Wed Jun 30 08:30:23 PDT 2004 T0223 Due 19 Aug 2004 Wed Jun 30 09:05:08 PDT 2004 Kevin Karplus T0223 looks like a comparative model for d.90.1.1, with 1nox as the best-hit template. Sat Jul 17 21:54:55 PDT 2004 Kevin Karplus None of the initial alignments used in try1 are very complete. The top hit (1nox) is a dimer, with the C-terminal strand domain swapped. There is a 4-strand antiparallel sheet (order 1,2,4,3, with strand 5 parallel to strand 1 of the other monomer). Sun Aug 8 13:15:30 PDT 2004 Kevin Karplus I haven't looked at this in a long time, but I'm beginning to doubt the classification of this as "comparative" rather than "fold recognition", despite the very small e-values for the 1nox and 1bkjA templates. I reran make to get the bys predictions and to create t04.many.frag. The big problem with this target is that our secondary structure predictions do not agree with the template hit. The t04 predictions fit much better than the t2k predictions, so I'll stick with t04. I'll probably have to make dimers to see whether the predictions are any good, and (maybe) add scaffolding for the domain-swapped strand. The t04 predictions like 1bkjA best, so maybe I should use that as the basis for the scaffolding. Tue Aug 10 16:12:35 PDT 2004 Kevin Karplus The try2-opt2 predictions look rather different from the try1-opt2 ones. Both dimerize without serious conflict, but the try2-opt2 structure has big holes in it. try1 scores better than try2, even with the try2 costfcn, but try2 is more consistent with the templates and with the secondary structure prediction. I tweaked the try3 costfcn until it barely scored try2-opt2 better than try1-opt2, and will run another prediction from the alignments. Wed Aug 11 16:10:08 PDT 2004 Kevin Karplus I don't like try3 at all. It looks bad by itself and it doesn't dimerize well. I'm close to giving up on this one---just submitting try1 and try2. Sun Aug 15 11:01:08 PDT 2004 Kevin Karplus Before giving up, maybe I should do an unconstrained optimization. (If all else fails, I'll try adding scaffolding constraints taken from the dimers.) Sun Aug 15 16:00:45 PDT 2004 Kevin Karplus try4 is the new best score for unconstrained cost fcn. It makes an ok dimer, with only a few bad clashes. I'll submit try4-opt2 best undertaker cost try2-opt2 different possibility try3-opt2 different possibility try1-opt2 full auto T0223-1nox-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template In essence, we had a comparative model for one domain, and new-fold predictions for the domain in which is inserted. The new-fold predictions are rather suspect. We probably should have allocated more time for this target. Fri Sep 24 21:03:04 PDT 2004 Kevin Karplus It looks like robetta2 beat us on this one, and our alignment to a template did better than our hand-tweaked models: name length missing_atoms rmsd rmsd_ca GDT smooth_GDT robetta-model2.pdb.gz 206 0.0000 15.6798 14.2747 -37.7427 -36.1078 model5.ts-submitted 206 695 12.2003 8.7848 -36.0437 -31.9693 template model3.ts-submitted 206 0.0000 17.6404 16.0809 -28.8835 -27.9836 model2.ts-submitted 206 0.0000 22.8165 22.1952 -28.1553 -26.6059 model4.ts-submitted 206 0.0000 22.5525 21.3095 -20.6311 -19.9116 full atuo model1.ts-submitted 206 0.0000 22.5749 21.3470 -20.6311 -19.9115 robetta-model4.pdb.gz 206 0.0000 20.6790 19.6160 -13.9563 -12.9452 robetta-model1.pdb.gz 206 0.0000 20.4890 19.3381 -13.9563 -12.9365 robetta-model3.pdb.gz 206 0.0000 20.9395 19.4141 -13.5922 -12.8825 robetta-model5.pdb.gz 206 0.0000 21.3137 19.9422 -10.9223 -10.1049