Wed Jul 2 09:19:52 PDT 2008 T0482 Make started Wed Jul 2 09:20:20 PDT 2008 Running on cheep.cse.ucsc.edu Wed Jul 2 10:15:00 PDT 2008 Kevin Karplus The top pdb-blast hit (2btjA, E-value 1.5) is not likely to be useful, as it is for a beta barrel that is 222 residues long. Since T0482 is only 120 long, it can't be the full barrel. The t06 alignment only finds homologs in other Pseudomonas species (and one Azotobacter), and they are very similar, so conservation signals are weak. The t04 and t2k alignments are quite similar. The HMMs are getting very poor E-values, so this looks like it will be a template-free prediction. Wed Jul 2 11:17:08 PDT 2008 Kevin Karplus The top three HMM hits are not in SCOP id length E-value 2glwA 93 2.2693e+01 1zowA 313 3.6072e+01 2ebdA 310 3.7815e+01 Several of the top hits below that are for c.95.1.2, the thiolase-like fold and superfamily, the Chalcone synthase-like family. The 2glwA is another Pyrococcus protein of unknown function. 1zowA and 2ebdA are very similar to each other and to the c.95.1.2 proteins, according to VAST, which puts them in the cd00830: KAS_III domain Ketoacyl-acyl carrier protein synthase III (KASIII) initiates the elongation in type II fatty acid synthase systems. It is found in bacteria and plants. Elongation of fatty acids in the type II systems occurs by Claisen condensation of malonyl-acyl carrier protein (ACP) with acyl-ACP. KASIII initiates this process by specifically using acetyl-CoA over acyl-CoA. The KASIII proteins seem to be mostly 2-domain proteins, with two thiolase folds. So we have two main contenders: 2glwA (which is another Pyrococcus protein) and the thiolase fold. The predicted secondary structure seems to be 3-4 antiparallel strands followed by 3 helices, which is not very compatible with the beta barrel in 2glwA which has 8 strands. Of course, the thiolase fold, with 7 strands and 5 or 6 helices in 170 residues is also not very compatible with the secondary structure prediction. This really looks likely to be a template-free prediction. Wed Jul 2 11:42:26 PDT 2008 Kevin Karplus The alignments that have bits of sheet are 1 2glwA 9 1q6hA 13 1g6oA 15 2oynA 16 2p19A none of which seem to be the thiolase fold. The thiolase alignments pulled out strands and helices, but not strands that were paired with each other. Wed Jul 2 12:52:00 PDT 2008 Kevin Karplus The try1-opt3 model looks like junk---no sheets formed. I may have to hand-add sheet costraints and try again. Sun Jul 6 09:22:03 PDT 2008 Kevin Karplus The MQAU quality assessment likes the Zhang-Server, BAKER-ROBETTA, and RBO-Proteus models. The MQAC quality assessment likes the Zhang-Server models, SAM-T08-server_TS1, PSI_TS5, RPO-Proteus_TS1, ... but the top GDT is predicted to be under 31%. I'll do MQAU1 and MQAC1 runs, but I don't expect much---I need to find proper sheet constraints to guide undertaker. Sun Jul 6 13:50:51 PDT 2008 Kevin Karplus The try1 costfcn likes MQAC1-opt3 best, then MQAU1-opt3. try1-opt3 is quite a bit worse. Sun Jul 6 13:58:50 PDT 2008 Kevin Karplus The MQAU1-opt3 model, from RBO-Proteus_TS1, actually has a decent sheet and helices that match the secondary structure prediction, and burial is not bad. The C-terminal helix may need some work, but this is the only promising model I've seen for this target. Perhaps I should give it to VAST, and look for other templates to force on the target. VAST finds that the MQAU1-opt3 model is similar to proteins like 2opdA=2opeA, 2ayuA, 1lrwA, 2poxA=2iovA, 1c9lA, 2z2rA, 1tl2A, 1cegA=1yqsA, 2ffgA, 2orcA=1orcA, ... (or, by length of alignment: 2cr9A, 2ra8A, 2eocA, 3bxoA, 2ffgA, 2gr8E=2gr8A, ...). 2opeA, 1tl2A, 1yqsA, 2ffgA, 2cr9A, 2ra8A, 3bxoA, 2gr8A are in template library, though none came up in the HMM hits. Thu Jul 10 10:58:56 PDT 2008 SAM-T08-MQAO hand QA T0482 Submitted Thu Jul 10 10:58:56 PDT 2008 SAM-T08-MQAU hand QA T0482 Submitted Thu Jul 10 10:58:56 PDT 2008 SAM-T08-MQAC hand QA T0482 Submitted Sun Jul 20 09:52:35 PDT 2008 Kevin Karplus I made try2.costfcn with the sheet and helix constraints from MQAU1. I started try2 to polish existing models (mainly MQAU1) and MQAX2 to see if there are any servers other than RBO-Proteus that come up with this structure. Sun Jul 20 10:02:01 PDT 2008 Kevin Karplus I also set MANUAL_TOP_HITS:=2opeA 1tl2A 1ysqA 2ffgA 2cr9A 2ra8A 3bxoA 2gr8A to make extra_aligments for the templates similar to MQAU1-opt3. The local alignments seem to be mostly empty, but the global alignments are full length (though scoring extremely badly). For try3, I'll do another run from alignments, but favoring these "manual" hits. The try3 costfcn will not have sheet constraints, but will have hbond_geom_beta_pair heavily weighted. Sun Jul 20 10:37:35 PDT 2008 Kevin Karplus MQAX2 is polishing Pcons_dot_net_TS4, which is probably BAKER-ROBETTA_TS1. try2-opt3 scores better than MQAU1-opt3, thinking it has reduced clashes, but Rosetta prefers MQAU1-opt3. Sun Jul 20 13:30:20 PDT 2008 Kevin Karplus Both the try2 and the try3 costfcn like try2-opt3 best. Rosetta likes MQAX2-opt3.gromacs0.repack-nonPC best. I like both the try2-opt3 and the MQAX2-opt3 models (both metaserver models). Sun Jul 20 13:46:34 PDT 2008 Kevin Karplus I started a try4 run to polish the MQAX2-opt3 model. I'm also doing an MQAX4 metaserver run, excluding RBO-Proteus, BAKER-ROBETTA, and Pcons_dot_net models, to see if there is anything else along the same lines. Sun Jul 20 15:59:47 PDT 2008 Kevin Karplus try4 polishes MQAX2-opt3 as expected, scoring better or the same on every component of the undertaker cost fcn (except sidechain), but rosetta still prefers MQAX2 to try4 (-opt3.gromacs0.repack-nonPC, in both cases). The difference is small for Rosetta, though. MQAX4 picks up GS-KudlatyPred_TS1. Sun Jul 20 16:14:11 PDT 2008 Kevin Karplus I should probably also try to optimize a model from the Zhang-Server models, since they scored best in MQAC quality assessment. Sun Jul 20 16:40:41 PDT 2008 Kevin Karplus On looking at the Zhang-Server models, I'm not sure they are worth rescuing---they have bad clashes and don't form the sheet very well. Even with the try5 costfcn which uses sheet and helix constraints extraxted from Zhang-Server_TS1 and Zhang-Server_TS2, the Zhang-Server models score poorly. The metaserver models I've already pulle dotu score better. The top server models with that try5 costfcn are BAKER-ROBETTA_TS1=Pcons_dot_net_TS4 USED for MQAX2, try4 RBO-Proteus_TS1 USED for MQAU2, try2 BAKER-ROBETTA_TS5 Pcons_dot_net_TS5 BAKER-ROBETTA_TS4 Pcons_dot_net_TS1 RBO-Proteus_TS5 RBO-Proteus_TS2 RBO-Proteus_TS3 GS-KudlatyPred_TS1 USED for MQAX4 The next model from a different server is FALCON_TS5 Sun Jul 20 17:04:30 PDT 2008 Kevin Karplus With try6 favoring the sheets and helices found in FALCON_TS1 and FALCON_TS5, the best-scoring serer models are BAKER-ROBETTA_TS1 Pcons_dot_net_TS4 BAKER-ROBETTA_TS5 Pcons_dot_net_TS5 RBO-Proteus_TS1 BAKER-ROBETTA_TS4 Pcons_dot_net_TS1 RBO-Proteus_TS5 BAKER-ROBETTA_TS2 Pcons_dot_net_TS3 FALCON_TS5 GS-KudlatyPred_TS1 RBO-Proteus_TS3 RBO-Proteus_TS2 So I'll try polishing the five FALCON models with try6.costfcn in order to get another possible model to consider. My big question for this target is still---why didn't SAM find any of the templates? Sun Jul 20 18:23:04 PDT 2008 Kevin Karplus try6 is based on FALCON_TS5, and it does not looks as good as the other metaserver models. I should try polishing try4-opt3.gromacs0.repack-nonPC and MQAX2-opt3.gromacs0.repack-nonPC, in an attempt to get the best model I can from the BAKER-ROBETTA_TS1 model. I should also try polishing the 3-strand variant of try3-opt3. Sun Jul 20 18:32:54 PDT 2008 Kevin Karplus Started polishing try4/MQAX2 gromacs optimized models as try7. Started polishing try3 models as try8. Sun Jul 20 20:25:38 PDT 2008 Kevin Karplus undertaker likes try7-opt3, but rosetta thinks that try7-opt3.gromacs0.repack-nonPC is worse than MQAX2. The residues that Rosetta doesn't like are L7, F2, L9, A32, G65, M1, S11. Perhaps cutting and pasting M1-S11 from MQAX2 to try7 and reoptiimzing might help. Sun Jul 20 20:37:37 PDT 2008 Kevin Karplus Acutally, I don't like how M1-L7 are set up in the models derived from BAKER_ROBETTA_TS1. The conserved F2, P4, and H6 do not seem to be doing anything interesting. I think that perhaps H6 should be interacting with Y26. The MQAX4 model has similar problems. The try8/try3 models don't cluster the conserved residues at all. The try2/MQAU1/RBO-Proteus model brings H6 and Y26 close, but has Y92, D93, D98, and F96 far away. Perhaps I should try again, with a different sheet constraint for the first strand than in try7. I do have a predicted Hbond for L12.O-L15.N. Extending this in a standard hairpin would give SheetConstraint H6 L12 N21 L15 Hbond L15 but this makes the turn too soon. There is another prediction for D16.O-D19.N, which extends to SheetConstraint H6 D16 R29 D19 Hbond D19 but this has the Hbonds on the wrong side of the D19-R29 strand. The str4 predictions suggest that evens hbond to evens, with I20 paired to V10 or L12. The Robetta-based models pair L12-I20, so this can be extended just a little to get SheetConstraint H6 L12 Y26 I20 Hbond I20 which would nicely pair the conserved H6 and Y26. Actually the sheet extraction already does the extension: SheetConstraint (T0482)D19 (T0482)Y26 (T0482)P13 (T0482)H6 hbond (T0482)Y26 1 I'll just up the weight of this sheet constraint, and add distance constraints for H6.NE2 and Y92.OH. I can't add another hairpin before this, despite the P4, G5 pair, since H6's hbonds are already taken. I am tempted to try to tuck F2 into the core, somewhere near Y26. Let's tie the CG atoms to be within 6 Angstroms. I'll put together try9.costfcn to try to force this behavior at the N_terminus. Try2 nearly has this behavior at the N-terminus, so let me put together a try2-try7 chimera, copying M1-V10 from try2 and the rest from try7. F2 and M1 will make horrible clashes, but perhaps they can be moved a bit to fix the problem. Rotating Y26 would help. Sun Jul 20 21:46:51 PDT 2008 Kevin Karplus I'm trying an optimization from chimera-try2-try7 as try9. Sun Jul 20 23:20:09 PDT 2008 Kevin Karplus undertaker likes try9, but rosetta hates it. Probably it has too many clashes still. I should come up with a new set of constraints for the N-terminus, up the clash and break costs, and re-optimize. F2 did not end up going where I wanted, but H6 did. Maybe I need constraints to pull it near V28 and A66. Once I get it near where I want it, we can try closing the gaps and removing clashes. Mon Jul 21 00:01:24 PDT 2008 Kevin Karplus try10 is like try9, but with different (better?) N-terminal constraints. It will start again from chimera-try2-try7. Mon Jul 21 06:45:28 PDT 2008 Kevin Karplus Once again, undertaker likes try10, and rosetta hates it. I rather like the packing of try10, so I'll do another optimization of it, with the N-terminal constraints turned off, but clashes and breaks turned up. Mon Jul 21 08:21:06 PDT 2008 Kevin Karplus OOPS. try11 tried to read try11 (instead of try10) models. Let's do that again as try12. Mon Jul 21 09:50:48 PDT 2008 Kevin Karplus try12 eliminated breaks, but still has some clashes. There is also a bit more foaminess than I'd like to see. I should try to squeeze it, though undertaker is not very good at packing things tighter without constraints. Mon Jul 21 21:30:17 PDT 2008 Kevin Karplus For try13, I turned up pred_nb11_back and the dry weights in an attempt to pack try12 a little tighter. I'll probably have to give up on this target soon, as I need to submit it tomorrow. Tue Jul 22 08:01:40 PDT 2008 Kevin Karplus try13 now scores better than try7 (with the try13 costfcn) I think I'll submit ReadConformPDB T0482.try13-opt3.pdb # < try12-opt3 < try10-opt3.pdb < chimera-try2-try7 # try7-opt3.pdb < try4-opt3.gromacs0 < T0482.MQAX2-opt3 < Pcons_dot_net_TS4 =? BAKER_ROBETTA_TS1 ReadConformPDB T0482.MQAX2-opt3.gromacs0.repack-nonPC.pdb # < Pcons_dot_net_TS4 =? BAKER_ROBETTA_TS1 # best rosetta energy ReadConformPDB T0482.try2-opt3.pdb #