Mon Aug 16 17:24:22 PDT 2004 T0282 Mon Aug 16 20:46:05 PDT 2004 Kevin Karplus This looks like a comparative model with 1gq6A (and other c.42.1.1 family) as top templates. Tue Aug 17 16:12:00 PDT 2004 Kevin Karplus The try1-opt2 model looks pretty good. There is a little looseness that could probably be fixed in a polishing run. The t04 alignment does a better job of highlighting the conserved D157-H159-D161 motif, but the overall conservation patterns are similar. Secondary structure predictions are very similar, so I'll use the t04 ones. I'll do a constrained polishing run for try2, then an unconstrained one for try3. Tue Aug 17 22:22:36 PDT 2004 Kevin Karplus I started the unconstrained polising run (try3), but if it does not succeed in closing the bad breaks T0282.try2-opt2.pdb.gz breaks before (T0282)L293 with cost 10.9728 T0282.try2-opt2.pdb.gz breaks before (T0282)G152 with cost 6.26168 T0282.try2-opt2.pdb.gz breaks before (T0282)I151 with cost 5.61004 T0282.try2-opt2.pdb.gz breaks before (T0282)D245 with cost 2.81212 T0282.try2-opt2.pdb.gz breaks before (T0282)G55 with cost 2.43221 T0282.try2-opt2.pdb.gz breaks before (T0282)A54 with cost 2.12052 T0282.try2-opt2.pdb.gz breaks before (T0282)D308 with cost 1.85245 T0282.try2-opt2.pdb.gz breaks before (T0282)I85 with cost 1.02705 then I'll probalby have to try again with constraints at the breaks, as I'm trying for T0279. Wed Aug 18 13:58:06 PDT 2004 Kevin Karplus The bad breaks are still there: T0282.try3-opt2 breaks before (T0282)L293 with cost 10.9661 T0282.try3-opt2 breaks before (T0282)G152 with cost 5.59607 T0282.try3-opt2 breaks before (T0282)I151 with cost 4.9281 T0282.try3-opt2 breaks before (T0282)D245 with cost 2.31869 T0282.try3-opt2 breaks before (T0282)G55 with cost 2.00121 T0282.try3-opt2 breaks before (T0282)A54 with cost 1.86006 T0282.try3-opt2 breaks before (T0282)G126 with cost 0.932277 T0282.try3-opt2 breaks before (T0282)G273 with cost 0.727452 T0282.try3-opt2 breaks before (T0282)D308 with cost 0.706295 The constraints worked well for T0279, so I'll try them here also. Thu Aug 19 15:06:07 PDT 2004 Kevin Karplus Still 3 bad breaks: T0282.try4-opt2.pdb.gz breaks before (T0282)I295 with cost 12.0266 T0282.try4-opt2.pdb.gz breaks before (T0282)Q309 with cost 1.07722 T0282.try4-opt2.pdb.gz breaks before (T0282)I307 with cost 0.729407 but try4-opt2 does score best with the unconstrained cost function. It looks like these breaks resulted from moving the previous bad breaks L293->I295, D308->Q309+I307 The gap at L293 moved the wrong way---it would have been better to unwind a little of the helix, say back to Y289. Let's put in gap-closing constraints for the new breaks (and M294), in addition to the old constraints, and try again. Fri Aug 20 13:40:25 PDT 2004 Kevin Karplus YOW! try5-opt2 has one horrendous break: T0282.try5-opt2.pdb.gz breaks before (T0282)D297 with cost 21.4346 T0282.try5-opt2.pdb.gz breaks before (T0282)I307 with cost 1.17039 T0282.try5-opt2.pdb.gz breaks before (T0282)K292 with cost 0.922252 The gap moved even FURTHER the wrong way! Let's do one polishing run with the unconstrained costfcn and be done. I tweaked try6.costfcn (adjusting soft_clashes and breaks) until try3-opt2 and try4-opt2 scored almost identically, so that crossovers between the two are likely. Fri Aug 20 21:04:42 PDT 2004 Kevin Karplus try6-opt2 has bad breaks, though not quite as bad as try3-opt2 (except for the first one): T0282.try6-opt2.pdb.gz breaks before (T0282)L293 with cost 11.3423 T0282.try6-opt2.pdb.gz breaks before (T0282)I151 with cost 3.72224 T0282.try6-opt2.pdb.gz breaks before (T0282)D245 with cost 2.42141 T0282.try6-opt2.pdb.gz breaks before (T0282)D308 with cost 1.61866 T0282.try6-opt2.pdb.gz breaks before (T0282)I307 with cost 0.977618 T0282.try6-opt2.pdb.gz breaks before (T0282)G126 with cost 0.820948 I'll have to think up a way to unwind the last half turn of the helix currently ending at L293. Sat Aug 21 07:56:24 PDT 2004 Kevin Karplus Maybe I could add a StrandConstraint for N291-M294, to try to force it straight (though it isn't a strand, just not a helix), as well as a strand constraint for the strand M294-E300, making sure to have gap-closing constraints before and after L293. The unconstrained costfcn likes try4 best, but if I add the constraints to try to close the gap at L293, the robetta models 1,2,3,4 score better, and try5 comes next. I'll do the optimization excluding the robetta models, since they do quite poorly other than at this gap. After the try7 run with the gap-closing constraints, I should probably do one more run unconstrained to polish the model. Sat Aug 21 11:00:51 PDT 2004 Kevin Karplus Well, try7 really botched it, pulling the strand out of the sheet entirely (and not unwinding the helix)! Let's try again, but keeping the sheet constraints from try6-opt2 needed to hold the strand in place, and not taking the fake strand constraint quite as far. I'll also make sure that the starting point is try6-opt2, by commenting out any models that score better with try8.costfcn in the ReadConformPDB lines. Sat Aug 21 14:26:27 PDT 2004 Kevin Karplus try8-opt2 has removed the terrible break, but still has some moderately big ones: T0282.try8-opt2.pdb.gz breaks before (T0282)I151 with cost 3.58145 T0282.try8-opt2.pdb.gz breaks before (T0282)D245 with cost 2.43321 T0282.try8-opt2.pdb.gz breaks before (T0282)K292 with cost 0.985878 T0282.try8-opt2.pdb.gz breaks before (T0282)G126 with cost 0.782637 Some of these were fixed in other runs (like try4-opt2). Should I do an unconstrained polishing run to try to close the remaining gaps, or should I superimpose the best models and make a chimera before polishing? try8-opt2, try6-opt2, and try4-opt2 are currently the best scoring with the unconstrained cost function. None of the models look good near I151. The way that try4-opt2 tried to fix the break is ugly. I could copy R150-G152 from the first undertaker alignment. Near D245, the try4-opt2 model looks pretty good, and I should paste in Q240-Y250. Perhaps better, I could paste in D244-Y250 from the second undertaker alignment. Near K292, the model from first alignment looks better than any of the tries, and I should paste in R285-M294. OOPS--the models from the first alignment is not complete there. The models from the alignment only look better because they don't have the inserted residues. I'll probably have to stick with try6 for this region. All the models look about the same near G126---not distinguished enough to be worth cutting and pasting (the cut-and-paste break would probably be a s bad as the existing ones). I made a patched model in decoys/try8-patched.pdb, but it scores very badly (clashes), so I'll just polish the existing models and call it quits. Sat Aug 21 20:25:57 PDT 2004 Kevin Karplus try9-opt2 has the following breaks (worst 3): T0282.try9-opt2.pdb.gz breaks before (T0282)I151 with cost 3.401 T0282.try9-opt2.pdb.gz breaks before (T0282)D245 with cost 2.11586 T0282.try9-opt2.pdb.gz breaks before (T0282)D308 with cost 0.551934 This is slightly, but not greatly, better than try8-opt2. It is probably not worth further fussing at this point. Rosetta still prefers try9-opt2.repack-nonPC, but it thinks all the models have terrible clashes. Undertaker agrees that there are bad clashes, but has not been able to relieve them: other-bump: 1.63694 Ang (T0282)E99.O and (T0282)A103.CB threshold= 2.77262 cost= 0.95765 other-bump: 1.73325 Ang (T0282)K225.O and (T0282)P229.CD threshold= 2.83689 cost= 0.947987 other-bump: 1.8996 Ang (T0282)C95.CB and (T0282)E99.CG threshold= 3.08387 cost= 0.945374 other-bump: 1.9057 Ang (T0282)E277.O and (T0282)P281.CD threshold= 2.83689 cost= 0.908109 other-bump: 1.87522 Ang (T0282)E51.CB and (T0282)G64.O threshold= 2.77262 cost= 0.904286 other-bump: 2.16768 Ang (T0282)C95.CB and (T0282)E99.CB threshold= 3.15526 cost= 0.894863 other-bump: 1.94919 Ang (T0282)P259.CD and (T0282)V274.O threshold= 2.83689 cost= 0.894786 self-bump: 1.63361 Ang (T0282)I307.N and (T0282)I307.C threshold= 2.19456 cost= 0.829862 other-bump: 2.43264 Ang (T0282)F227.CA and (T0282)H235.NE2 threshold= 3.19726 cost= 0.805999 other-bump: 1.89817 Ang (T0282)D53.O and (T0282)N59.O threshold= 2.47194 cost= 0.794986 other-bump: 2.19516 Ang (T0282)E99.O and (T0282)A103.CA threshold= 2.83765 cost= 0.785689 other-bump: 2.15488 Ang (T0282)A68.CB and (T0282)G124.O threshold= 2.77262 cost= 0.779609 neighbor-bump: 1.90557 Ang (T0282)I307.N and (T0282)D308.N threshold= 2.42906 cost= 0.76691 other-bump: 2.1131 Ang (T0282)P43.O and (T0282)P86.CG threshold= 2.68408 cost= 0.761904 other-bump: 2.19867 Ang (T0282)T144.C and (T0282)V146.N threshold= 2.74083 cost= 0.733515 other-bump: 2.28852 Ang (T0282)G49.O and (T0282)N92.CA threshold= 2.83765 cost= 0.724845 The two bad proline problems come from putting a proline in mid-helix, a most unlikely event! The bad E99 problems seem to come from a poor placement of the E99 sidechain. The chance of my fixing these problems without doing great damage to the model seems slight, so I think I'll let them go. If I change the unconstrained costfcn to put more weight on clashes, try9-opt2 still comes out best. (The robetta models all have worse clashes, so I shouldn't be too concerned.) I'll submit try9-opt2 best unconstrained try8-opt2.repack-nonPC best Rosetta energy try1-opt2 full auto T0282-1gq6A-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template T0282-2cevA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 2nd template