Mon Jun 24 09:22:25 PDT 2002 This looks like a homology model with 1lba or 1aroL (both d.118.1.1 SCOP domain) as the template. 25 June 2002 Kevin Karplus For some reason, the undertaker try1 run has picked out a totally different template---one that has strands where helices predicted and other bad properties. T0141.try1-opt.pdb looks like a good protein, but has little respect for the predicted secondary structures. I'll have to try again, perhaps tightening up the initial set of alignments to try, by resetting BEST_EVALUE and NUM_BEST and remaking the best-scores.rdb file. T0141.try2-opt.pdb has blown the beta sheets apart, and has lots of random coil. In T0141.undertaker-align.pdb, models 5 and 6 (1lba) have a reasonable structure with a few breaks that need to be closed. The 2ry structure agrees with the prediction. Let's start from T0141-1lba-2track-global-adpstyle5.pw.a2m.gz as a seed alignment, and see what happens. None of the input alignments have a helix for 40-47, so that region never gets a helix in undertaker. T0141-try3-opt.pdb blows the beta sheet apart and looks like molten globule or worse. 29 July 2002 Jonathan Casper Adding SCWRLing and pred_alpha to the undertaker script. 11 Aug 2002 Kevin Karplus Reran make with the new template library. No changes among the top hits. Modernized score function (try5-opt-scwrl now scores best). Modernized undertaker script and reran, starting from just alignments, not seed conformation. 12 Aug 2002 Kevin Karplus try6-opt-scwrl is new best score. Aside from R71-A74 having peeled away from the sheet, it looks pretty good. There are some breaks still to close. Hbonds for parallel sheet are C76.N-G114.O, C76.O-E116.N Missing Hbonds A74.O-G114.N A74.N-S112.O V72.O-S112.N Adding these 3 hbonds doesn't change what the low-cost decoy is. 13 Aug 2002 Kevin Karplus Best score on try7 is try7-opt-scwrl. This looks ok EXCEPT for R71-A74, which have got turned at a funny angle. 14 Aug 2002 Kevin Karplus Best score is now try8-opt, but R71-A74 is still having problems. Adding another constriant still leaves it as best score. 15 August 2002 Kevin Karplus R71-A74 STILL having problems in try9-opt-scwrl, which is new best score. I tried breaking at 55,65, and 74, moving 55-65 and 66-74, in the hopes that OptSubtree would put the pieces back together better. Fri Aug 16 11:47:36 PDT 2002 Kevin Karplus Current best score is try10-opt-scwrl, but the chunk from H69 to A74 is even worse placed than usual. According to the undertaker-try10.log, the constrainst for this chunk were # Constraint# added constraint atoms: O(562) at 0 0 0 in T0141:V 72 (71) N(885) at 0 0 0 in T0141:S 112 (111) min_dist=2.3 desired_dist=2.7 max_dist=3.2 # Constraint# added constraint atoms: N(570) at 0 0 0 in T0141:A 74 (73) O(889) at 0 0 0 in T0141:S 112 (111) min_dist=2.3 desired_dist=2.7 max_dist=3.2 # Constraint# added constraint atoms: O(573) at 0 0 0 in T0141:A 74 (73) N(899) at 0 0 0 in T0141:G 114 (113) min_dist=2.3 desired_dist=2.7 max_dist=3.2 # Constraint# added constraint atoms: CB(566) at 0 0 0 in T0141:S 73 (72) CB(695) at 0 0 0 in T0141:P 89 (88) min_dist=4 desired_dist=6 max_dist=8 # Constraint# added constraint atoms: CB(548) at 0 0 0 in T0141:R 71 (70) CB(868) at 0 0 0 in T0141:D 110 (109) min_dist=4 desired_dist=6 max_dist=8 The only H-bonds formed in try10-opt-scwrl are S73.O-G114.N A74.N-S112.O Maybe I need some constraints on the other side---CB constraints on A74 and V88 perhaps. Tweaking the parameters to make breaks worse makes try9-opt-scwrl score better than try10-opt-scwrl. Maybe I should do another quick run from try10-opt-scwrl to try to get this chunk back into place. Fri Aug 16 17:04:23 PDT 2002 Kevin Karplus The chunk has still not fallen into place. Perhaps it is time to remove constraints and reoptimize, to let it fall where it may---at least jiggling will then remove some of the bad clashes. 17 Aug 2002 Kevin Karplus try12-opt-scwrl is new best, but the chunk is still not in place. Perhaps part othe problem is that the protein is "knotted", with the N-terminus (1-25) passing through the loop of W95-N109. Maybe I should move both of those off regions into space, to give them a chance to reform without the knot. 17 Aug 2002 Kevin Karplus Foo, try13-opt managed to reknot itself. Maybe I should copy A94-R107 from some random conformation, just to get it out of this wedged state. I copied from T0141.rand.17.20.pdb to make T0141.cut-and-paste.pdb, and ran try14, which seems to be getting new best scores, so has probably unwedged itself. 18 Aug 2002 Kevin Karplus try14-opt is new best score. The strand has rejoined, though it still bends more than I'd like, and the knot has been removed. The structure is still pretty "foamy" though. Upping the break penalty and the constraint weight should help straighten the strand (if that is possible) and close up some of the gaps. Let's do try15 with CloseGap and InsertFragment high, and OptSubtree back to normal levels. 18 Aug 2002 Kevin Karplus In try15-opt, the strand is still a bit bent, so it doesn't make the hbonds I'd like to see. Missing are G114.N-A74.O S112.O-A74.N S112.N-V72.O These are in the constraint list already, so undertaker must be doing the best it can to get them. The best score right now is for try15.8.90.pdb, not try15-opt. Maybe I should do one more run, with break turned DOWN, so that the strand can be flexed without being constrained by where the loop currently is. With the break weight down, try15.9.90 scores best. 19 Aug 2002 Kevin Karplus try16-opt is new best score. It is still a bit foamy, but better on the "foaminess" costs (dry5, dry6.5, dry8, dry12, phobic_fit) than previous runs. The strand still bends awkwardly, making a bit of a hole. Let's try turning constraints way up, and make OptSubTree and InsertFragment much more probable, to try to get the strand to settle into place, even though it will have to (temporarily) make a bad break. I'll make try17 be a real short run, so that I have more time to look at the results. Mon Aug 19 14:21:50 PDT 2002 Kevin Karplus try17-opt-scwrl is new best. It has a mid-strand break that I don't like in a neighboring strand, and still has not unbent the bent strand. Breaks are pretty bad in this version. If we turn the break penalty up high (to 25), the best model is try17.2.50, which has some bad clashes. Let's try a polishing run (with breaks turned up high) on several of the intermediate runs from try17 as well as try17-opt-scwrl. Mon Aug 19 17:08:05 PDT 2002 Kevin Karplus try18-opt has screwed up big-time, introducing bad clashes. Increasing the clash weight substantially makes try17-opt-scwrl score best, and try18.3.50 next. Let's redo the optimization with this version of the score function. 20:43 try-19-opt-scwrl is new best score. It looks a bit foamy to me, and try16-opt has a better dry6.5 score. Perhaps I should turn down the clash and break penalties a little, and reoptimize from some of the best. Tue Aug 20 12:58:55 PDT 2002 Kevin Karplus try20-opt-scwrl is new best score. The stupid strand is still bent, making a loose spot, but I give up on trying to unbend it---I'll have to think about ways to do that sort of directed change more easily. try20-opt-scwrl is still foamy, but until I can straighten the strand, I don't think I can get rid of the foam. Let's just submit it and be done. As a second model, let's submit try16-opt, which has better density, but worse clashes.