Mon Aug 16 17:24:22 PDT 2004 T0280 Mon Aug 16 20:35:06 PDT 2004 Kevin Karplus Comparative model with 1ecfA (and other c.61.1.1 templates). Tue Aug 17 15:15:47 PDT 2004 Kevin Karplus try1-opt2 looks pretty good, except for the packing of the N and C terminal helices. I think that V26-G29 probably should be a strand rather than a helix, In fact, in the first couple of models in the t2k.undertaker-align file, they ARE in a sheet. Although all the models have the same fold, there seem to be two main classes of alignments, and it looks to me like try1-opt2 picked the wrong class (that used for models 5,10,12, and 13 in t2k.underaker-align). If I pick up sheet constraints from some of the other alignments, I should be able to select the right alignments in try2. In the critical region, the t2k multiple alignment has more conservation than the t04 one, so I'll rely more on it for constraints. For try2, I'll include lots of sheet constraints, including conflicting ones, with "bonus" keywords. My hope is that try2 will come up with a consistent alignment that looks good, so that I can the do a polishing run without sheet constraints. I'll include the better rr constraints in try2 also. Tue Aug 17 22:18:08 PDT 2004 Kevin Karplus try2 was pretty much a waste of time. The try1-opt2 model scores better with the try2 costfcn than try2-opt2. I guess I'm going to have to choose the alignment that I most want to mimic, and force those sheet constraints. I'm probably too tired to do that tonight---I'll try it tomorrow (hoping to be more awake). It may be necessary to set up several different runs and try them in parallel. Wed Aug 18 15:02:39 PDT 2004 Kevin Karplus One alignment that looks pretty good is model 3 in t2k.undertaker-align. I'll try picking up align3.sheets. Thu Aug 19 17:46:24 PDT 2004 Kevin Karplus try3-opt2 looks pretty good. E48-G61 are a bit questionable. The unconstrained costfcn likes try1-opt2 better than try3-opt2, despite worse breaks and clashes. The pred_alpha and phobic_fit are better. I'll up the weights of clashes and breaks a little, so that try3 is preferred and use that as the try4 polishing run. Fri Aug 20 12:59:55 PDT 2004 Kevin Karplus try4-opt2 now scores best (with both try4 and unconstrained costfcns). It still has the questionable section E48-G61, but I doubt that further polishing will accomplish much. There are bad breaks in try4-opt2: T0280.try4-opt2.pdb.gz breaks before (T0280)A183 with cost 2.06916 T0280.try4-opt2.pdb.gz breaks before (T0280)A182 with cost 1.59626 T0280.try4-opt2.pdb.gz breaks before (T0280)L22 with cost 1.25666 T0280.try4-opt2.pdb.gz breaks before (T0280)R108 with cost 1.03028 T0280.try4-opt2.pdb.gz breaks before (T0280)V184 with cost 0.887597 T0280.try4-opt2.pdb.gz breaks before (T0280)P63 with cost 0.839448 The first two bad breaks are in a region predicted to be helical, but made into very short hairpins. I'm surprised that this region has not been reshaped already by undertaker. It may be worth adding some gap-closing constraints and a helix constraint to clean up this region. Actually, helices are SO weakly predicted that it is better not to have a constraint for them--this is more likely to be a turn between strands. I've added the gap-closing constraints to try5, and am repotimizing try3 and try4 runs. Fri Aug 20 17:11:59 PDT 2004 Kevin Karplus try5-opt2 has no bad breaks and is the best-scoring models with unconstrained.costfcn (beating try4-opt2 by a tiny amount). try4-opt2.repack-nonPC still has the best Rosetta energy. I'll submit try5-opt2 best unconstrained try4-opt2.repack-nonPC best rosetta try1-opt2 full auto T0280-1ecfA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best t02 template T0280-1o5oA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best t04, 10th t02 Thu Nov 18 20:16:35 PST 2004 Martina Koeva Here are the smooth GDT scores for this target: best sam-t04 32.0870 best submit 32.0773 model1 32.0773 auto 26.4340 align 30.8845 robetta best 37.2053 (robetta model 2) robetta1 20.6707 This target would definitely be an outlier in terms of the comparison between robetta model 1 and our model 1, but robetta model 2 did beat our models.