t0134 11 June 2002 Kevin Karplus T0134 is the delta-adaptin appendage domain. Its closest fold-recognition template is 1b9kA, the alpha-adaptin appendage domain. The scores are not overwhelming, but given the similarity of names, it is almost certainly the right template. The FSSP file (1qtsA.fssp) has several proteins with Z-score>7. Most are only 100 residues long at about 2.4-3 angstroms, but 1e42A has a 202-long alignment at 4.9 Angstroms. (1b9kA has 99% residue id to 1qtsA, with 0.7 RMSD). 13 June 2002 Kevin Karplus In undertaker-try2, it looks to me like there are several strands floating off in space that should be folded against the existing sheets. I think that the second sheet is getting blown apart to close breaks---I have to either get the hydrogen bond working, or add manual constraints. Strand 63-67 is predicted to be helix, so maybe should be removed? The predicted strand at 213-220 should probably pack against the sheet, as should 140-148. We may have to create an alignment by hand 20 June 2002 Kevin Karplus I modified undertaker to be able to output PDB files for alignments in the alignment library today (see show-align.script and T0134.undertaker-align.pdb). Looking at the alignments, it is clear that many of them cluster well (not surprising, since many are from the same templates). We can discard some outliers, based on their poor appearance in the cluster. The "script stride" command can't be used yet, because it uses sequence-based numbering, but the T0134.undertaker-align.pdb file uses the proper pdb numbers (which don't start at 1 for this target). Bad models include 10, 60, 2, which stick way out from the rest. model 2 is called T0134.t2k-2track-undertaker.a2m:1dt9A model 10 is called T0134.t2k-2track-undertaker.a2m:1qu6A model 60 is called T0134-1dm8A-simpleSW-adpstyle5.pw.a2m.gz:1dm8A I really need "script stride" to see which have bad alignments (strands and helices in the unpredicted places). 25 June 2002 Kevin Karplus T0134.try3-opt.pdb looks like it has some good points---it remains compact and the beta sheets are not blown apart, but a lot of the strands are predicted to be something else, so I'm still not happy with it. We should probably redo the "best" scoring with tighter thresholds, since we now believe we know the correct SCOP domain. The scripts stride, dssp, str, and alpha now work OK, as I fixed the way of generating them to look for the residue numbers in the blank.pdb file. 26 June 2002 Kevin Karplus try2-opt still scores the best (better than try3 or try4), but it has the wrong residue numbering. Perhaps we should try starting from all existing decoys, to mix and match the best parts? I like the way try4 looks a little better than try3. 30 July 2002 Jonathan Casper I'm re-running try4 with SCWRLing and pred_alpha information. 11 Aug 2002 Kevin Karplus Running make with the updated template library picked up a new best hit (1kyfA). I made the try-alignments target, updated the undertaker.script file and define-score.script, and reran undertaker. 13 Aug 2002 Kevin Karplus Best score is now try7-opt-scwrl, and it seems to be based mainly on 1kyfA. It looks pretty good, except that the N-terminal strand (S888-T893) and the linker between the two domains (C1003-I1008) are predicted to be strands but have become helices. It looks like the Hbonding positions of the second strand are V906, V908, and I910, though try4 has T907-A909-V911 bonding to the first strand. The antiparallel strands I want to match are YVKMTD and QVTVAIVLE 889> yvkmtd 913< elviavtvq I'm not sure how to align them (other than putting K891, E913, and Q905 on the same face). Let's try again, starting from all the try*-opt decoys, with lots of CrossAndInsert. I still don't know what constraints would be good to put in to get better alignments. When I start getting a clue there, I might want to run again from just the alignments. 14 Aug 2002 Kevin Karplus Best score is now try8-opt-scwrl. I still wonder if we might be best off making a hand alignment to get the strands where they are predicted to be---or perhaps using a 2-track HMM with a much higher weight than usual on the 2ry track? I don't see much point to reoptimizing at ths point unless we start from a different alignment. 15 Aug 2002 Kevin Karplus I added the try-heavy target to the Makefile (and heavy-str to Make.main) so that the pairwise alignment with a heavy str track could be made and included in the alignment set. I tweaked the score parameters so that breaks would not be too expensive (I'll have to raise it again later), and started try9 from scratch. Note: the heavy-str target in Make.main could serve as a model for the other 25 types of pairwise alignment we try, instead of having them hidden in a PERL script. Hmm---it looks like try9 is not going to use those heavy alignments, preferring the other alignments (probably because they specify more of the chain). Perhaps I should also start a try10, that uses only the heavy alignments in the TryAllAlign, but allows the others after the TryAllAlign? I like the aligment in try10-al12---let's just hope that the optimization doesn't throw it away. (If I run into trouble, I can try turning off InsertAlignment.) Fri Aug 16 13:18:14 PDT 2002 Kevin Karplus Best score is try8-opt-scwrl, and try10-opt-scwrl looks terrible, but I still like the overall look of try10-al12, but * The S918-I919 break is bad. * The L976-K977 break is bad * The region around T991-L995 is a mess * The gap T1009-T1010 is a bit big * The N1042-L1043 is huge, but could be fixed by moving L1043-H1052 * The S1064-M1065 gap is large * The K1081-G1082 gap is large. Let's try another run, with InsertAlignment turned off. I doubt that it will help much, but who knows? 17 Aug 2002 Kevin Karplus try11-opt-scwrl barely beats try9-opt, but is beaten by try7-opt-scwrl and try8-opt. Unfortunately, try11-opt-scwrl is a terrble mess, with all the beta sheets blown up. I think we should submit try8-opt as our first and try10-al12 as our second model, and give up on trying to reconcile them.