Mon Jun 5 11:51:21 PDT 2006 T0315 Make started Mon Jun 5 11:52:41 PDT 2006 Running on lopez.cse.ucsc.edu Mon Jun 5 12:41:57 PDT 2006 Kevin Karplus Strong full-length BLAST hit 1j6oA 39%id over 252 residues E-value 7.7e-50 Also strong full-length hits to 1yixA 1zzmA 1xwyA No other strong blast hits. Mon Jun 5 13:38:41 PDT 2006 Kevin Karplus The HMMs are getting strong hits for 1yixA, 1j6oA, 1xwyA, 1zzmA. The only one of these in SCOP is 1j6oA (c.1.9.12). There are weaker hits for a related family (c.1.9.3). Mon Jun 5 20:55:30 PDT 2006 Kevin Karplus The best hits are 1j6oA, 1yixA, 1xwyA, 1zzmA (subsequent hits are much weaker), but the top 25 are all TIM barrels, from the same superfamily. The try1 run seems to have settled on 1xwyA as the template, though it initially favored 1j6oA. Looking at the final try1-opt2 result though, it looks like 1j6oA is really the main template. There is a bad break between L169 and N170 and an even worse one between E92 and M93. Other breaks are minor. It might take a bit of polishing to produce a decent model without breaks, but this looks otherwise fairly straightforward. Wed Jun 21 11:13:18 PDT 2006 Kevin Karplus try1-opt2 scores better with the unconstrained costfcn than try2-opt2, but try2-opt2 scores better with try2 costfcn. Our automatic server SAM_T06_server_TS1 beats both on unconstrained.costfcn. Best other servers are FOLDpro_TS2-scwrl FOLDpro_TS1-scwrl Zhang-Server_TS4-scwrl ROBETTA_TS4 BayesHH_TS1-scwrl ROBETTA_TS3 3Dpro_TS1-scwrl Pmodeller6_TS4 ROBETTA_TS1 PROTINFO_TS3-scwrl ROBETTA_TS2 PROTINFO_TS3 RAPTORESS_TS4-scwrl try2-opt2 has pretty much closed the breaks---the worst one left is T0315.try2-opt2.pdb.gz breaks before (T0315)V165 with cost 0.684487 Wed Jun 21 11:24:43 PDT 2006 Kevin Karplus Both models and the top several server models are all in agreement about the structure and alignment. I'll do one more polishing run and call this done. Wed Jun 21 13:20:25 PDT 2006 Kevin Karplus try3-opt2 has improved the score slightly, but the movements are very small. I think that we have gone past the point where the cost functions are of any use. Rosetta likes repacking try2-opt2 better than try3-opt2. I'll submit try3-opt2 try2-opt2.repack-nonPC (Rosetta's favorite) try1-opt2 align 1 (from 1j6oA) align 2 (from 1yixA) Wed Jun 21 13:26:19 PDT 2006 Kevin Karplus So submitted. Thu Mar 22 13:16:06 PDT 2007 Kevin Karplus According ot our evaluations, our best submitted model was model3=try1-opt2, which did not fare well against the servers. The best servers were getting GDT aound 95%, while we only had around 87%. Was it that undertaker chose the wrong template? VAST lists the top templates as 1zzmA, 1xwyA, 1j6oA, 1yixA, 1bf6B, ... We had 1j6oA, 1yixA, 1xwyA, and 1zzmA as our top hits by quite a bit. So we either messed up the alignment or we damaged it in building the loops. The alignments in the undertaker-align.under script were no better than models we submitted, so it is likely that we did not damage them (much)---so either they were bad or we were just not as successful at loop building as others. The alignments look ok compared to the VAST structural alignment, so it looks like the problem is in the loop building.