Wed Aug 7 19:33:19 PDT 2002 t0171 Obvious homology model for 1a8q, 1a8s, 1cr6B, ... (c.69.1.12 or c.69.1.11). Since both families are scoring so well, maybe I should look at SCOP to see what is the difference---this may help decide which alignments to use. We may have too many alignments with strong scores, and so I may have to tighten the threshold in the Makefile. 9 Aug 2002 Kevin Karplus T0171.try1-opt-scwrl looks pretty good. There are still a few breaks to close, so let's run again. 10 Aug 2002 Kevin Karplus try2-opt-scwrl is new best. Breaks have been reduced, but not eliminated. The beta sheet remains intact, so lets just run again. Hmm---I haven't been running with scwrl_each, so let's turn that on. We can also turn up the break and clash penalties a bit, and dry5 (to concentrate on tighter packing). (Try2-opt-scwrl still best with new score weighting.) 22 Aug 2002 Kevin Karplus try3-opt is new best. Probably should do one more polishing run tonight and try the VAST alignment trick in the morning. 23 Aug 2002 Kevin Karplus try4.8.100 is new best (not try4-opt). I'm submitting it to VAST, to see if I can get a better alignment to polish---run VS30765 password casp5t0171 Hmm, some of the helices looked sloppy enough that VAST didn't see them, but I think the rest may be enough to get the right template. Fri Aug 23 17:58:15 PDT 2002 This VAST run is taking forever! 20:58 OK, it's done. Best hits are 1a88A 2Ang over 247 and 1a8uA (1.7Ang over 250). Also high are 1brt, 1broA (1.6 over 248), 1a8s, 1a8q, 1bez, ... I picked up the 1a88A and 1broA alignments and hand edited them to position the gaps a bit better. I'll do a run using just these alignments as starting points, and check in the morning to see whether try5 or try4 turned out better with the score functions. I'll submit whichever it is. 24 Aug 2002 8:03 Best score is still try4.8.100. try5-opt scores worse than many of the try3 runs. Let's do a polishing run with lots of crossover, then submit whatever ends up best. 17:11 Best is now try6.4.110. It sure looks like C31 and C246 should have a disulphide bond---that pair are so close. With a disulphide bond score added to the score function, the best scoring is still try6.4.110. Let's do ONE more polishing run, with the disulphide bond score on. 20:44 Best is now try7-opt, and the disulphide is formed. Let's do one more hour of polishing, to try to close up the remaining gaps. I increased the break penalty, the disulphide bonus, and dry8 slightly, which moved try7.0.80 to the top of hte list. Try8 will start with try7-opt and the 3 results of the try7 iterations. It should be a short polishing run. 21:42 I made a pretty good guess about how many generations I could run in an hour---within a couple of minutes! Other than the rather strange loop at S112-D123, which sticks out leaving a hole underneath it, try8-opt does not look too bad. Ther problem was there in try7 also, so it is not a new introduction. There is also a problem with the burial of K184, as it looks like that whole helix needs a half turn. Perhaps a constraint between L86.CD2 and I182.CG1 would help flip it, though I'd probably have to break it by moving V175-D187 and repacking. If we're going to break and repack, we might want to grab F111-W117 and P118-V124 and move them also, so that the funny loop can get repacked. 23:31 In try9-opt, K184 has gotten out, but the helix has been lost and has not gotten packed into position. I'll try a few hours of optimization, but I suspect that this try is doomed, and I'll have to go back to try8-opt. If try10 DOES work, then I'll probably want to do one polishing run starting from both try8-opt and try10-opt. 25 Aug 2002 9:48 Kevin Karplus In try10.2.100 (the best-scoring of the try10 run), D174-L184 has been stretched out into a long extended region (rather than the predicted helix), and there is a big hole exposing W22, L59, L61, L83, L86, V87. Try8 also had trouble with this area, but try8-opt still scores best. If we score without the try9.constraints, the best scores are for try8-opt (and try7.0.80, which it is largely based on), but for the try10 runs the best are try10.0.100, try10.2.100, try10.1.100. Increaseing the wet6.5 weight (and the dry5 and dry6.5, to keep the chain from coming apart), moves try10.2 before try10.0. Let's do a crossover run with try8-opt, try10.2,100, try10.0.100, try10.1.100, and try10.7.100. 13:03 It looks like try11 mainly improved slightly on try8-opt. If I had more time, I'd work more on this, but I'm way behind schedule, so I'll just submit try11-opt.