Tue Jul 27 11:12:38 PDT 2004 T0251 DUE 15 Aug 2004 Wed Jul 28 10:30:01 PDT 2004 Kevin Karplus This looks like fold-recognition with c.47.1 as the superfamily. Thu Jul 29 02:59:43 PDT 2004 Kevin Karplus There are almost no homologs found, so rr predictions are pretty useless on T0251. The fold-recogniton prediction produces very good alignments in the t2k.undertaker-align.pdb file, but the sheets are not quite forming hbonds, so get lost in try1-opt2. We should probably pick up some constraints from the alignments to keep things together. MANUAL_TOP_HITS:=1egr 1fo5A 1ego 1kte 1grx 1qfnA 1jhb 1hyuA 1t1vA 1j0fA 1thx 1r26A 1iloA Thu Jul 29 09:50 PDT 2004 Kevin Karplus Observation: The reason (at least a reason) try1 was so messed up was that the alignment all-align.a2m.gz was not getting properly created before try1 was run. I moved try1 to no_align1, and am rerunning try1 after creating all-align.a2m.gz properly. Fri Jul 30 05:24:36 PDT 2004 Kevin Karplus try1 gets 3 of strands into a sheet and it looks like the s2 strand may fit on the edge, perhaps s1 || s2 maybe M1-D7 || F39-D45. C10 and C13 may be part of metal-binding (or ss bond?) Mon Aug 2 11:37:08 PDT 2004 Kevin Karplus try2 gets almost 4 strands in sheet. Helix K20-K31 has its buried side facing out. I think that we may want to add a sheet constraint for K31-K33 antiparallel to T42-K40, with T42 hbonding. On the other end of the shee we should hadd L73-M76 antiparallel to A82-E79. Wed Aug 4 02:14:22 PDT 2004 Kevin Karplus The sheet is looking sort of ok in try3, but I wonder if I have the right toplogy. The protein is looking rather foamy. Fri Aug 6 08:29:42 PDT 2004 Sol Katzman Created a script pointed to by 'strands' that defines the strands S1,S2,S3,S4 and sheet1 (as they are in try3-opt2). In the 1kte template, the highly conserved cysteines (C10,C13 in T0251) do indeed form an SS-bond, holding them in the first turn of a helix. In try3 these residues were not part of the first helix, so in addition to the SSbond, add a HelixConstraint to include them. Also note that in try3, the M1 residue is not part of strand S1 and the Hbonds to strand S1 are shifted in phase by one residue, compared to try2. The try3 sheet looks better to me. As far as the topolgy, looking at template 1kte and 1hyuA(residues 1-102) I think one side of the sheet is supposed to be buried by a) the helix between S1 and S2 (some of C10-K31) b) the final C-terminal helix (some of K89-L99) The other side of the sheet can be buried by: c) the helices between S2 and S3 (some of D57-E65) (in 1kte, there is also an N-terminal helix to use here) (in 1hyuA, there is replication of the domain) We also need to get the hydrophobics in these helices to face the sheet. These constraints are attempts to do that: item (a) Constraint I22.CD1 A6.CB Constraint L26.CD1 ?? Constraint L29.CD1 V2.CG2 item (b) Constraint I91.CD1 I74.CD1 Constraint I95.CD1 V81.CG1 Constraint L99.CD1 I37.CD1 ?? item (c) Constraint L58.CD1 I44.CD1 Constraint I61.CD1 V2.CG1 Constraint I64.CD1 L73.CD1 Fri Aug 6 21:45:13 PDT 2004 Sol Katzman Did not get the SS bond in try4, but otherwise the model did a lot of what was desired, in terms of moving two helices to one side of the sheet, and one helix to the other side of the sheet. The model does have a lot of breaks and is somewhat foamy. For try5 use the HelixConstraints from try4, increase the weight of the SheetConstraints from try3 (since they are better than those from try4), increase the weight of the constraint that defines the desired SS bond. Finally, increase the weight of Break (20 -> 40). I note that Kevin fetched all the robetta models. For try5, use read-pdb.under, eliminating TryAllAlign. Sat Aug 7 15:07:40 PDT 2004 Kevin Karplus try5 looks pretty good to me. I made an "unconstrained" cost function that includes the disulphide. It scores try5-opt2 best. I'll run a try6 that optimizes existing models using an unconstrained cost function. First, I'll make extra_alignments to make sure that all the alignments are present (they were not), and remake all-align.a2m.gz It may be worthwhile to give try5-opt2 (or try6-opt2 when it is done) to VAST and see if we can come up with a better initial alignment to work from. It may not be worth it at this point though, as we have so many targets and so little time. Sun Aug 8 08:26:31 PDT 2004 Kevin Karplus try6 looks slightly better than try5 and scores best. Rosetta prefers try1-opt2.repack-nonPC. I submitted try6 to VAST: job VS60572 password T0251try6 When the VAST job is done, we should pick up the alignments and see if they help any in creating a new model. Sun Aug 8 15:25:36 PDT 2004 Kevin Karplus The best VAST p-value is for an alignment to 1kte, which is already one of our templates. I'll save that as 1kte/1kte-T0251-try6-vast.a2m Next best is 1su9A=1st9A. Third is 1fb0A=1fb6A. Fourth, 1h75A There are many others (total 140 found), but these should be enough coverage of the thioredoxin fold, which was our fold-recognition hit. I'll add these templates to MANUAL_TOP_HITS and make extra_alignments. try7 will be another run from alignments, with a cost function based on try6 sheets. I just noticed that t04.many.frag was never made, so I'll do a make to create it also. Mon Aug 9 16:00:24 PDT 2004 Sol Katzman As discussed at out group CASP meeting, try7 is terrible (lost 2 strands from the sheet, helices sticking out into space etc.) But try7 does score best on Kevin's unconstrained costfcn, on which it does better than try6 due to sidechain,bystroff,clashes,break, but worse than try6 on dry8,dry12,phobic_fit,pred_alpha. So I have tweaked unconstrained costfcn by increasing phobic_fit and dry8,dry12 while decreasing break (100 -> 60) The two bad breaks in try6 are K40-Y41 and Y71-P72. In an effort to avoid these breaks without having to use a strong weight on breaks, modify the SheetConstraints to include K40 and P72 for try8. Wed Aug 11 10:14:42 PDT 2004 Sol Katzman The try8 model had the same bad breaks as try6, with no change in the sheets. For one last try, I will use the same constraints as in try8, but for try9, will not read in the existing models; instead use TryAllAlign. Wed Aug 11 13:14:19 PDT 2004 Kevin Karplus I turned the sidechain weight way down in unconstrained.costfcn---it was being grossly overweighted. try7-opt2 still scores best on the unconstrained costfcn, but try8 is not hopelessly far behind. Tweaking a little more (increasing phobic_fit and the pred_alpha* weights, and lowering soft_clashes and breaks) moves try8-opt2 up until it just beats try7-opt2. For try10, I'll use the unconstrained costfcn plus the sheet constraints from try8-opt2.sheets. Wed Aug 11 20:26:54 PDT 2004 Sol Katzman It looks like try9 did resolve the two breaks at Y41,P72 but introduced a new break at N50 as it attempted to unify the helix D51-I64. The unconstrained cost fcn likes try10 best by a good margin, followed by try8,try6,try7,try9. Rosetta likes try1, then try3,try7,try10 Probably we should submit this now. Thu Aug 12 06:35:56 PDT 2004 Kevin Karplus I think I'll do one polishing run with an unconstrained cost function, not expecting much improvement. Thu Aug 12 11:43:32 PDT 2004 Kevin Karplus try11-opt2 scores a bit better than try10-opt2. submit: try11-opt2 best with unconstrained costfcn try1-opt2 full auto try1-opt2.repack-nonPC best Rosetta energy try3-opt2.repack-nonPC 2nd best Rosetta energy, best with try3 costfcn, slightly different solution? T0251-1egr-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m Sun Aug 15 17:56:13 PDT 2004 Kevin Karplus OOPS, I just edited try11.under by mistake (I wanted to edit T0237/try11.under.) I don't have a back-up file, but I only changes the OptConform parameters, so I should be able to restore them from try11.log.