Fri Aug 6 10:03:23 PDT 2004 T0264 DUE 27 Aug 2004 Fri Aug 6 22:01:54 PDT 2004 Kevin Karplus comarative model with 1pjqA as the template. try1-opt2 looks fairly good. We may just want to tighten the packing and declare it done. The rr constraints look fairly good, and may help a little in getting things to pack better, particularly between the two domains. I'll do one more run from the alignments (try2), then do a polishing run on all existing models. Sun Aug 8 16:18:05 PDT 2004 Kevin Karplus try2 looks horrible. The secondary structure is all messed up. The problem may be that the rr constraints were given too much weight. For try3, I will use a costfunction like try2, but with the rr constraints getting less weight, by scaling up the other constraints. Sun Aug 15 10:30:17 PDT 2004 Kevin Karplus try3 has gone off the deep end again. Luckily, try1-opt2 is still the best scoring with unconstrained.costfcn, so I'll just do a polishing run from the existing models with the unconstrained costfcn, perhaps turning down phobic_fit a little. Sun Aug 15 16:26:38 PDT 2004 Kevin Karplus try4-opt is new best score with unconstrained costfcn. it doesn't look bad, though I think that R61-E64 should be another parallel strand on the sheet. I'm not sure which way it should face though. I'm not confident enough of that prediction to go to a lot of trouble to try to make it happen. I'd submit (if I had to do it today) try4-opt2 best undertaker try4-opt2.repack-nonPC best Rosetta energy try1-opt2 full auto T0264-1pjqA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 T0264-1cbf-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 The two template alignments have the strand attached (though with quite different alignments), so I should be able to pick up the sheet constraint from align1.sheets or align2.sheets. I think I like align2.sheets better, since it matches the 2ry prediction: SheetConstraint (T0264)D34 (T0264)V38 (T0264)A60 (T0264)E64 hbond (T0264)V36 I'll do a run with constraints (try5) then another run from existing models without constraints (or with weakened constraints, if try5 is not best). Sun Aug 15 20:25:21 PDT 2004 Kevin Karplus The try5 run has finished, and it doesn't score quite as well as try4 on unconstrained cost fcn. The 4.4-point difference comes from soft_clashes, wayback, phobic_fit, sidechain, the pred_alphas, and the hbonds. try5-opt2 does look much better to me, so I'll optimize it using the unconstrained costfcn, leaving try4 out of the optimzation. Mon Aug 16 07:52:16 PDT 2004 Kevin Karplus try4 still scores very slightly better than try6 on the try6 (unconstrained) cost function, with the difference coming largely from "sidechain", but try4 also has more helical hbonds. I suspect that P44-K51 should be more helical in try6, but I don't think undertaker is likely to improve that. A bigger problem may be the exposure of G185, S189, and L192, which shoud be buried----that whole helix needs to turn over to pack better. It does pack better in try4, so maybe I should make a try6-try4 chimera. I did that, crossing over from try6 to try4 just before G122. The chimera scores best on the try6 costfcn. Maybe I should do one more polishing run. Mon Aug 16 20:57:41 PDT 2004 Kevin Karplus try7-opt2 is best after the polishing run, but Rosetta prefers try4-opt2.repack-nonPC I'll submit try7-opt2 best undertaker unconstrained try4-opt2.repack-nonPC lowest Rosetta energy try1-opt2 full auto T0264-1pjqA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template T0264-1cbf-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 2nd best Thu Nov 18 23:47:34 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 44.1205 (try-opt1-scrwl/try1-opt1) best submit 43.0970 (this shoulf be the same as the full auto, model3) model1 40.1910 auto 43.1026 align 30.8999 robetta best 55.3691 (robetta model3) robetta1 54.9459