Tue Jul 6 08:05:24 PDT 2004 T0225 Due 23 Aug 2004 Tue Jul 6 12:21:57 PDT 2004 Kevin Karplus Looks like a comparative model with c.2.1.2 as the SCOP fold: Tyrosine-dependent oxidoreductases. Best score is for 1r66A, which seems to be a full-length alignment with some loops to build. All the models from alignments in the T0225.t2k.undertaker-align.pdb file are very similar. There will be a little loop closing and sidechain packing to do, but I think that this will be a pretty easy target. Tue Jul 6 18:25:18 PDT 2004 Kevin Karplus The top 29 hits (all c.2.1.2) were used as templates for try1, and the resulting model looks quite good. The large hydrophobic-lined hole is a binding pocket for the ligands in the crystals. There are a few small gaps to close and we may want to add a sheet constraint (K2-T6, E26-T30, hbond I5) to get the edge sheet firmly connected. For try2, I took the sheet constraints from try1-opt2.sheets and the helix constraints by using rasmol (select helix, show selected) and then trimming the helices back to just the part predicted to be helix by t2k str2 predictions. From karplus@soe.ucsc.edu Tue Jul 6 19:13:33 2004 Date: Tue, 6 Jul 2004 19:13:31 -0700 From: Kevin Karplus To: ggshack@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: MI working great on T0225 On T0225, where we have a good comparative model, the t2k-thin30.mi constraints file seems to be working great---the top constraints are all met: Constraint F105.CB H205.CB -10. 7.0 12.0 1.15928 Constraint Y152.CB S270.CB -10. 7.0 12.0 0.937908 Constraint T6.CB N57.CB -10. 7.0 12.0 0.835217 Constraint W150.CB G180.CA -10. 7.0 12.0 0.790444 Constraint V191.CB C206.CB -10. 7.0 12.0 0.7295 Constraint Y152.CB S184.CB -10. 7.0 12.0 0.703713 Constraint T181.CB K209.CB -10. 7.0 12.0 0.677302 Constraint N78.CB Y125.CB -10. 7.0 12.0 0.63897 Constraint W150.CB P182.CB -10. 7.0 12.0 0.637143 This is a phenomenal success rate for contact prediction! It probably helps that there are so many diverse sequences in the family. You might want to look at the first one though--- F105.CE1-H205.CE1 is only 3.88 Ang (contact), but the CB atoms are 9.95 apart. This is an example of why I think you should train for sidechain centroids, rather than CB atoms. Of course, the opposite phenomenon seems to be happening for T181-K209, but that one may be a result of a poor rotamer choice. -------------------------------------------------- Wed Jul 7 07:07:49 PDT 2004 Kevin Karplus The mutual information constraints are good for this target, but took a very long time to compute. It's really to bad that this has such close homology to known structures that we don't need the mutual information constraints! The try2-opt2 model looks pretty good. There are 2 bad breaks: before D36 and before K230, and lots of smaller ones. We should probably do a polishing run from the existing models with the breaks and clashes turned up and constraints turned down. Wed Jul 7 17:02:43 PDT 2004 Kevin Karplus Try3 looks pretty good, but still has a break before D36 (and clashes between L35 and D36). Thu Jul 8 06:21:57 PDT 2004 Kevin Karplus try4 seems to have reduced the L35-D36 clash, though not eliminated it. I think this is basically done, unless we want to use the mutual-information constraints to try to improve sidechain rotamers. That would require looking at the top 30 or so constraints, making sure that they are near in the structure and changing them into constraints on distal atoms of the sidechains. Mon Jul 19 17:06:29 PDT 2004 Kevin Karplus Looking at the conserved residues, I think we may want to make an Hbond between E115.OE1 and R147.NH2. Looking at the top mutual information contact predictions using the new rr script, it looks like we might want to extend the sheet constraint to pair C58 and I99 more closely. (Or perhaps just a CB distance constraint for parallel strands of 5.14) No--things are a bit loose there, so a sheetconstraint is in order. We may want to add a distance constraint for I56.CD1 and I196.CD1, which are currently 4.42Ang apart, but should probably touch (< 4Ang). I don't think I want to try to move F204 and R147 to the same side of the sheet, even though they have a predicted contact, Burying a histidine to expose a phenylalanine sounds like a really bad idea. I'll do a polishing run with a couple of extra constraints and with the dry weights turned up, to try to get tighter packing. Mon Jul 19 21:58:39 PDT 2004 Kevin Karplus try5-opt2 has slightly better score than try4-opt2 with the try5 costfcn, and with unconstrained costfcn. Rosetta still prefers the repacked version of try4-opt2. Note for model 4 that we are submitting: Model 4 is SCWRL sidechain replacement for the alignment T0225-1n2sA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5, using our second-best-scoring template 1n2sA. The best-scoring template (1r66A) was observed in PDB, but was not in our template library, so we could only do sequence-based alignments to it, which are not reported as models in the automatic method. We did not think it worth the trouble to add 1r66A to our template library, as the models we were getting looked good enough without the full set of alignments for 1r66A. Wed Sep 22 11:33:34 PDT 2004 Kevin Karplus Using GDT score, our best model is try5-opt2 89.2857% The order for submitted models is model1 88.3036 this should be try5-opt2 why the difference? model2 88.1250 best rosetta energy robetta3 86.2500 model4 86.1607 robetta4 85.8929 robetta5 85.8036 model3 85.6250 full auto robetta2 84.9107 robetta1 80.6250 model5 65.0893