Wed Jul 17 12:49:18 PDT 2002 t0157 Thur 18 July 2002 Kevin Karplus The top-hit (1hjrA) is not a very strong score (E-value 4.6e-02 overall, and 4.19e-04 for the str target model) and the next few hits are different SCOP families, but the structure in T0157.try1-al12.pdb looks pretty good. One of the helices gets broken in try1-opt, but that can probably be fixed by adding a constraint from D41 to E51. The try2 run failed with a segmentation fault after about 50 minutes. I'll try running it (with the same random number seed) under gdb, to see if I can find the problem. I think the problem is fixed, but I accidentally ran try2 twice simultaneously, which probably screwed up the record keeping. The try2-opt run has exploded the beta sheet, but has all the helices intact. Sigh--it looks like I'll need a lot of constraints to keep things together! I can probably pick a few of the hbonds from try1-opt and use them to hold things together. 12 August 2002 Kevin Karplus Reran make with new template library. No change to top hits, but some small changes in the weak hits. Modernized define-score.script and undertaker.script and reran, with only the one constraint added after try1. If this blows up, I'll have to add hbond constraints. 18 August 2002 Kevin Karplus The top score is now try3-opt-scwrl. It is fairly compact and has not blown up. I think we can do just one more run to tune things up and it'll be ready to submit. Or, should I try all alignments for top several templates? Maybe I should---I'll do try4 from a larger set of alignments, then do a polishing run from try3-opt and try4-opt. 18 Aug 2002 Kevin Karplus try4-opt is new best score, but try3 and try4 are QUITE different structures! (The difference in score is tiny). try3-opt-scwrl is a beta sandwich, with 2.5 strand in each slice (one strand runs diagonally and is in both slices. try4-opt has a single beta sheet of 5 strands, though the last two strands are a little out of position and are not making hbonds. The top-scoring hit with fold-recognition is 1hjrA, which matches the 5-strand single-sheet prediction, as does our 4th hit (again c.55.3). The two hits in between are too fragmentary to support a fold decision. This means that I'll believe the try4-opt fold more strongly than the try3-opt-scwrl one. In the T0157-1kcfA alignment, we have a fairly complete-looking fold, so I'll borrow Hbond constraints from it: anti-parallel L5-G20 A7-A18 D9-G16 I15-Q36 V17-K34 V19-A32 parallel L6.N-E56.O L6.O-I58.N F8.N-I58.O F8.O-G60.N I57.N-V91.O I57.O-L93.N V59.N-L93.O With these constraints (in try5.constraints) the best score is for try4-al10.16.80, which has already begun to lose the Hbonds, so let's run again from the alignments with this score function. Mon Aug 19 13:28:44 PDT 2002 Kevin Karplus best score now try5-opt, which looks pretty good. Let's turn up the packing parameters (and turn down breaks) to do some polishing. 21:51 try6-opt-scwrl now best, and looks pretty good. Let'ts try tweaking parameters, and seeing if it remains best. Increasing break and clash penalties doesn't budge it. I'll submit try6-opt-scwrl as first model, and try3-opt-scwrl as second model (as a backup, in case the 5-strand sheet is wrong).