Thu Aug 8 21:53:16 PDT 2002 t0176 9 Aug 2002 Clear homology to 1jrmA. 10 Aug 2002 It seems that try1-opt has blown up a beta sheet. One problem might be that 1jrmA is not in the template library---I'll add it and rerun the scoring, so that we cna get a better set of alignments. I'll also add some Hbonds from the T0176-1jrmA pdb file: A7.O R18.N (not A7.N though) T9 V16 N11 G14 D13.O N87.N (not D13.N though) L15 I85 L17 I83 L19 K81 I21.N R79.O Q69 I86 V71 K84 E73 Q82 S29.O A41.N V31.N K39.O Adding these Hbonds still leaves try1-opt-scwrl as the lowest cost, but increasing the weight of the constraints makes try1.15.60 the best. I think we should re-run without a seed conformation, just using the alignments. Perhaps this should wait until 1jrmA has been added to the template set. I'm adding a bunch of new templates from FSSP (I gave up on waiting for them to update the FTP site, and parse the ids from the web site instead), including 1jrmA. 24 Aug 2002 Kevin Karplus Updated the score function to make dry6.5, dry8, and break more important. Modified undertaker.script to try only the 1jrmA alignments (including the heavy-str one I just added), but use the other alignments in the specific fragment library. With the adjusted score function, try1-opt-scwrl comes out best. Let's do another run with the changed initial conditions. 25 Aug 2002 Kevin Karplus 16:26 try2-opt-scwrl is new best score. Now the beta sheet is staying together but there is a bad mid-strand break that needs to be fixed between V70 and V71. The V71:K84 hbonds look ok to me so we want to add Q69:I86 hbonds. Hmm---those are already in try2.constraints, so let's just increase constraint weight, leaving try2-opt-scwrl as top score. Fri Aug 30 16:37:43 PDT 2002 try3-opt is new best with try2.constraints and without. The midstrand break is fairly small now. Let's polish. 30 Aug 2002 22:33 Kevin Karplus The best is now try4-opt. With breaks turned way up, try4-opt is still the best. It looks like it will be difficult to heal this break, but I'm willig to make one more run with CloseGap turned up and JiggleSubtree turned down. Since I can't see any way to unbury K24, let's have it make a salt-bridge with D28. I've also added the "nonalpha-hydrogen-bonds" scoring function from Jonathan's version of undertaker, to see whether it works. With the modified score function, try4.8.160 scores best. The K24.NZ comes the right distance from D28.OD2, but K24.CD and K24.CE clash with D28.CG and D28.OD1. Maybe the salt bridge should put K24.NZ equidistant from OD1 and OD2. Should there be a water distance in between? how big is that? I'll make the spacing an extra Anstrom, and hope that's enough. 31 Aug 2002 Kevin Karplus 00:16 The try5 run seems to be doing ok with the Hbond scoring, but it never appears to change the Hbonds very much, and the first iteration seems to have loosened the packing to reduce the clashes. I'm doing another (try6) run that will not bother with the Hbonds and will try to remove breaks without such a large clash penalty. 31 Aug 2002 Kevin Karplus With the try6 score function, the best score is for try6.4.200, which still has small breaks. The differences between all the models now is quite tiny, so let's submit try6.4.200 22:19 oops, that was scored with buggy code. With the old version of undertaker, the best is try6-opt. Let's resubmit that one instead.