Mon Jun 12 08:53:18 PDT 2006 T0326 Make started Mon Jun 12 08:55:02 PDT 2006 Running on lopez.cse.ucsc.edu Mon Jun 12 09:00:08 PDT 2006 Kevin Karplus Superb full-length BLAST hit to 2ghrA 56% identity over 301 residues, (E-value 4e-96). Nothing else comes close, so this may be a single-template predictions. VAST claims high structural similarity between 2ghrA and 1o1yA, 1wl8A, 1gpmA, 1qdlB, 1cs0B, 1gpwB, 1t36B, ... so we are likely to have multiple templates. These are mainly phosphatases, so we should look for the active site and make sure to add constraints to the sidechains there. The CASP organizers said "One of the targets to be released on Monday, T0326, is a quite high sequence similarity target. We will ask our high accuracy assessor to pay special attention to side chain accuracy in predictions for this target. So, please do your best in refining side chains for it." Mon Jun 12 09:57:07 PDT 2006 Kevin Karplus The t06-w0.5 model does indeed find 2ghrA as best, but 1wl8A, 1gmpA, 1o1yA, ... are the next highest hits, so it is finding the structural homologs as well. The fully automatic run will probably have too much emphasis on constraints, which are not all that useful for this level of similarity (unless there is a big insertion, which seems unlikely given the lengths). Mon Jun 12 11:54:08 PDT 2006 Kevin Karplus There are at least 23 templates in the t2k library for c.23.16.1 (79 sequences for the family in scop, thoug hsome many are redundant copies). There will be plenty of templates to provide variability in the backbone (if any such variability is needed). Make started Mon Jun 12 21:44:15 PDT 2006 Running on cheep.cse.ucsc.edu Make started Mon Jun 12 22:29:13 PDT 2006 Running on cheep.cse.ucsc.edu Mon Jun 12 22:47:35 PDT 2006 Kevin Karplus I had to restart the job, because the mucking around with the Make.main file I did this afternoon killed the initial run. It looks like Q146, C142, H234, and E236 are conserved residues at the active site. We should probably look at the corresponding residues in 2ghrA to make constraints on them. Tue Jun 13 07:43:20 PDT 2006 Kevin Karplus The try1-opt2 run moved the C-terminal helix slightly, and made up something for the extra N-terminal stuff to do. We should make a dimer based on 2ghr before trying to do further optimization, as the N-terminal stuff may be blocking the dimerization interface. Unfortunately, 2ghr has only one chain, and relies on the REMARK 350 field to create the symmetries. I will have to modify undertaker to read these rotation matrices and create the transform. I've never been sure which way the rotation matrices are written in pdb (using row or column vectors). Given the way the offset is written, I assume it is matrix*col_vector + col_vector. For 2ghr, it doesn't much matter as the rotation matrix is symmetric (swap x and y and negate z). Tue Jun 27 15:32:15 PDT 2006 Grant Thiltgen I started try2 on orcas. Try2 is an optimization run of try1-opt2, with increased weights on the dry weights, near_backbone, clashes, sidechain, break, soft_clashes, phobic_fit, and hbond_geom. Wed Jun 28 17:03:38 PDT 2006 Grant Thiltgen I started try3 on orcas. I dropped phobic_fit back down to normal and I increased the t04 and t06 pred_alpha constraints. I did something to try2 that pulled the protein apart, so I'm trying again. Thu Jun 29 14:05:58 PDT 2006 Kevin Karplus The sidechain weights in try3 are too high---the tiny differences in preference for different rotamers is swamping out much more important signals. Thu Jun 29 14:24:12 PDT 2006 Kevin Karplus I picked up the dimer 2ghrAB.pdb from the PQS website: http://pqs.ebi.ac.uk/pqs-doc/macmol/2ghr.mmol Now let's see if I can remember how to make a dimer from a specific PDB file. Thu Jun 29 14:31:58 PDT 2006 Kevin Karplus No problem, there were comments in the example file I used to create make-dimer1.under that explained how to use ReadTemplatePDB to read in the pdb file. I did have to read it in by chains, instead of by models, since I got the dimer from PQS rather than from PDB, and they use different conventions for the multimers. Thu Jun 29 14:37:59 PDT 2006 Kevin Karplus The dimer makes it very clear that we have the junk at the N-terminal in the wrong place in try1-opt2. I'll also check try3-opt2 as a dimer, and see if it does any better. Thu Jun 29 14:50:43 PDT 2006 Kevin Karplus try3-opt2 has almost exactly the same problem and try1-opt2. I'll try doing an optimization of the dimer to see if this stuff will move out of the way (I've upped the weight of ReduceClash considerably), but if it doesn't, then we'll probably have to ProteinShop it out of the way. Thu Jun 29 15:14:12 PDT 2006 Kevin Karplus Oops, I forgot to put "multimer 2" in the dimer/try1.under file, so it will optimize the two monomers separately. This may be good or it may be terrible. I'll start a try2 with the same costfcn, fewer alignments, and with multimer 2 in the OptConform commands. Thu Jun 29 15:17:46 PDT 2006 Kevin Karplus dimer/try2 started on lopez Thu Jun 29 17:43:23 PDT 2006 Kevin Karplus dimer/try2 does seem to be getting lower costs than try1, so it was important to tell undertaker to treat it as a dimer. Thu Jun 29 22:26:03 PDT 2006 Kevin Karplus dimer-try2-opt2 does not look too bad. The N-terminal region is junk, but at least it isn't in the way of the dimer. Perhaps I3 and V5 should fold down close to V177 and C291? I think that there is a strand wrong in all the monomers. Strand R227-G233 should be flipped over (slid over one residue) relative to its partner strand Y219-N223 to make the predicted residue-residue contacts make more sense. How much to I trust rr, though? All the top alignments have the same alignment for this pair of strands. Do I dare to mess with it? The top alignments have SheetConstraint (T0326)Y219 (T0326)K224 (T0326)T232 (T0326)R227 hbond (T0326)V221 87.7554 If I go down the alignment list far enough, I find #sheet constraints from conformation 7 T0326-1g2iA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:1g2iA SheetConstraint (T0326)A216 (T0326)V218 (T0326)G233 (T0326)V231 hbond (T0326)G217 13.9646 (a shift of 2, which doesn't help) and #sheet constraints from conformation 8 T0326-1ox6A-t04-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:1ox6A SheetConstraint (T0326)V220 (T0326)K224 (T0326)T232 (T0326)Q228 hbond (T0326)A222 10.9586 (a shift of 1 with a flipping of the hbonds) Perhaps I should try this constraint. No, I don't think so. When I look at the actual RR constraints, I see that the coloring pattern of the rr script was very misleading. We have constraints for I229---Y219 / F140----F230---V220---A211 / X A144 V231---V221---L210 which seem to indicate that if any strand is flipped in the current models it is Y219-K224. The pairings of V220 indicate the it belongs with F140, A144, F230, A211, G148---all of which cluster on one side of the sheet. If we pair V220 with F230, which seems to be the center of the complex, and keep the anti-parallel sheet, we get SheetConstraint E208 A211 N223 V220 Hbond E208 SheetConstraint V218 N223 T232 R227 Hbond Q228 This does have an unusual separation 4 end to the hairpin (with hbonding at separation 6, though, which is not so unusual). Thu Jun 29 23:24:44 PDT 2006 Kevin Karplus I started a new optimization from alignments with these constraints (plus the other sheetconstraints from try1-opt2, plus neural-net constraints and the top few rr constraints) as try4 on cheep. When it is finished, I'll have to dimerize it, maybe make a chimera with dimer/try2-opt2 to get N-terminus out of the way, and optimize it as a dimer. Thu Jun 29 23:50:36 PDT 2006 Kevin Karplus Just noticed that try4.costfcn had some of the bad try3 weights, so I killed try, fixed the weights, and will rerun try4. Fri Jun 30 08:05:58 PDT 2006 Kevin Karplus try4-opt2 scores well and doesn't look too bad. There are a few gaps to close. It does *NOT* succeed in flipping over the strand---I must not have an alignment that does that, so I'll need to create a hand alignment. I made a dimer form try4-opt2 with make-dimer3.under, but the N-terminal part interfers with dimerization, so I'll have to paste on the N-terminus from dimer/decoys/try2-opt2. I made decoys/chimera-dimer2-try4.pdb.gz from M1-D32 of dimer/decoys/T0326.try2-opt2.pdb.gz chain A I33-H304 of decoys.T0326.try4-opt2.pdb.gz Fri Jun 30 10:15:54 PDT 2006 Kevin Karplus I made dimer/decoys/dimer-chimera-try4-2ghrA.pdb using make-dimer4.under I'm thinking about how to fix up 2ghrA/T0326-2ghrA-hand.a2m Looking at the sequences, it looks really hard to flip the strand over---we have a long gapless alignment block, and the residues that may be important all match very well---sliding any part over by one is going to make the matching much worse, as well as creating gaps that might be hard to close. OK, I'm going to disbelieve the hints from the RR signals and just use the alignments we have---no hand alignment. I'm trying to optimize dimer/decoys/dimer-chimera-try4-2ghrA.pdb (dimer/try3 started on cheep) After that, I'll probably want to do a polishing run on the dimer models, and perhaps a polishing run as monomers. Sat Jul 1 07:34:14 PDT 2006 Kevin Karplus Started dimer/try4 on cheep to polish dimer models Sat Jul 1 11:23:16 PDT 2006 Kevin Karplus I'm not convinced that the dimer/try4 has the right dimerization---the N-terminal region is not well-packed against the rest. Let me try optimizing chain A of the dimer as a monomer, perhaps with constraints to pack A11 and L15 near L55 and I3 near V67. Sat Jul 1 11:43:20 PDT 2006 Kevin Karplus Started try5 on cheep to optimize just chain A of dimer/try4-opt2. Sat Jul 1 14:51:18 PDT 2006 Kevin Karplus It looks to me like we could improve try5 by adding a SheetConstraint SheetConstraint P2 P6 Q64 T68 Hbond V65 (There is another possiblity which might be harder to form, in which the strand G19-M23 is between these two strands: SheetConstraint P2 P6 G19 M23 Hbond F21 SheetConstraint G19 M23 Q64 T68 Hbond V65 ) Sat Jul 1 15:07:05 PDT 2006 Kevin Karplus I set up try6 to optimize try5 using the first SheetConstraint, and I'll set up try7 to optimize using the pair of SheetConstraints. Sat Jul 1 15:13:24 PDT 2006 Kevin Karplus try6 started on cheep try7 started on shaw Sat Jul 1 21:13:51 PDT 2006 Kevin Karplus Neither try6 nor try7 manages to make an extra strand at the N-terminus, and Rosetta prefers repacking try5-opt2. The unconstrained costfcn prefers try6-opt2, because of slightly better packing and hbonds. I should probably optimize try6-opt2 once in the context of a dimer, and declare it done. The N-terminus is probably wrong, but I don't think we have time to come up with a better one. I made the dimer from 2ghrAB using make-dimer5.under, and called it dimer/decoys/dimer-try6-2ghrA.pdb.gz I'll optimize it as dimer/try5 Sat Jul 1 21:28:11 PDT 2006 Kevin Karplus dimer/try5 started on lopez Sat Jul 1 21:30:03 PDT 2006 Kevin Karplus I scored all the server models with try5.costfcn (the monomeric, not the dimeric costfcn). Zhang-Server_TS5 does surprisingly well. I might want to score it with the try6 and try7 costfunctions also, and polish it up. Hmm---it looks more like it is trying to satisfy try6 than try7. Sun Jul 2 09:16:18 PDT 2006 Kevin Karplus Hard deadline tomorrow on this target, so I think I should just send off what I have now, though I'm still not happy with either the N- or the C-terminus. I haven't seen anything by Grant or Navya on this target since Wednesday, so I assume they have dropped work on it. Sun Jul 2 09:40:31 PDT 2006 Kevin Karplus Submission done, with comment The T0326 target had a fairly easy comparative model for the core, but the N- and C-termini are still rather uncertain. The initial models did not work well in a dimer based on 2ghr[AB], as the N-termini conflicted. Optimization in a dimer context tucked the N-terminus out of the way, but did not really pack it tightly against the rest of the monomer. (The 2ghr[AB] model has an interaction between the N-termini that does not seem very feasible with the extra residues of T0326.) Model 1 is dimer/try5-opt2 chain A, the last optimization in a dimer context that we had time for. Model 2 is dimer/try2-opt2 chain A, an early optimization in dimer context. Model 3 is dimer/try1-opt2 chain A, the first optimization in dimer context. Model 4 is try1-opt2, the fully automatic model (which has serious clashes in a dimeric context) Model 5 is sidechain replacement by SCWRL on a alignment to 2ghrA. Sun Jul 2 09:42:36 PDT 2006 Kevin Karplus I don't plan to do more work on this target, but if Grant and Navya do more on it, we can resubmit up to early Wednesday morning. Mon Sep 4 15:36:13 PDT 2006 Kevin Karplus We seem to have done pretty well on this target. model 1: GDT 78.9% RMSD 3.63, RMSD_CA 3.12 Zhang-Server_TS1: GDT 79.9% RMSD 9.65, RMSD_CA 10.13 We messed up T74-F88 and rotated G8-T24, but otherwise got the model pretty much right. I think GDT is misleading here, as the Zhang model really messed up the N-terminus. We did not submit our best models (try[567]-opt2.gromacs0). (Actually, we haven't evaluated the dimer models yet, other than the monomers based on them.) Zhang-Server_TS5 (which we had noted scored well in undertaker) was indeed the best of the server models, with RMSD 3.63, almost identical in score to our model1.