Tue Jul 16 10:27:25 PDT 2002 t0155 Pretty obvious homology modeling target: 1dhn and 1b9lA as models--- both are in the t2k alignment. The alignments to 1dhn are gapless. The try1-opt structure looks good, except for the predicted strand on the C terminus, which is not near the sheet (though it is straight), and a bit of curling of the N-terminal strand. Both problems could probably be fixed with constraints. We have an Hbond O I5 to N V35, so extending back two hbonds would be: N I5 O V35 O D3 N I37 At the C-terminus the penultimate strand is 112:AVVIRRS and the ultimate strand is 124:GWVV. I'd be tempted to pair this V113 V127 V114 V126 I115 W125 and V114 is the residue with the exposed N and O for hydrogen bonding, so let's try V114 N V126 O V114 O V126 N Try2 looks much better and almost has the strand in position. It seems to be getting hung up on R116 though, so I'll try SCWRLing the conformation, reoptimizing, and reSCWRLing, to see if that will give it the necessary freedom to repack. Try3-opt scores a little better than try3-scwrl-opt, but both score better than try2-opt and try2-opt-scwrl. The final strand has not hbonded properly with the others yet. The H bonds should be for R116, V114, and A112. Let's try R116 with R123, V114 with W125, and skip A112. 18 July 2002 Kevin Karplus I was looking at the homolog 1b9lA, and see that it is part of a tetramer, with the 4-strand sheets lining up to for a barrel. This means that there should be nothing on the "inside" face of the sheet, and we have to decide whether the barrel is made up of 4-strand sheets (as in 1b9lA) or of 5-strand sheets. If 5-strand, then the entire firth strand has to be there, to make a continuous barrel. try5-opt-scwrl has the beginnings of a fifth strand, but the C-terminus winds up into a helix on the wrong side of the sheet. I wonder how we could use constraints (or other mechanisms) to keep the C-terminus on the correct side of the sheet (if we decide to go with 4 strands instead of 5). QUESTION: 4 strands or 5? 3 August 2002 Kevin Karplus Let's try a very weak constraint putting G130 6Ang from A78, to get the C-terminus on the right side of the sheet. We can also fix the constraints for the 5th strand to match what we see in try5-opt and modernize the rest of the score function by including phobic_fit, pred_alpha2, and dry5. try5-opt-scwrl still scores well, but try5-al2.12.20.pdb is the best scorer. Let's do another iteration from both of them. 4 August 2002 Kevin Karplus try6-opt is new best, with try6-opt-scwrl close behind. Strand 4 makes nice hairpin to strand 5, but has come unstuck from strand 3 and drifted over a bit. I need to add constraints hbonding V93-R117, V95-I115, V97-V113, K99-V111 With the new constraints and some tweaking of the weights for the cost functions, the new best is try6.9.80. Let's adjust the constraint on the C-terminus to put L72.CD1 near V133.CG2. Best is still try6.9.80. Let's try another run from the best few. 5 August 2002 Kevin Karplus New best is try7-opt. It's not perfectly packed, but it looks pretty good. Oops--I forgot that if we have the fifth strand we should have the ENTIRE strand for good dimerization. Adding new constraints for the full C-terminal strand still scores try7-opt best, so let's do another run from there. 8 Aug 2002 Kevin Karplus In try8-opt, the C-terminal strand is still not quite straight, and the strands seem a little too far apart (one can see through the space-filling model). Let's add a constraint to straighten the strand and reduce the desired spacing from 5.12 to 5 and re-optimize. 9 aug 2002 Kevin Karplus I don't see much change from try8-opt to try9-opt. The best new score is actually for try9.16.100. Unless I can think of an improvement to make in the score function, we seem to have reached the point of diminishing returns on this target. 12 Aug 2002 Kevin Karplus Reran make with new template library---no changes to top hits. 12 Aug 2002 Kevin Karplus Tried turning off constraints in score function to see which model looks best. Even with constraints off try9.16.100 looks best. Let's try an optimization run without constraints starting from the best few models. Note: the constraints will still be present (to guide OptSubtree) but will not be included in the score function. 13 August 2002 Kevin Karplus New best score is try10-opt. packing looks pretty good, but Cterminal strand has curled up a bit. let's put back c-terminal constraints. With the c-terminal constraints, try10-try9-12.9.80 scores best. Let's do a new run from try10-opt and some of the good scoring ones with the constraints on. 14 August 2002 Kevin Karplus Best score is now try11.0.90 (not try11-opt). In fact, 19 of the intermediate files for try11 score better than try11-opt! This is probably the result of restoring the breaks poorly when re-reading the .pdb files. Perhaps I should do another run, with constraints turned way down in score function, and CloseGap made more probable. Increasing the break penalty and the dry6.5 and dry8 weights moved try11.11.90 to the top 15 Aug 2002 Kevin Karplus Try12-opt is slightly better score than try11 runs in internal scoring, but when output and read in again it does poorly. Best score is now try12.14.90. I'll do one more short run, tightening the CB constraints for the anti-parallel strand on the end. 15 Aug 2002 Kevin Karplus try13-opt is now the best, but it does not seem to be a big change from try12. Further runs seem likely to get miniscule returns, at least until we think up a way to make the gap closing better. I will submit try13-opt.