Fri Jul 7 09:11:32 PDT 2006 T0363 Make started Fri Jul 7 09:12:33 PDT 2006 Running on lopez.cse.ucsc.edu Fri Jul 7 23:59:18 PDT 2006 Kevin Karplus No hits with BLAST in PDB, best is 1y9uA, E-value 7.8. Modest hits (E-value 2.e-04) to d.15.* (1vjkA, 1oqqA, ...) Fri Jul 14 16:08:35 PDT 2006 Navya Swetha Davuluri, Sylvia Do, Crissan Harris, Cynthia Hsu The target matches the top alignments perfectly from residues 14-84, but there isn't much to work with for the two terminal helices. We ran the conserved_t06 script and found that the exposed residues were mainly conserved ones, indicating an active site. We also checked to see if there were His and Cys residues. No Cys residues were found, so we set "maybe_ssbonds" to 0. We found a Histidine tag in the N-terminal helix, so we also set "maybe_metal" to 0. [Wed Jul 26 09:38:13 PDT 2006 Kevin Karplus maybe_metal only affects cys residues. ] Our main concern in this model was the large break between 23 and 24, so we raised the "break" weight to 100. try2 is currently running on shaw. Mon Jul 17 08:43:22 PDT 2006 Navya Swetha Davuluri, Cynthia Hsu The breaks in try2 were fixed, but the residues from Y20-Y25 formed a helix. We've decided to extend the beta sheet constraints on thsese strands, as follows: SheetConstraint (T0363)Q12 (T0363)Y19 (T0363)D33 (T0363)Y25 hbond (T0363)I13 5 Hopefully this will force the helix to form beta sheets with a turn. We also raised the overall weight of constraints to 20. try3 is currently running on lopez. Mon Jul 17 10:24:15 PDT 2006 Navya Swetha Davuluri, Cynthia Hsu A break was formed before residue G46, which also broke the helix from residues 37-49. Examining the best-models.pdb in RasMol indicated that these residues should be forming a helix. We also failed to turn residues 19-25 into a turn. To deal with the helix from T37 to S49, we copied the helix constraint from try1-opt2.helices pertaining to these residues and set the weight to 5. We also edited the residues in our sheet constraint to more accurately reflect the beta sheet desired, as followed (note that the constraint was also raised to 10): SheetConstraint (T0363)Q12 (T0363)A20 (T0363)V32 (T0363)R24 hbond (T0363)I13 10 The weight on "break" was increased to 180. We also uncommented the dssp-ehl2 and the rr.constraints. try4 is currently running on lopez. Mon Jul 17 12:50:51 PDT 2006 Navya Swetha Davuluri, Cynthia Hsu Under Professor Karplus's suggestion, we began polishing try4-opt2. We raised "soft_clashes" to 80. We also transferred in the sheet constraints from try4-opt2.sheets and lowered the weight on the sheet from Q12-A20 and V32-R24 back to 1. try5.under was set to polish the gromacs version of try4. try5 is currently running on lopez. Mon Jul 17 15:11:31 PDT 2006 Navya Swetha Davuluri, Cynthia Hsu try5 scored worse than try1 did in rosetta, but it scored the best in unconstrained scores. In view of this, we decided to attempt a polishing run on try1's gromacs model so that we could better close the breaks. For the try6.costfcn, we copied the try1.costfcn, and then changed the constraints as follows: "soft_clashes" was raised to 80, "break" was raised to 200, "dry5" was set to 20, "dry6.5" to 25, and "dry8" was set to 20. try6 is currently running on lopez. We also thought it might be worth trying to improve try5, given that it scored significantly worse in try1 in bad_peptide. For this reason, we copied try5.costfcn to try7.costfcn, and raised "n_ca_c" to 10. "break" was also raised to 200. try7.under accepted try5-opt2 as its input model. try7 is currently running on shaw. Tue Jul 18 11:17:47 PDT 2006 Cynthia Hsu I've updated looked at the unconstrained scores and the Rosetta scores, and modified superimpose-best.under as follows: ReadConformPDB T0363.try7-opt2.pdb ReadConformPDB T0363.try6-opt2.gromacs0.repack-nonPC.pdb.gz ReadConformPDB T0363.try5-opt2.gromacs0.repack-nonPC.pdb.gz //similar to try7, but more favored by Rosetta. ReadConformPDB T0363.try3-opt2.pdb ReadConformPDB T0363.try2-opt2.pdb I'm concerned because neither try3 nor try2 are very high scoring, but we don't have five diverse models. I'm not sure what I should do at this point. Thu Jul 20 11:50:59 PDT 2006 Cynthia Hsu Examining try3 in the near and burial script, it occurred to me that part of its low scores was due to the unraveling of its helices, which caused regions to be exposed that shouldn't have. I copied try3.costfcn to try8.costfcn, and first raised "soft_clashes" to 60 and "break" to 150. I also lowered "sidechain" to 4, hoping that this would allow Undertaker to focus more on closing the break. I also reincluded the dssp-ehl2 constraints and the rr constraints. I added the following constraints on the helices: //from try3-opt2.helices (but modified) HelixConstraint (T0363)M1 (T0363)H6 1 HelixConstraint (T0363)A18 (T0363)L27 1.5 // will force the helix to turn inward, burying certain regions. HelixConstraint (T0363)V38 (T0363)S57 3// combines two helices currently separated HelixConstraint (T0363)D89 (T0363)G97 1 I briefly considered removing the helix constraint from A18 to L27, as our preference had been for this to be a hairpin turn, but I felt that this might give us a chance to improve the score of try3 for a slightly diverse model. try8 is currently running on orcas. I also attempted to further modify try2, adding the following helix constraint to try9.costfcn for the same reason as I did in try8: HelixConstraint (T0363)V38 (T0363)S57 3 I also raised "soft_clashes" to 40. try9 is currently running on orcas. Thu Jul 20 13:24:25 PDT 2006 Cynthia Hsu try8 had an interesting fold that does not appear contradictory to the burial and near scripts. Interestingly enough, part of the region from A18-L27 remained as a sheet, and instead of forming an entire helix, just formed a loop. try9 also produced a uniqe fold. Undertaker scored it the fifth highest, after try7, try5, try6, and try1. In Rosetta, the try8 gromacs version was the third favorite (after try6 and try1). try1 could not be included as a best-model because it was almost identical to try6, so try3-opt2 was left as the fifth choice model. Thu Jul 20 14:17:49 PDT 2006 Kevin Karplus It may be worth looking at the top few server models: SAM_T06_server_TS1, ROBETTA_TS1, ROBETTA_TS5, ... I fetched all 10 robetta models, in case they are promising starting points. Thu Jul 20 16:27:09 PDT 2006 Cynthia Hsu Looking at the score-all+servers.unconstrained.pretty file, I found that only tries 7, 5, and 6 scored better than the top SAM_T06 model. However, comparing our top models with those of the servers, I found few differences. Mon Jul 24 12:51:36 PDT 2006 Cynthia Hsu Looking at try7, I found that it still appeared extremely foamy in some regions, particularly in the way the helix from residues 37 to 55 was poorly packed against the beta sheet. I considered this, and decided to experiment with increasing phobic fit, which I increased to 4. I also raied "dry5" to 20, "dry6.5" to 25, and "dry8" to 20. try10 was a polishing run that took try7-opt2 as the input model. try10 is currently running on lopez. Mon Jul 24 14:07:18 PDT 2006 Cynthia Hsu try10 looks fine under the burial script but still is slightly foamy under the near script. However, it is currently the best-scoring of all the models. Mon Jul 24 15:54:29 PDT 2006 Cynthia Hsu Given that try3 was a very poorly scoring model, and the fact that I am concerned about not having five good models to present, I've decided to attempt to fix the breaks in try3 and hope that it may produce a unique but decently scoring fold. I raised "soft_clashes" to 40, and lowered "sidechain" to 3. I also added the following helix constraint: HelixConstraint (T0363)T37 (T0363)S49 8 However, I set the overall weight of "constraint" back to 10, in hopes that this would help in fixing the break. Tue Jul 25 12:20:10 PDT 2006 Cynthia Hsu try11 still has several large breaks in try2, before T58, S65, S57, although they are less significant (they are around 1 angstrom, and are not visible in RasMol). In view of this, I decided to copy try11.costfcn to try12.costfcn, set "soft_clashes" to 60, and "break" to 140. I set up try12.under as a polishing run that would try11-opt2. try12 is currently running on lopez. Tue Jul 25 14:25:26 PDT 2006 Cynthia Hsu try12 effectively eliminated all breaks from the model. However, examining the unconstrained scores, try12 does significantly worse than all other models, so it may be best to discard this model. Instead, I decided that it may be to our advantage to polish the try11 model. Specifically, the dry weights, phobic fit, n_ca_c, and bad_peptide scores were all very poor, and soft_clashes could also be improved. I copied try11.costfcn to try13.costfcn, then raised "break" to 110 and "soft_clashes" to 60. I also raised "n_ca_c" to 10 and "bad_peptide" to 13. I raised the dry weights by a significant amount, as follows: dry5 25 dry6.5 30 dry8 20 dry12 6 I also raised the "constraints" to 12, in hopes that this will also improve the packing. try13 was set up as a polishing run of try11-opt2. try13 is currently running on lopez. I found that try12.under had an error, in which it used itself as the ReadConformPDB. I considered rerunning it with the same cost function, but decided against it as similar changes existed in try13 already. Tue Jul 25 16:24:38 PDT 2006 Cynthia Hsu try13 did slightly worse than try11, despite being a polishing run. I decided at this point that my attempts to improve the score of try11 were futile. Examining the scores, I was surprised to find that try12 was actually Rosetta's favorite. I decided that it would be in our interest, because of this, to rerun try12 as try14, but with the appropriate input file (try11-opt2). try14 is currently running on camano. Wed Jul 26 09:42:54 PDT 2006 Kevin Karplus When you have a model that is good except for small clashes or breaks, it may be worthwhile to do an undertaker optimization from the gromacs0.repack-nonPC version of the model. I'll try doing a polishing run on try12-opt2.gromacs0.repack-nonPC, just to see whether anything improves. Wed Jul 26 09:53:44 PDT 2006 Kevin Karplus try15 started on lopez to polish try12-opt2.gromacs0.repack-nonPC. try16 started on lopez to polish from all existing models (probably mainly try7-opt2). Wed Jul 26 10:18:16 PDT 2006 Kevin Karplus Although try16 started out concentrating on try7-opt2, it repeatedly shifted attention to try10-opt2. Wed Jul 26 10:23:42 PDT 2006 Kevin Karplus I finally looked at try12----it is complete junk with no sheets at all. Scrap that, and scrap try15 also. Other than those trash models, we are seeing only small differences. The cores of our models are essentially all the same, with different termini and slightly different loops. It looks like try16-opt2 will be the new best-scoring, but I don't know about what rosetta will like. If it doesn't like try16, I should try polishing try6-opt2.gromacs0.repack-nonPC Wed Jul 26 10:43:34 PDT 2006 Kevin Karplus try17 started on cheep, polishing try6-opt2.gromacs0.repack-nonPC with the same costfcn as try16. (as it turns out rosetta *doesn't* particularly like try16-opt2 as a base.) Wed Jul 26 11:36:26 PDT 2006 Navya swetha Davuluri rosetta doesn't like try17-opt2 either. I am going to polish try6-opt2.gromacs0.repact-nonPC by increasing weight on dry8 to 25 and on dry6.5 to 35. try18 started on peep. Wed Jul 26 12:40:37 PDT 2006 Kevin Karplus try18-opt2 does fairly well, though undertaker prefers try16, try7, try10 and try17. Rosetta likes decoys/T0363.try18-opt2.gromacs0.repack-nonPC.pdb better than anything other than the trashy try12 and try15 runs. I'll replace try6-opt2.gromacs0.repack-nonPC with T0363.try18-opt2.gromacs0.repack-nonPC in superimpose-best.under. I think we have reached the point of dimishing returns, and I'll prepare the final submission. ReadConformPDB T0363.try16-opt2.pdb ReadConformPDB T0363.try18-opt2.gromacs0.repack-nonPC.pdb.gz //favored by Rosetta, but has a break before S65 ReadConformPDB T0363.try3-opt2.pdb //poor-scoring, but adds diversity to the model. ReadConformPDB T0363.try9-opt2.pdb ReadConformPDB T0363.try8-opt2.pdb try16-opt2: best-scoring with break-sensistve costfcn (try16), < try10-opt2 < try7-opt2 < try5-opt2 < try4-opt2.gromacs0.repack-nonPC < alignments (1vjkA) try18-opt2 .gromacs0.repack-nonPC < try6-opt2.gromacs0.repack-nonPC < try1-opt2.gromacs0.repack-nonPC < alignments (1vjkA) try3-opt2 < alignments (1fm0D) try9-opt2 < alignments (1vjkA) try8-opt2 < alignments (1vjkA) I don't know that try3, try8, and try9 are the best three models to add to the top 2 (try16-opt2 and try18-opt2.gromacs0.repack-nonPC), but they'll do. There are, at least, some noticeable differences in the loops and termini. Maybe I should look at try14 also, as it scores well with some costfcns. It has a *very* different handling of the loop F21-Y26, and I think that dropping try9 to include it might be a good idea. try14-opt2 < try11-opt2 < alignments (1vjkA) Wed Jul 26 13:20:23 PDT 2006 Kevin Karplus Submitted with comment For T0363, we had modest fold-recognition hits to 1vjkA and similar proteins. We optimized mainly with undertaker, ocassionally running through gromacs and repacking sidechains with rosetta to get out of a local minimum. We have selected 5 models based partly on score and partly on the desire to have a variety of different loop configurations. Model 1 is try16-opt2, which appears ot have done the best job of closing gaps. try16-opt2 < try10-opt2 < try7-opt2 < try5-opt2 < try4-opt2.gromacs0.repack-nonPC < alignments (mainly 1vjkA) Model 2 is try18-opt2.gromacs0.repack-nonPC, the model that rosetta liked best of all the backbones for which we had it repack sidechains (not counting a couple of models that had no beta sheets). try18-opt2.gromacs0.repack-nonPC < try6-opt2.gromacs0.repack-nonPC < try1-opt2.gromacs0.repack-nonPC < alignments (mainly 1vjkA) Model 3 is try3-opt2, which scored rather poorly, but was one of the few models that did not seem to be based mainly on 1vjkA. try3-opt2 < alignments (1fm0D) Model 4 is try8-opt2, which is included mainly for diversity of loops. try8-opt2 < alignments (1vjkA) Model 5 is try14-opt2, which is included mainly for diversity of loops. try14-opt2 < try11-opt2 < alignments (1vjkA)