Wed Jun 16 10:17:14 PDT 2004 T0205 Due 5 July 2004 Wed Jun 16 10:38:37 PDT 2004 Kevin Karplus T0205 looks like fold recognition, with 1nfjA, 1nh9A, 1h0xA, 1udvA, 1ukkA, 1ml8A, 1i50k as top hits. Actually, only 1nfjA and 1nh9A seem to be coming up really consistently, but 1h0xA does very well in the template library. Wed Jun 16 13:47:25 PDT 2004 Kevin Karplus The try1 model is not bad, though there are some disagreements with t04 predicted secondary structure. The conserved_t2k residues do cluster fairly well in the hydrophobic core. Hmm--it looks like t04 did not get many homologs for this one, so the t2k predictions are more useful. The N-terminus may need a bit of work still and we could pack a bit tighter, but otherwise this looks pretty good. The positioning of the C-terminal helix is still mostly guesswork also. Wed Jun 16 20:49:41 PDT 2004 Kevin Karplus Foo! try2 broke the sheet. I'll probably need to add the Hbond constraints for residues 22-25 from try1-opt2: Hbond (T0205)N21.O (T0205)E50.N # 2.79897 Hbond (T0205)I23.N (T0205)E50.O # 2.78206 Hbond (T0205)I23.O (T0205)S52.N # 3.03505 Hbond (T0205)V25.N (T0205)S52.O # 2.77787 Thu Jun 17 06:21:41 PDT 2004 Kevin Karplus try3 put the strand back into the sheet, but gave up the initial hairpin that try2 had formed. Maybe I need more hbonds for the initial strand. Thu Jun 17 13:09:02 PDT 2004 Kevin Karplus try4 is somewhat worse scoring than try3. The strand V75-M80 has curled up into a helix. Question: is M80-S82 a turn, or there a long strand to pair antiparallel with K97-K105? One possibility is for L51 and A53 to Hbond parallel with I100 and I102, keeping the part predicted to be most buried together. But most of our initial alignments have an antiparallel pairing for M80 and K99, extending back to E76-T103 and forward to S82-K97 or V84-V95. Let's stick with that for now. Thu Jun 17 17:45:12 PDT 2004 Kevin Karplus Try5 really looks like it wants to pair I102 with V49, I100-L51, A98-A53. Let's try that for try6. I'm rather dubious about it though, since it is not compatible with the inital alignments. So let's simultaneously try a run with hbonds similar to try3's. Fri Jun 18 17:18:27 PDT 2004 Kevin Karplus try7 is not too bad, but I don't really like the orientation of E76-I83. We may need to flip over the strand or move the bulge around. try6, with the hbonds specified for a flipped-over hairpin looks surprisingly good. It didn't quite form V49-I102, L51-I100, A53-A98, but almost. Nothing is tightly packed though---there is a lot of space between the helices and the strands. I don't think this one is worth pursuing further---it would be too hard to get decent density. Wed Jun 23 21:01:24 PDT 2004 Kevin Karplus We may want to do a polishing run on try7 -- use Rosetta to repack residues for the top 4 or 5 models (at least try7-opt2, try3-opt2, try1-opt2, and try8-opt2), then do a run that has the TryAllAlign commented out and has ReadConformPDB to get initial conformations. The cost function should probably be a copy of the unconstrained.costfcn. I think that trying to fuss with the beta strand alignment at this point is unlikely to result in substantial improvement. Fri Jun 25 12:22:19 PDT 2004 Kevin Karplus The DNA-binding proteins this is homologous to are dimeric. We might want to try superimposing out best model so far on the dimer and seeing whether there are bad clashes either in the dimer interface or with the DNA-binding pocket. Fri Jun 25 13:08:12 PDT 2004 Kevin Karplus I did this with the make-dimer-try7.under script using 1h0x as the homolog. The dimer alignment is based on the t04 str2+CB_burial_14_7 alignemnt. I used 1h0xA, rather than 1nfjA, since 1nfjA is a monomeric model. Observation: the C-terminal helix is badly placed interfering with both the dimerization interface AND the DNA-binding site. (The DNA binding site is the saddle-shaped beta sheet, with K19 and K20 sticking into the minor groove.) Maybe I need to constraint F110 to something near L104, L113 near L102, V128 very roughly near A57? Fri Jun 25 17:47:03 PDT 2004 Kevin Karplus try9 did tuck the big helix out of the way, though not quite where I wanted it, but it let the small N terminal helix drift in. Perhaps I should add salt bridges for E2 and E3 to K29 to hold the N terminal helix out of the way? Maybe a salt bridge for E119 to K29 would help hold the Cterminal helix better, and moving V128 close to M56. Actually, I like the close approach of E76.OE2 and K121.NZ, so I think I'll try to enforce that instead. Sat Jun 26 07:00:58 PDT 2004 Kevin Karplus try10-opt2 looks pretty good, though it doesn't score much better than try9-opt2, even with the try11.costfcn, which was tweaked to make try10-opt2 look good. try9-opt2 gets more hbonds, but does not satisfy the constraints as well. With unconstrained.costfcn, the best scorers are try9, try7, try3 and try10. Repacking try10-opt2 does not hurt it much, so it may be time to do a polishing run with the try11.costfcn starting from previous models. I wonder if I should try repacking the dimer---that might help get E66, K69, and K78 into better rotamers, since they are hiding behind the other half of the dimer. Unfortunately, undertaker can't handle dimers yet, so I'd have to go with rosetta repacking. Let's do the polishing run first. Sat Jun 26 10:01:00 PDT 2004 Kevin Karplus Try11-opt2 scores almost as well try9-opt2 with the unconstrained cost function, and tucks the helices nicely out of the way. The main remaining problem is that some of the charged residues get buried in the dimerization interface. Try11-opt2.repack-nonPC does not score nearly as well, but is the best-scoring of the repacked models. Rosetta likes it best, though there are still some clashes it doesn't like. We could do another polishing run, with soft_clashes and break turned up. Maybe this is reaching the point of diminishing returns though---perhaps we should just declare T0205 done and move on? One more iteration won't hurt ... Sat Jun 26 14:29:49 PDT 2004 Kevin Karplus try12-opt scores better than try11-opt, and Rosetta likes the repacked version better also. I think this model is basically done, unless we want to try a Rosetta repack of the dimer. From bbarnes@ucsc.edu Sat Jul 3 17:53:31 2004 Date: Sat, 3 Jul 2004 17:53:12 -0700 (PDT) From: Bret D Barnes To: Kevin Karplus cc: sol@soe.ucsc.edu, ggshack@soe.ucsc.edu, learithe@soe.ucsc.edu, martina@soe.ucsc.edu, marcias@ucsc.edu, rph@soe.ucsc.edu Subject: T0205: Repacking Dimers with Rosetta Kevin, I wanted you to look at the repacked dimer for T0205. Its in the decoys directory called T0205.dimer-1h0x-try12.repack.pdb. The original dimer is also there (T0205.dimer-1h0x-try12.pdb). Let me know if that is what you wanted, or if there are any problems. -Bret -------------------------------------------------- Sat Jul 3 18:42:22 PDT 2004 Kevin Karplus The repacked dimer does look better than the un-repacked one. The lysine K78 that was tucked away behind is now making some good Hbonds. I'll take the first chain of the repacked dimer and make it our number 1 model. Hmm--I may have to retract that statement. I'm now seeing very little difference. I'll look at it again after supper. Sat Jul 3 20:08:00 PDT 2004 Kevin Karplus The Hbonds are K78.NZ:A to T81.O:B and K78.NZ:A to T81.OG1:A (the T81 in each chain). The mutual-information measures would not pick up this relationship, because we did not check such a small chain separation. There IS a different rotamer for K78 in the repacked version, chosen to make the pair of Hbonds. I'm submitting the 5 models: 1 T0205.chainA-from-repacked-dimer12.pdb 2 T0205.try12-opt2.pdb 3 T0205.t2k.undertaker-align.pdb model 1 T0205-1nfjA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m 4 T0205.t2k.undertaker-align.pdb model 2 T0205-1h0xA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m 5 T0205.t2k.undertaker-align.pdb model 3 T0205-1nh9A-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m Wed Sep 22 11:12:12 PDT 2004 Kevin Karplus Using the GDT evaluation, the order is robetta2 76.3441% try1-opt1-scwrl 72.3118% our best model model5 64.2473 model4 61.8280 model3 60.7527 robetta1 60.7527 model2 52.4194 model1 50.5376 robetta3 45.9677 robetta4 45.4301 robetta5 43.0108 I should have submitted the full auto prediction: try1-opt2 68.8172 Robetta beat us on this one (either comparing best of 5 or the model1 predictions), but we did not submit our best models. Fri Sep 24 17:03:28 PDT 2004 Kevin Karplus Looking at the GDT curves and the smooth_GDT scores, it is clear that robetta had 2 models that were good, and we had 3, but the robetta models did beat us and we messed up good alignments: name length missing_atoms rmsd rmsd_ca GDT smooth_GDT robetta-model2.pdb.gz 130 0.0000 6.4407 6.5057 -76.6129 -71.4679 T0205.try1-opt1-scwrl 130 0.0000 4.3055 3.1306 -71.5054 -68.0529 T0205.try1-opt2.pdb.gz 130 0.0000 6.9888 6.7410 -67.4731 -64.7141 full auto model5.ts-submitted 130 449 3.2723 2.6127 -61.8280 -58.9488 align model4.ts-submitted 130 379 3.7769 3.1154 -61.2903 -57.8946 align robetta-model1.pdb.gz 130 0.0000 5.0410 3.5719 -61.2903 -57.1057 model3.ts-submitted 130 388 4.2015 3.1087 -59.1398 -55.9924 align model1.ts-submitted 130 0.0000 6.4006 5.3173 -50.2688 -47.8306 model2.ts-submitted 130 0.0000 6.4211 5.3173 -50.8065 -47.8200 robetta-model3.pdb.gz 130 0.0000 7.6659 6.8175 -44.8925 -42.2957 robetta-model4.pdb.gz 130 0.0000 8.0924 7.3510 -43.5484 -41.3109 robetta-model5.pdb.gz 130 0.0000 7.6484 7.1444 -43.2796 -40.8375