Tue Aug 6 14:31:50 PDT 2002 t0170 Note: T0170.doc.html says no disulphides and monomeric. Not an obvious homology target. Best scores are for 1b2vA (d.35.1.1), but 2igd (d.15.7.1) also seems possible. Let's grab alignments from 1b2vA, 1dk0A, 2igd (skip 1igd and 2igh, which are identical sequences as 2igd, skip 1pgx which is 99% identical), 1g59A, 1gln, 1gb4, 1em7A Hmm---undertaker seems to be liking 1g59A best, though that is a.97.1.1 and c.26.1.1. If it looks bad, I'll have to remove those alignments from the undertaker initial set, and try again. 7 Aug 2002 Kevin Karplus This seems to be a small, all-helical protein. We may want to increase the weight for pred_alpha2. 8 Aug 2002 Kevin Karplus Lowest cost is for try2.1.80, which is somewhat lower than third best (try2-opt). The helices are NOT staying straight, so maybe we should add constraints. K12-E25 should be straight, as should W34-I41 and L52-Q66. Adding constraints Constraint 96 205 18 19.5 25 // CAs K12 E25 Constraint 273 335 9 10.5 14 // CAs W34 I41 Constraint 421 536 19 21 27 // CAs L52 Q66 still makes try2.1.80 score best. Let's try again, still not using a seed conformation, but with the constraints added and with no initial alignment sampling, so that more random conformations will be sampled. 22 Aug 2002 Kevin Karplus Current best is try3-opt-scwrl, beating robetta models handily. Let's do a polishing run. Fri Aug 23 16:23:12 PDT 2002 best is now try4-opt. I don't like the burial of K57, and helix A3-L22 is bent at K12. Increasing wet6.5, decreasing phobic_fit, increasing constraints, to try to get straighter helix and and unbury K57. Try4-opt still scores best. Let's try another run starting from a random conformation. 24 Aug 2002 Kevin Karplus try4-opt still scores best. try5-opt has ok burial, but C-terminal helix juts out rather than packing well. Doubling constraint weight to 0.1 still leaves try4-opt on top. Let's try a run from all the try*-opt files, with lots of crossover. 24 Aug 2002 Kevin Karplus 23:17 try6-try4.14.100 is new best. It looks pretty good, except for the burial of K56 and K57. Perhaps I need to break off the helix and move it so it can repack? 29 Aug 2002 Kevin Karplus 15:26 Try 7 didn't work. Best still try6-try4.14.100. Perhaps I need to do another optimization run with wet6.5 turned up from here. 16:55 With wet6.5 turned up to 3 (much higher than usual), try6-try4.14.100 still scores best. 20:58 The best scoring is now try8.2.100, but K57 is still buried. Opps-I see that the length constraint for the last helix was mis-typed, and so never used. Tweaking the length constraints and putting in this last one, still leaves try8.2.100 as best. Let's add a constraint for F60.CD1 and W34.CD2, to give the helix some incentive to turn over. I've started 2 runs---one from try8.2.100 (try9), the other from a random conformation (try10). Fri Aug 30 09:36:35 PDT 2002 With the try9+helix constraints, the best score is for try9.7.150, which STILL buries K57. This one also scores best with just the helix constraints and with no constraints. try10-opt is much foamier and buries K56, but it has a clean helix for L53-T67, starting 3 earlier than try9.7.150. Let's try cutting and pasting the clean helix into try9.7.150 and reoptimizing. try11 will be a repacking run with helix constraints and try9 constraints. try12 will be a repacking run with just helix constraints. Fri Aug 30 14:06:57 PDT 2002 Kevin Karplus With just the helix constraints, try11-opt scores best K57 is not buried, but K56 only avoid being buried by exposing a lot of hydrophobic residues in the middle, and turning L50 and L53 to the outside. try11-opt also scores best with the try9 constraints on as well. If we turn wet6.5 down from 3 to 1 and use only the helix constraints, the best score is still for try11-opt. Let's re-optimize with helix-only constraints from try11-opt, try12-opt, and try9.7.150. 31 Aug 2002 09:18 Kevin Karplus Current best is try13-opt, but it is a bit pried open to avoid burying K56. If I turn down wet6.5 from 1 to 0.3, the best-scoring is still try13-opt. If I turn breaks up from 15 to 35, the best-scoring is still try13-opt. If we add hydrogen_bonds to the scoring function, the best is try13-try11.9.200, which looks way too loose. It isn't the Hbonds that do it though, but the interacton of break, clashes, and the wet and dry parameters. Balancing clash avoidance with tight packing is always rather tricky. Let's reoptimize (without hydrogen-bond scoring) from try13-opt, try13-try11.9.200, try13-try11.0.200, try12-opt, try11-opt. 15:46 Best is now try14-opt. It looks pretty good, though a bit looser than I'd like. If we turn of the helix constraints, turn down pred-alpha2, and turn up phobic_fit (basically twiddling parameters randomly), the best is still try14-opt. Turning up dry8, and turning break down to 10, the best is try14.1.200. Let's do one more optimization run with these parameters. 21:47 Current best is try15.3.200 Upping the break weight to 30 and turning on helix constraints makes try14-opt score best. I like the looks of try14-opt better. Upping dry8 to 8 to encourage folds that don't expose F19 and W9 so much leaves try14-opt on top. I don't seem to be accomplishing much with this one any more. Let's submit model1 try14-opt model2 try6-try4.18.100 model3 try5-opt 25 November 2002 Got the real pdb file from CASP5 today (1h40). I tried to do rmsd measurements, but # command:# Reading conformation from PDB file T0170.real.pdb Bad short name: OT1 for alphabet: pdb_atoms Bad short name: OT2 for alphabet: pdb_atoms PDB file T0170.real.pdb has 69 residues and 564 atoms, but target has 69 residues and 568 atoms, so can't set conformation. I'll need to do a plot of score vs. rmsd. 26 November 2002 I made changes to undertaker and scored the models. The best fit is robetta3.pdb.gz---NONE of what we did improved things. At least model1 better than model2 better than model3. RMSD-CA RMSD robetta3 6.46541 7.21132 model1 try14-opt 12.6002 13.3334 model2 try6-try4.18.100 13.1256 13.9822 model3 try5-opt 13.5117 14.0476