Tue May 23 09:04:41 PDT 2006 T0297 Make started Tue May 23 09:05:59 PDT 2006 Running on lopez.cse.ucsc.edu Tue May 23 09:07:53 PDT 2006 Kevin Karplus Moderately strong full-length hits with blast to 1fxwF, 1es9A, 1wab, 1bwp, 1bwr, 1bwq. (25-26% id oer 185-201 residues). Tue May 23 09:34:21 PDT 2006 Kevin Karplus The t06 alignment has many sequences (2257), so there should be plenty of signal for secondary structure and mutual information computations. There are 15 pdb files in the t06 multiple alignment: 1fxwA=1wab, 1es9A, 1bwr, 1bwq, 1bwp, 1vyhA, 1vzgA, 1u8uA, 1j00A, 1yzfA, 1z8hA, 1vjgA, 1pp4A=1k7cA=1dexA, 1ese=1esd=1esc, 1zmbA Tue May 23 11:34:38 PDT 2006 Kevin Karplus As expected, top hits with target models include 1wab, 1es9A, aand other c.23.10 superfamily members. I'm still waiting for the template-library search to finish---it seems to have gotten quite slow lately, perhaps because of the number of files that need to be opened. Tue May 23 11:53:20 PDT 2006 Kevin Karplus The t06 template library, though still rather incomplete, does score 1es9A at the top. Tue May 23 14:27:05 PDT 2006 Kevin Karplus After all fold-recognition scoring is done, the top hits are all c.23.10 members, with 1es9A, 1ivnA, 1wab, 1yzfA, ... scoring on top. All the models from alignment are pretty much in agreement about the beta sheet, with only small variations in how the helices pack against it. We should be able to pick out a decent conformation from the available set. It looks like try1 is favoring 1fxwF as the template, even though that scored rather low in the "best-scores" list. It was blast's favorite, though, so this should not come as a surprise. Tue May 23 14:48:39 PDT 2006 Kevin Karplus Looking at the models from the alignments in T0297.undertaker-align.pdb (which don't include 1fxwF as a template), I see that burial and residue-residue contacts are very well predicted, not that we need these aids for this target. Tue May 23 17:52:25 PDT 2006 Kevin Karplus The try1 run failed, probably because I changed undertaker when it was running (without renaming it to "undertaker-in-use" or something first). I have restarted the make. Tue May 23 20:07:23 PDT 2006 Kevin Karplus On the second try1 run, undertaker seems to be favoring 1es9A. Tue May 23 20:53:59 PDT 2006 Kevin Karplus The try1-opt2 result is not too bad, but there are a couple of bad breaks, and the N-terminal long helix seems to be rotated a bit, as the stripe that is predicted to be buried is on the *outside*. The RR contacts are quite good in this model, but we don't really need them. For try2, I'll turn them off and just use the sheet and helix constraints trom try1-opt2 (except for the first helix, which I'm not sure I believe in). The conserved polar residues D39, S40, N93, D94, D190, H193 cluster, and probably should have distances fixed from some of the templates. I won't bother for try2 (still seeing if I can close the gaps), but I should do it for try3. Wed May 24 07:01:51 PDT 2006 Kevin Karplus try2-opt2 manages to reduce (but not eliminate) breaks and clashes, but the packing terms are not quite as good. The first helix is still there and still rotated so that the more hydrophobic face is exposed. The bad breaks are before P100 and E49. The rest are not so terrible. try1-opt2 had at least 5 breaks that were as bad as these two try2-opt2 (including before P100 and E49, so crossover won't fix it). Make started Tue May 23 17:50:11 PDT 2006 Running on lopez.cse.ucsc.edu Thu May 25 15:42:35 PDT 2006 Kevin Karplus Score the server models, look for things worth copying. Do polishing run starting from existing models. (Server models not available yet.) Sat May 27 19:10:43 PDT 2006 Kevin Karplus I picked up the server models and scored them with the try1 costfcn, and SAM_T06_servr_TS1 scored better than any other model. Robetta_TS4 was almost as good as the SAM models. The robetta models get about the same constraint scores as the SAM ones. If polishing from server models works, we might want to try a polishing run with all the server models (so that crossover has a wider pool to draw on). Sat Jun 10 04:06:50 PDT 2006 Kevin Karplus At yesterday's meeting, the try2-opt2 model was discussed. It looks good, except for the unsupported helix at the N-terminus. In the top-scoring template (1es9A), the N-terminus wraps around the protein, covering currently exposed hydrophobics L69 and A110. We could add constraints for V3-I110 and L6-L69 which might stretch the N-terminus out in the same way. We also probably want to make E137-A163 a single, contiguous helix, as the corresponding part is in 1es9A. I have put such constraints into try3.costfcn and am running try3 from the initial alignments on lopez. Sat Jun 10 07:05:39 PDT 2006 Kevin Karplus try3-opt2 does bend the helix over as requested, but doesn't score quite as well with the unconstrained costfcn. For try4 I'll weaken the constraints on the N-terminus, and increase clash and break costs, then polish existing models. I'm going to give up on merging E137-A163 into one helix---undertaker seems quite insistent on making it be 2. I tweaked the weights of the try4 costfcn until the try3-opt2 models scored just a little better than the try2-opt2 models, so that polishing would concentrate more on the try3-opt2 models, but allow crossover from the try2 models. Run started as try4 on shaw. Sat Jun 10 09:07:49 PDT 2006 Kevin Karplus try4-opt2 looks pretty good, though Rosetta prefers repacking try2 (probably because the exposed helix is easy to repack). The unconstrained cost function also slightly prefers try2, probably because the n_ca_c and sidechain costs are a bit wors for try4-opt2. I should probably polish try4 using a cost function that has no constraints, but high clash and break weights (make sure that try4-opt2 scores best still). Starting such a run as try5 on shaw. Sat Jun 10 11:03:34 PDT 2006 Kevin Karplus Foo! try5-opt2 went back to straightening out the helix. I'll have to put the try4 constraints back in. try6 (with just the N-terminal constraints) started on shaw. Sat Jun 10 13:40:41 PDT 2006 Kevin Karplus try6-opt2 looks pretty good. It scores best with try6.costfcn, and is almost as good as try5-opt2 with the unconstrained costfcn. Rosetta like repacking try6-opt2 best. Sat Jun 10 13:52:41 PDT 2006 Kevin Karplus I'll submit try6-opt2 try5-opt2 align1 from undertaker-align (1es9A) align2 from undertaker-align (1ivnA) align3 from undertaker-align (1yzfA) Sun Sep 24 12:02:13 PDT 2006 Kevin Karplus Our model2 (try5-opt2) looks great (better than any of the servers). try5-opt1 is even slightly better. Our model1 is somewhat worse (a few servers beat it). GDT measure alone gives a somewhat different picture, with some of the servers getting better GDT (though worse Hbonds).