Fri Jun 20 09:02:20 PDT 2008 T0464 Make started Fri Jun 20 09:03:15 PDT 2008 Running on cheep.cse.ucsc.edu Fri Jun 20 09:38:05 PDT 2008 Kevin Karplus T0464 is almost an ORFan, with only 13 sequences found in NR by target06, and those all quite similar. There is not enough variation to see strong conservation, but there is a bit more conservation on the hydrophobics than the polars, and P12, G36, and P70 are somewhat more conserved than the rest of the residues. The t2k alignment actually finds more homologs, most of which are missing a chunk from about E20 to about L50. This might indicate that E20 and L50 are close and that there is an insertion of about 30 residues. The conservation patterns on the ends look pretty good, so this does look like a real hit, not just random alignment. Fri Jun 20 09:47:33 PDT 2008 Kevin Karplus The w0.5 HMMs get some strong hits: ID Length Target_Evalue 2jnyA 67 3.19e-06 2pk7A 69 3.02e-05 2hf1A 68 0.000347 2js4A 70 0.000818 2jr6A 68 0.01188 2jnyA and 2pk7A are not in the template library, so I might have to add them as extra templates. Only 2hf1A is in the template library. Fri Jun 20 10:43:57 PDT 2008 Kevin Karplus The alignments in undertaker-align are very short and not very useful. I'll start a try2 run that pays less attention to alignment constraints, but uses the 5 good hits (not just 2hf1A). Fri Jun 20 11:00:21 PDT 2008 Kevin Karplus It looks like E22 and E52 are coming out paired, as are L20 and L54, so my initial conjecture about there being an insertion into the fold looks reasonable. The structure of the insertion is not clear, though try1 is proposing a helix and a hairpin, while try2 is proposing a pair of helices. Fri Jun 20 11:42:37 PDT 2008 Kevin Karplus For try3, I'll make a costfcn that has sheet and helix constraints for the parts that try1 and try2 agree on, but has breaks and clashes turned up a bit to get better closure. Try3 will still run from alignments to sample loop possibilities for the insertion. Since this is such a short protein, I'll probably do at least 5 runs from alignments, to get a bit more sampling of the inserted loop. try4 will use the same costfcn as try3, but will omit the 2hf1A alignments. Fri Jun 20 12:38:46 PDT 2008 Kevin Karplus try3-opt3 scores very well. try4-opt3 poorly. Despite this, try3-opt3 seems to have an uncloseable gap between L50 and H51. For try5, I'll do the same sort of optimization as for try3, but with a higher break penalty in the costfcn. Fri Jun 20 13:00:47 PDT 2008 Kevin Karplus For try6, I'll try putting L26 and L29 near I55, to get some burial for the dry face of the helix. I also simplified the sheet and helix constraints, to include just the consensus of the previous models. I think that some of the models may have extended the helix in the loop too close to the end of the strand, so that it was not flexible enough to move around. Fri Jun 20 14:16:47 PDT 2008 Kevin Karplus try6 does not manage to satisfy the pack_helix constraints---it doesn't even do as well as try3. Fri Jun 20 14:54:31 PDT 2008 Kevin Karplus For try7, I'll do a polishing run to reduce clashes and breaks. I suspect that it will just end up polishing try5-opt3.gromacs0, which has 0 breaks. I may want to do another run that excludes try5 and try7 from the optimization, to find a "next-best" model. Tue Jun 24 10:10:22 PDT 2008 Kevin Karplus The MQAC quality assessment favors the Zhang-Server, SAM-T06-server, BAKER-ROBETTA, and Pcons_dot_net. The MQAU quality assessment favors SAM-T08-server, Zhang-Server, SAM-T06-server, pro-sp3-TASSER, and BAKER-ROBETTA. I'll do meta-server runs with both the try1 and the try7 costfcns. Tue Jun 24 16:03:12 PDT 2008 Kevin Karplus MQAU1 < SAM-T06-server_TS1 MQAC1 < SAM-T06-server_TS1 MQAU7 < SAM-T06-server_TS1 MQAC7 < SAM-T06-server_TS1 Not much variety from the metaservers--they all liked the same model. The order with the try7 costfcn is ReadConformPDB T0464.MQAU7-opt3.pdb ReadConformPDB T0464.MQAC7-opt3.pdb ReadConformPDB T0464.MQAU1-opt3.gromacs0.repack-nonPC.pdb ReadConformPDB T0464.try7-opt3.pdb ReadConformPDB T0464.MQAC1-opt3.gromacs0.pdb ReadConformPDB T0464.try5-opt3.gromacs0.pdb ReadConformPDB T0464.try1-opt3.gromacs0.pdb ReadConformPDB T0464.try3-opt3.pdb ReadConformPDB T0464.try2-opt3.gromacs0.repack-nonPC.pdb ReadConformPDB T0464.try6-opt3.pdb ReadConformPDB T0464.try4-opt3.gromacs0.pdb Looking for DIFFERENT models, we get MQAU7-opt3 (MQAC7, MQAU1, MQAC1) try7-opt3 (try5 < align) try1-opt3 < align try3-opt3 < align try2-opt3 < align try6-opt3 < align try4-opt3 < align I'm not sure what to do next: a polishing run on the MQA* and try7 models? try to generate more distinct models? polish the "next best" models: try1 and try3? a metaserver run excluding SAM-T06-server? Tue Jun 24 16:24:54 PDT 2008 Kevin Karplus I started a metaserver run (MQAUC7) with the union of the top 10 models from MQAU and MQAC, but excluding the SAM server models. After removing duplicates, this left 10 models to start from. Tue Jun 24 17:46:50 PDT 2008 Kevin Karplus MQAUC7 is based mainly on Pcons_multi_TS1 It does not score as well as MQAU7, MQAC7, or MQAU1, but it beats (barely) try7. I like the C-terminal helix of MQAUC7-opt3, but I think I like the A34-H51 loop better in try7. Actually, I'm undecided about the C-terminus: the MQAU7 strand is also appealing. Wed Jun 25 09:05:09 PDT 2008 Kevin Karplus I accidentally redid the MQA computations, because the CASP people had re-released the server tarball. The metaserver run was not redone, since everything was judged "up to date" (no dependency on the server files, I guess). There *were* some changes in the MQAC scores, but not one that changed which 10 models were on top. Sat Jun 28 15:42:06 PDT 2008 SAM-T08-MQAO hand QA T0464 Submitted Sat Jun 28 15:42:06 PDT 2008 SAM-T08-MQAU hand QA T0464 Submitted Sat Jun 28 15:42:06 PDT 2008 SAM-T08-MQAC hand QA T0464 Submitted Fri Jul 4 10:29:17 PDT 2008 Kevin Karplus I should do MQAX1 and MQAX7 runs excluding the SAM servers, to see if there is anything else out there that I like, though I already have 7 rather distinct models. Given that this is an NMR model, I would not be at all surprised to find that the parts I can't seem to get a handle on are also parts that are floppy in the real world. For my final choosing, I should probably put together a costfcn that includes sheets and helices from all the distinctly different models, and see what comes to the top. Fri Jul 4 13:04:50 PDT 2008 Kevin Karplus Something wierd: MQAX7-opt3 scores poorly with try1.costfcn, but MQAX7-opt3.repack-nonPC scores well. A lot of the scoring differences are for functions that should not have been affected by only changing sidechains! I'll run the make decoys/score-all.try1.pretty command again, to see if it is a repeatable error. Yes, it is. Fri Jul 4 13:14:04 PDT 2008 Kevin Karplus I think that either the gromacs or the rosetta repacking runs stepped on each other, so that files are not correctly attributed. I'll remake the gromacs and rosetta runs. Fri Jul 4 13:18:51 PDT 2008 Kevin Karplus OK, now the MQAX7 values agree more. try7.costfcn likes MQAX7-opt3 best, but try1.costfcn does not. try1.costfcn likes MQAX1-opt3 best, which comes from the sane Pcons_multi_TS1 model that MAQX7 and MQAUC7 comes from. I think I like the way MQAU7 looks better than MQAX7, so I'll try making a new costfcn (try8) that favors it. Fri Jul 4 13:33:50 PDT 2008 Kevin Karplus try8 started on peep. Fri Jul 4 17:43:53 PDT 2008 Kevin Karplus try8-opt3, based on MQAU7-opt3, scores well with the try8 costfcn though MQAX7-opt3 still scores better with the try7 costfcn, thanks mainly to differences in the helix constraints. Fri Jul 4 17:53:32 PDT 2008 Kevin Karplus OK, I don't think I'll be making any great improvements now. I'll take 5 decent but different models and submit them. ReadConformPDB T0464.try8-opt3.pdb # < MQAU7-opt3 < SAM-T06-server_TS1 ReadConformPDB T0464.MQAX7-opt3.pdb # < Pcons_multi_TS1 ReadConformPDB T0464.try7-opt3.pdb # < try3-opt3 < align(2jwyA) ReadConformPDB T0464.try1-opt3.gromacs0.pdb # < align(1sgpI) ReadConformPDB T0464.try2-opt3.gromacs0.repack-nonPC.pdb # < align(2hf1A)