Fri Aug 16 16:03:44 PDT 2002 t0191 Homology model 1gpjA Aug 26, 2002 Oscar Hur Saw hole in the structure in try1. Run more iteration to see if the structure improves. Aug 28, 2002 Oscar Hur Increase JiggleSubtree and OptSubtee and ReduceClash Read in try1, try2, and try3 Run try4 Aug 29, 2002, Thu Oscar Hur This structure has similar problems as with t0192. There are many breaks in the structure. The breaks are in the pairs of amino acid residues of: 28-29, 85-86, 101-102, 147-148, 174-175 I increase the values to: CloseGap 10, OptSubtree 2 JiggleSubtree 2 And Run try5. Aug 30, 2002 Fri Oscar Hur Have difficulties with breaks. Breaks still exist between: 30-31, 85-86, 101-102, 147-148, 219-220 Increase the value: InsertFragment to 30 InsertSpecificFragment to 25 Run try6 1 Sept 2002 Kevin Karplus 18:07 I increased the burial weights and the break weight in define-score.script, and rescored the existing models. Best is now try6-opt. This has more serious problems than just a few breaks---at least three of the strands are not attached to the sheet. I'll turn pred_alpha2 weight way up (from 1 to 4), to see if there are any models that have these pieces straight, even if not attached. If not, we may want to do another run from the alignments to try to pick out some better pieces. Even with pred_alpha2 way up, the try6-opt model is preferred. Let's do another run with no initial conformations, to see if we can get the strands to appear. If not, we'll probalby have to start adding constraints. 2 Sept 2002 15:00 Although try6-opt scores better than try7-opt-scwrl, try7-opt-scwrl has not had much polishing, so its score is quite good. Unfortunately, it does not look so hot---there are 5 good strands in a sheet, and the rest are pretty scattered. Try6-opt got 5 1/2 strands, so try7 is not much better than try6 (fewer strands wound into helices). In try6, the strands we have in the sheet are a nice alpha-beta alternation, with the parallel strands in the order 321456 (count E175-K178 as the third strand at the edge of the sheet). 1gpjA has such a sheet as one domain, so it looks like we've got multiple-domain problems again. The domain looks like it is about G107-I245 (or I261, if a final helix has just been pried off the sheet). We should do subdomains t0191-1-110 t0191-101-260 t0191-241-282 16:56 Kevin Karplus The alignments and fragment build for the subdomains just finished. Let's try run try8, with the larger set of alignments to work from. Wed Sep 4 12:16:53 PDT 2002 Kevin Karplus Although oscar did not write anything in the README file, he said at today's meeting that he though T0191 was basically done. Current best score is try6-opt, which has the N-terminal strands messed up. Try7 and try8 don't fix the problems. I think we need to add constraints. s1 V8-I12 s2 Y36-D42 s3 G62-T66 s4 V90-I95 s5 K99-I101 s6 N125-G130 ... from s6 on, we have a fine sheet. We have an alternation s1 helix s2 helix s3 helix helix s4 turn s5 helix s6 helix s7... From this I guess orientation s1 > aktKVIGLIghp s2 > glnYVYVAFPVlp s3 > lgivGFNVTIph s4 > gaVNTIKIed s5 < nyGIAKg s6 < The existing sheet is all parallel 11^10^9^6^7^8 Let's at least try to assemble s1^s2^s3^ and the s4-s5 hairpin. 4 > aktKVIGLIghp 33 > glnYVYVAFDVlp 58 > lgivGFNVTIph 88 > gaVNTIKIed 104< nyGIAKg With these try9.constraints, the best score is for T0191.try8-al6+1qnf-T0191-local-adpstyle5.pw.a2m.gz:1qnf.0.40.pdb which has started to form the strand2-3 pairing, and looks like it could get the strand1-2 pairing easily. There is even an hbond V37.O, T66.N, which indicates a different register (V37-V65). I don't like that alignment as well, since it winds the strongly predicted turn at L58,G59 up into a longer helix. Let's do another optimization, with these constraints, from T0191.try8-al6+1qnf-T0191-local-adpstyle5.pw.a2m.gz:1qnf.0.40.pdb T0191.try8-al6 T0191.try8-al4 T0191.try8-al2 T0191.try7.15.60.pdb T0191.try7-al2.pdb T0191.try3.8.60.pdb Wed Sep 4 18:01:10 PDT 2002 Kevin Karplus try9 is still pretty messy---let's turn break penalties down and reoptimize to see if stuff can pack better. Then we'll have to look for what to stuff where and add constraints to squeeze things into place. 21:13 try10 is about 5/6 done, but it seems to be relying too heavily on OptSubtree. On the next round I'll have to turn that down. 4 Sept 2002 Kevin Karplus 23:05 try10-opt is doing fine on the middle of the protein, but the domain 1-110 is messed up, and the C-terminal helices have gotten detached from the sheet. I'm not sure how to fix the helices, but I can clean up the N-terminal domain by shifting the first strand (I think), and by adding length constraints on the helices and the parallel strands. I'll also try putting the hairpin parallel to the first strand // 58 > lgivGFNVTIph // 33 > glnYVYVAFDVlp // 4 > aktKVIGLIghp // 88 > gaVNTIKIed // 104< nyGIAKg 5 Sept 2002 Kevin Karplus 17:09 try11 is running very long optimizations, but does not seem to be accomplishing anything. It is relying heavily on JiggleSubtree, which makes moves too small to fix the problems of N-terminus. Fri Sep 6 12:38:29 PDT 2002 Kevin Karplus try11-opt is new best score (with try11.constraints), but all the problems of try10 still seem to be there. Let's try assembling the N-terminal domain by itself in directory t0191-1-110. That should run faster, and we can cut-and-paste it back in later. I'll start two jobs there---one trying all alignments (try1), the other starting from random starts (try2). Both will have the constraints I was trying to use in try11 on the whole domain. Perhaps I should try assembling the second domain independently also? It is looking pretty good right now, except for the drifting of the Cterminal helices. Perhaps I should just do a copy to start from? There's not much point in doing it for 101-260, since the extra helices come AFTER that. Maybe there is some point, since the first and last helix of the domain are drifting out of place. When pasting back together, the best cutpoints are probably after G102 and near F259 (depending where the helix starts getting frayed). I the current try11-opt, the continuing helix starts at about V257, so it would be nice if the helix continued up to there. When pasting back together, we may want to add some helix length constraints that cross the cut, so that the helix can be reassembled quickly. The 101-260 try1 is making score improvements over what was provided to it, though (predictably) the insertion of all alignments did not make any improvements. I think that the alignment we had is basically right, and we just have to fold in the C-terminal helix. Fri Sep 6 13:59:54 PDT 2002 For 1-110, try1 and try2 have come up with somewhat different models, with trh try1 model scoring much better (probably because of longer polishing). I'll run try3 from both, but it will probably favor try1, so I may want to do another run from just try2 (this is try4). I may need to modify constraints to pull the strands together in the try2 attempt---I'll check that when tries 3 and 4 are done. Sep 7 11:14 Kevin Karplus For 1-110, the try3-opt model scores better than the try4-opt, both with the try1 constraints and with no constraints. Try3-opt makes a 5-strand sheet 3^2^1^4^5v. It looks pretty good, but the helix P45-L58 has a couple of bad breaks. try4-opt has 2^1^3^4^ with strand 5 at right angles. It looks like it could be fixed, but try3 may be somewhat better. We may want to do two cut-and-paste models when we are done. -------------------- For 101-260, the final helix did not fold against the sheet---probably in part because of K243, which would then be buried. I suspect that K243 is off by one---It should be next to D223 or V221, not M222. Let's try 217> edmVVMDLiy 240> vnaKTINgl and run try2 from scratch (which may have troubles, since it has not had the polishing of the piece imported from the full protein). -------------------- 13:01 For 1-110, I made up new constraint sets (try5 and try6) corresponding to the structures in try3-opt and try4-opt, and am doing separate optimizations of each style. -------------------- 13:33 For 101-260, try2-opt-scwrl does not score quite as well as try1-opt, because the pred_alpha2 score is not quite as good, though almost everything else is better. Unfortunately, it got the improvement, not by flipping over the last strand and packing the helix tighter, but by breaking off the end of the last helix and unwinding it. Perhaps I should submit try1-opt to Vast (job VS31325 password casp5t0191) and use Cn3D to hand edit the alignment. That way I could get a decent alignment without too much manual effort. ---------------------------------------- 16:00 For 1-110, the try6-opt arrangement looks better than the try5-opt arrangement, and scores better with no constraints. Should I cut-and-paste the final hairpin (A89-A110) from try5 into try6 and reoptimize? (why not? try7). For 101-260, the vast run has finished and clearly picked out 1gpjA -------------------------------------------------- 17:45 For 1-110, try7-opt is clearly worse than try6-opt. Let's polish try6-opt once more, then paste it together with whatever we get from 110-260 and the final helices. 18:13 For 101-260, the 1gpjA alignment provides essentially the same result in try3 as we had in try2 and try1. The 1gpjA helix IS unfolded from the sheet---it is the beginning of a a very long helix. try3-opt scores better than try1-opt is most parameters, but not pred_alpha2. Turning down pred_alpha2 moves try3-opt to the top. I think this domain is ready for cut-and-paste. 7 Sept 2002 20:59 Kevin Karplus try12-opt put the domains together without disrupting them too much, but I don't think it did it quite right---I'd like helix Y103-E116 to stay in place, with L114.CD2 touching I191.CG1 and CG2 and A110.CB touching L142.CD1 We'd also like the hydrophobic patches to approach each other---maybe we can add some weak constraints to encourage that. Trying something along these lines in try13. We should probably take the hydrophobic packing constraints back out after getting the domains close, so that the details of the packing guide the docking. 23:49 try13 did not pack the domains well, leaving a large gap between G109 and A110, and having the C-terminal helices stick out into the solvent, plus yanking the strand V240-I245 off its sheet. The hydrophobic patches I was trying to get close to each other are on opposite sides of the protein, both exposed. 8 Sept 2002 08:47 Kevin Karplus try15-opt doesn't have large gaps, but also didn't pack the hydrophobic faces the way I wanted. There aren't many buried charges though, so this packing may be ok. For try16, I'll remove the hydrophobic patch constraints and turn the constraints down and dryness up to pack better. Then I'll submit. 12:35 try16 works ok, but I'll try adding some new hydrophobic constraints to try to tilt the docking a bit. I'll run for a while, but definitely submit tonight.