Tue Aug 6 14:30:39 PDT 2002 t0168 This looks like a fold-recognition problem, with e.3.1.1 as the most likely fold. The top-scoring sequence (1i2sA) is too new to be in our copy of fssp, but it appears in the on-line version with representative 1btl. Since all our top hits are to homologs of 1btl, we can have some confidence that we've got the right fold. 7 Aug 2002 Kevin Karplus T0168.try1-opt-scwrl does not look too bad, exept for the helical pieces 130-179, which are all broken up. We should just run for longer, with fragment insertion turned up, to see if we can clean up that area. We may want to look at some of the close CYS pairs, to see if we want to make them disulphides. T0168 is said to be glutaminase, and the only glutaminase I find in PDB is 4pgaA and its homologs. The 1btl (beta lacatamase) fold does seme to have other functions (like d-alanyl-d-alanine transpeptidase and peptidoglycan sythesis). Thu Aug 22 14:27:28 PDT 2002 Kevin Karplus I added 4pgaA as an extra template in the Makefile and will rerun undertaker from scratch including the 4pgaA alignments in the initial TryAllAlign run. Fri Aug 23 13:34:04 PDT 2002 try2-opt-scwrl is new best, beating try3-opt-scwrl (which beats all the try1---a fairer comparison, since try2 was polished from try1). Try3-opt-scwrl has a few bad breaks---the region S256-G274 looks badly displaced. try2-opt-scwrl also has trouble with this region, having created a knot. If we increase the break penalty (from 7 to 15), try2.17.80 scores best, but it has the region A260-G270 badly placed, so that if the gaps could close a knot would be formed. Let's try submitting try2.17.80 to VAST, and see if it can clean up our alignment: job VS30770 password casp5t0168 Fri Aug 23 17:06:13 PDT 2002 Best VAST hits are for 1mfoA (1.9RMSD over 204 residues) or 1dy6A (1.6RMSD over 196 residues) or 1bul(1.8 over 203) (all of these are 1btl homologs). I grabbed the 1mfoA, 1dy6A, and 1bul alignments and tweaked the gaps a bit by hand. Now let's do another run with just these alignments for TryAllAlign. 24 Aug 2002 Kevin Karplus try2.17.80 still scores best---quite a bit better than the try3-opt-scwrl results. Both still have some trouble with the placement of the c-terminal helix--it interferes with K43-Q48 in try3-opt-scwrl, and with A260-S269 in try2.17.80. Try4-opt scores slightly worse than try3-opt, avoiding the conflicts by blowing up the beta sheet. Maybe we should try optimizing from both try2.17.80 and try3-opt-scwrl, trying to get the good parts of each. 24 Aug 2002 23:13 Damn, I forgot to include "use_all" in the OptConform command, so all the optimization in try5 was done from scratch! 25 Aug 2002 Kevin Karplus Best score now try6-try2-17.8.60, with try6-opt-scwrl close behind. The Cterminal helix is still knotted, and a lot of helices in the middle are rather disordered. There is a strand in the 1mfoA alignments that is missing from the latest model. Let's add constraints for it (try7.constraints) and try reoptimizing, including TryAllAlign. antiparallel: Y259 A261 V263 V265 V279 A277 I275 G273 With this constraint added, try5-opt-scwrl (which was optimized from scratch) is now best. Unfortunately, try5-opt spoiled the other side of the strand G273-V279, so let's add those constraints also to try7.constraints. Now the best score is for try5.2.60. I wonder about the possiblity of a disulphide at C52-C293. They're pretty close on the same side of the sheet, but I'd probably have to slide one of the strands over by 2 to make them line up. Hmm sliding Q290-I302 up 2 looks like it would make the ends easier to close also. I've made that shift in T0168-moved7.pdb (based on try5.2.20), and that now gets the best score with the try7ss.constraints. Let's reoptimize from there, with fairly high CloseGaps. 25 Aug 2002 Kevin Karplus try7-opt is new best, but it seems to be having trouble closing the loops. It is trying to form a knot with P280 on the wrong side of A255. Let's try moving P280-F289 out of the way and reoptimizing with to repack. I suspect that C250 and C205 also form a disulphide, so let's move V200-A206 so that the disulphides are free to find each other with OptSubtree. 22:42 New best is try8-opt, but it is trying to knot things again, near the putativ C205-C250 disulphide. I'll have to figure out how to unknot things there. We may need to have a score function that detects attempted knotting---it is one of the major risks of allowing the backbone to be broken and reformed. Too tired tonight---I'll try looking at this in the morning, if I have time. 27 Aug 2002 11:08 Let's try sliding V200-V279 two residues along the sheet, changing the register of the strand interface between G270-L278 and S291-G299 27 Aug 2002 19:55 try9-opt-scwrl is new best score. It seems pretty good except for the bad break befor Y259. Let's optimize from here with lots of insertfragment and reduce break. 29 Aug 2002 11:40 try10-opt-scwrl looks good in the core, but the loops at K237-K258, C250-S282, and L190-V200, are all messed up, no closing and trying to knot (try9-opt had toruble here also). The helix and strand regions for T173-L180 are rather broken. The last run did a lot of insert fragment and not much optsubtree---perhaps both are needed. I should probably chop up P281-Q290, moving E286-Q290 down to fit with S191, and breaking into P281-A283, R284-R285, to allow this loop to reform in a better shape and place. Without constraints, the best model is try6-try2-17.8.60. It looks better to me than many of the constrained models. Maybe I should go back to try7.constraints, and not try to slide the strand to form the disulphide. With try7.constraints, the best is try5.2.60. Let's start a new run with the try7 constraints. 29 Aug 2002 Kevin Karplus 16:04 try11-opt scores best with the try7 constraints. It looks ok, except for I229-S256. There should be a hairpin 223> hdgYHPir 236< piVQEh and G256-F262 should be a helix. Perhaps the strand we are seeing should be shortened. It is one that we put in from the alignment with 1mfoA. I removed the strand constraints for that helix and added some length constraints to help it coil (try12.constraints). try11-opt-scwrl is still the favorite. Adding some CB constraints for the tiny hairpin (still try12.constraints) leaves try11-opt-scwrl at the top. Let's reoptimize it. 29 Aug 2002 Kevin Karplus try12-opt is new best score. I think there should be 2 SSbonds: C52-C293 C85-C205 Let's try adding them and seeing what happens. try12-opt still scores best. Fri Aug 30 09:18:44 PDT 2002 try13-opt scores best. The disulphides have both formed (well, C52-C293 needs a little rotamer work) and the backbone is not terribly tweaked. Let's do a polishing run with the constraints turned down somewhat to finish it off. Fri Aug 30 15:52:34 PDT 2002 try14-opt looks a bit foamy around A243, but no more so than try13-opt. try14.0.120 scores slightly better than try14-opt. If I turn down wet6.5 (which may be interfering with tight packing) and turn off constraints and ssbonds, the best-scoring model is try14.0.120 again. If I turn ssbonds on, but not any constraints, try14.0.120 again scores best. Let's do one more polishing run on try14.0.120 without constraints (but with ssbonds), with the wet6.5 weight down. 30 August 2002 Kevin Karplus 22:17 Best score now try15-opt, but it broke a strand at C52-V53 to make one of the disulphides. When the SSbonds are taken out, the best score is still try15-opt. I doubt we'll improve it much, so let's submit it.