13 June 2002 Kevin Karplus T0133 is a homology modeling target, with clear homology to 1hg5A. I think that the helix at 182-201 is slightly misplaced in the alignment chosen---moving it a couple of residues later may help. 25 June 2002 Kevin Karplus Looking at T0133.undertaker-align.pdb and T0133.try2-opt.pdb, I see that none of the alignments we're trying have strands in the predicted places. The tiny bits of beta we see are all in areas predicted to be helical. T0133-try1-opt.pdb scores slightly better than T0133.try2-opt.pdb. I'm not sure what to do to make this homology model better---there are lots of regions that still look a bit random, and we may be better off just using one of the alignments (which one?). 14 July 2002 Kevin Karplus Trying again, with the new fragments from the improved fragfinder. The scores have gotten a little better, but predicted secondary structure still doesn't match at all. Is the secondary structure prediction wrong, or is the alignment terrible? 17 July 2002 Kevin Karplus try4-opt-scwrl is the new best score, but it has unstacked the helices at 28-58 from the ones at 61-97, leaving a big hole in the protein. Maybe I should add a weak constraint to put I39 near F86. 303 CG2 ILE A 39 681 CE1 PHE A 86 18 July 2002 Kevin Karplus try5-opt-scwrl doesn't seem to do any better---it makes only minor improvements over try4-opt-scwrl. Maybe I should try again without using try4-opt-scwrl as a starting point. 19 July 2002 Kevin Karplus try6 seg faulted after a failed scwrl run, but the best of the 14 or so iterations was T0133.try6-al10+1jqcA-T0133-local-adpstyle1.pw.a2m.gz:1jqcA.10.20.pdb try7-opt-scwrl is the new best-scoring run, but it still has the helices moved apart, making a hole. Let's try again, using only the random conformation as a starting point and inserting all alignments. 23 July 2002 Kevin Karplus try8-opt-scwrl has kept the helices together--in fact it may have banged I39 and F86 into each other too hard, as they clash a little. Try7-opt-scwrl still scores much better than try8-opt-scwrl. Changing the score function to use the new pred_alpha cost function does not make try8 beat try7. I'll start from try8-opt-scwrl with the new function and see what happens. Although try9 scores better than try8, it still is not close to try7. OOPS---I mis-typed "pred_alpha2" in the cost definition, so the alpha costs were never used. Let's try again, starting from try9-opt, which does score somewhat better than try8. Try10-opt-scwrl does better than try9, but still not close to try7-opt-scwrl. 23 July 2002 Kevin Karplus try10 looks terrible also. I think I need to go back to try7 and start working forward from there. 24 July 2002 Kevin Karplus try11-opt is the new best scorer, doing very slightly better than try11-opt-scwrl, which is slightly better than try7-opt-scwrl. I think that this may be done now---I don't have any ideas how to improve it, except to remove the constraint for I39-F86. 24 July 2002 Kevin Karplus After removing the constraint and reoptimizing, the best scorer is T0133.try12-opt (not try12-opt-scwrl, which scores worse than try11-opt). The differences are very small. 11 August 2002 Kevin Karplus Modernized the score function with phobic_fit and dry5, and reran make with new template library---1hg5A is still best, but picked up 1gw5A as another good hit. Will run undertaker again with alignments but no seed conformations. 13 August 2002 Kevin Karplus Top score is now try13-opt-scwrl. There are a few breaks that would be nice to fix---tweaking the parameters to make breaks more important still leaves try13-opt-scwrl as the best score. 15 Aug 2002 Kevin Karplus try14 run took about 12 hours, and the best score is now try14-opt-scwrl, but I haven't looked at it yet. 22:26 try14-opt-scwrl looks pretty good, though there are still some breaks to fix. Fri Aug 16 13:11:05 PDT 2002 Kevin Karplus try15-opt is new best score. It is probably not worth fussing with this any more. I'm going to submit try15-opt.