. It ranks the servers HHpred2_TS1-scwrl HHpred3_TS1-scwrl SPARKS2_TS1-scwrl RAPTOR_TS1-scwrl SP4_TS1-scwrl BayesHH_TS1-scwrl SAM_T06_server_TS1 PROTINFO_TS1 Ma-OPUS-server_TS1-scwrl RAPTOR-ACE_TS1-scwrl SP3_TS1-scwrl Bilab-ENABLE_TS3 RAPTOR-ACE_TS1 SP3_TS1 ROKKY_TS2-scwrl SPARKS2_TS1 PROTINFO_TS1-scwrl SAM_T06_server_TS1-scwrl Zhang-Server_TS3-scwrl ... Sun Jun 18 10:32:30 PDT 2006 Kevin Karplus Optimization of server models started as try8 on lopez. Sun Jun 18 11:08:31 PDT 2006 Kevin Karplus Although HHpred2_TS1-scwrl starts out best in try8, it repeatedly gets overtaken by SPARKS2_TS1, probably as a result of gap closure in the SAPRKS2 model. Sometimes they trade places being first, and someimes HHpred2_TS1-scwrl beats out SPARKS2_TS1. Sun Jun 18 12:26:07 PDT 2006 Kevin Karplus try8-opt2 does not score quite as well with try8.costfcn as try7-opt2 (or even try6-opt2). The breaks are a bit smaller though. With the try7 costfcn, try8 is second only to try7. There is a difference between try7 and try8 on the alignment of one strand: I271-G275. try7-opt2 has SheetConstraint I271 G275 E291 G295 hbond W273 1 try8-opt2 has SheetConstraint I271 V274 T292 G295 hbond E272 1 The burial is clearly better in try8. I'll do another polishing run, with the try7 costfcn, but with the sheet constraints from try8. I think we may need more sampling of different alignments, because we *should* have been able to pick out this strand alignment. I tweaked try9.costfcn until try7-opt2 and try8-opt2 scored almost identically. I'll set crossover high, so that the best parts of each model can be kept. Sun Jun 18 14:24:21 PDT 2006 Kevin Karplus In try9, both try7 and try8 have turns being the best, but the try9-opt1 model is based on try7-opt2. Sun Jun 18 14:31:06 PDT 2006 Kevin Karplus T0298 has a helix sticking out and clearly needs to be optimized as a dimer. I'll make a dimer based on 1tb4[AB]. I've set up dimers from try7-opt2 and try8-opt2, and when try9 finishes, I'll set up one from try9-opt2, then do an optimization with the dimers. Things have moved so little from the templates, that even the un-optimized dimers look pretty good. Sun Jun 18 16:20:31 PDT 2006 Kevin Karplus try9-opt2 (based mainly on try7-opt2) is now our best-scoring monomer. Rosetta likes repacking it best also. I'll make a dimer from it and start a dimer optimization. (dimer/try1 started on lopez) Sun Jun 18 21:21:14 PDT 2006 Kevin Karplus I'll submit as monomers ReadConformPDB T0298.try9-opt2.pdb # best scoring ReadConformPDB from-dimer-try1-opt2-A.pdb # optimized in dimer context ReadConformPDB T0298.try8-opt2.pdb # optimized from servers ReadConformPDB T0298.try1-opt2.pdb # fully automatic ReadConformPDB T0298.try6-opt2.pdb # earlier, less refined version of model 1 Sun Jun 18 22:06:29 PDT 2006 Kevin Karplus So submitted, and dimers submitted: dimer try1-opt2 dimer-try7-1tb4A dimer-try9-1tb4A dimer-try8-1tb4A Sun Jun 18 22:15:52 PDT 2006 Kevin Karplus Dimer submissions were rejected: PFRMAT TS TARGET T0298 OLIGOMER # ERROR! There is no target name: T0298_oligomer # TARGET Txxxx # was expected