Tue Aug 13 12:45:55 PDT 2002 t0183 31 Aug 2002 kevin Karplus 19:00 Obvious homology model for 1jcjA Best score try1-opt-scwrl. Modifying score function to up dryness and breaks, and to include hydrogen bonds makes try1.1.60 best. There is a bad break in try1-opt-scwrl between V86 and T87, probably as a result of misalignment of T87 and V88, or of extending the helix at E79 for too long Currently have Constraint (T0183)T87.O (T0183)D117.N 2.2 2.7 3.2 Constraint (T0183)V89.N (T0183)D117.O 2.2 2.7 3.2 Constraint (T0183)V89.O (T0183)V119.N 2.2 2.7 3.2 80> gtdvKVVTVVgfp 110> hsgadEIDMVINvg probably should have 60> strfHGVCVnpc 80> gtdvKVVTVVgfp 110> hsgadEIDMVINvg Putting these constraints (as CB constraints) into try2.constraints and adding them to the score function leaves try1.1.60 as the best. It has the correct alignment for part of the V89:M118 connection. Let's start over from the alignments, with these constraints. 1 Sept 2002 Kevin Karplus 11:44 Try2 crashed because I was using the buggy new version of undertaker. Going back to the previous version for try3. try2.0.120, the current best score, looks like it needs one more strand to the barrel---probably F16-K20 14> reFYEFKpvr 60> nrfHGVCVnpc 80> gtdvKVVTVVgfp 110> hsgadEIDMVINvg 145> kgkvVKVIIEtcy 173> agahFVKTstgf 203> demgVKAsggi 14> reFYEFKpvr 2 Sept 2002 11:52 Kevin Karplus (try3 was run from try2.constraints, as was try4, but try5 was run from try5.constraints) With the try5.constraints, try5-opt scores best. Strand G63-C65 looks like it needs some Hbonds to snap it into place, as does Y17-K20, but they are in roughly the desired places. G206-S210 also needs hbonds to finish forming a TIM barrel. We seem to be short a strand---let's try inserting one 14> reFYEFKpvr 35> iehtnlkpfa 60> nrfHGVCVnpc 80> gtdvKVVTVVgfp 110> hsgadEIDMVINvg 145> kgkvVKVIIEtcy 173> agahFVKTstgf 203> demgVKAsggi 14> reFYEFKpvr These constraints are try6.constraints. With them, try5-al10.4.120 and try5-1l4.3.120 score best. Let's do try6 as a run from scratch, and try7 as a run from the best several existing models, both using try6.constraints. 2 Sept 2002 14:52 Kevin Karplus try6 is done, try7 is still running. Current best (with try6.constraints) is try7.0.120, though try6-al10.1.60 is nearly as good. try7.0.120 has 6 good strands, and is trying hard to put in the first two. try6-al10.1.60 also has 6 good strands, but seems to be having less success with the two difficult ones. 20:54 try7 run finished. Best (with try6 constraints) is now try7-opt. With no constraints, best is still try3-opt, which has 7 pretty good strands. Tomorrow, when I'm more awake, I'll try copying the constraints accurately from try3-opt and try to add in the 8th strand. I think it shows more promise than try6 or try7. Wed Sep 4 16:40:54 PDT 2002 Kevin Karplus I didn't work on T0183 yesterday after all--I'd better start on it today. I used Jonathan's version of undertaker to extract the hbonds from try3-opt, then cleaned them up to make try8.constraints. But ttry3-opt is missing a strand---let's try to guess where it goes and add that to try8-opt. The 7 strands are I35-T38 predicted to be helix H62-N67 K84-V89 E115-D136 V148-I152 H176-T180 M205-S210 weakly predicted All the strongly predicted strands are present, and there seem to be a few possible places for the barrel-closing strand: 230 IGTSS 242 GEERYGG 14 REFYEFKP Of these, the N-terminal seems most promising Let's try akyREFYEFkp saIEHTnlkpf nrfHGVCVnpc gtdvKVVTVVgfp sgadEIDMVInvg kgkvVKVIIEtcy agahFVKTstgf demgVKASgg ........akyREFYEFkp With try8.constraints, try6-al10.1.60 scores best. Let's do a run with the try8.constraints from several of the better models. The conserved residues are marked in script conserve---the conserved charged residues cluster very nicely in the middle of the barrel in try3-opt. 4 Sept 2002 21:02 Kevin Karplus The try8 run produces a fair barrel, but the first strand is still having trouble. I tried moving it over two, to get more charges into the center of the barrel and to get to a straighter part of the backbone (based both on visual inspection and the alpha prediction). I also added helix length constraints, to try to make sure the helices stay wound up. I'll run try9 for 2 hours, then see whether I want to do a long run with those constraintsor modify them again. 4 Sept 2002 23:48 Kevin Karplus Strand Y17-V22 has been yanked out of the barrel somewhat. A34-F43 doesn't seem to have a straight strand within it. P42-F61 is rather broken up. G206 to G212 may have to back up a couple of spaces. I'll try the following alignment next: kyREFYEFkp saIEHTnlkpf nrfHGVCVnpc gtdvKVVTVVgfp sgadEIDMVInvg kgkvVKVIIEtcy agahFVKTstgf demgVKASgg ......yREFYEFkp as try10.constraints. Thu Sep 5 13:48:28 PDT 2002 try10-opt looks pretty good, except that K12-F19 is wound into a helix instead of being straight. I have a straightness constraint, so it may be that just optimizing some more will fix it. Fri Sep 6 10:55:24 PDT 2002 Kevin Karplus try11-opt scores better than try10-opt, but hasn't fixed the helical piece. Maybe I need a straightness constraint directly on E15-F19. Fri Sep 6 12:33:55 PDT 2002 I may have turned OptSubtree down too much in try12, it has not succeeded once in the first 84 generations. Let's simultaneously run try13, with OptSubtree turned up a bit. Fri Sep 6 13:31:38 PDT 2002 try13 doesn't seem to be any faster than try12 at making improvements in the score function. 16:54 But on re-reading in the -opt files, try13-opt was better than try11-opt, and try12-opt was worse. The first strand has straightened out some, but still needs work. I'll add more constraints in the hope that it pulls things together. 21:43 try14-opt is a little better, but E15-F19 is STILL wound into a helix. I'll up the constraint weight and try again. 23:25 I'm still not getting that strand straight. Either undertaker has gotten stuck or I've got some constraints wrong. Maybe I should give this to VAST to try to align to a real TIM barrel. job VS31282 password casp5t0183 7 Sept 02 00:49 Vast job still running---I'm going to bed. 08:25 Vast like the homolog to 1jclA, and puts the extra strand at the end, not the beginning. I extracted the VAST alignment's hbonds into T0183-1jclA.hbonds, and sorted the backbone hbonds into try16.constraints. Incidentally, C65 is conserved from 1jclA, and does point into the center of the barrel, so that strand is probably oriented correctly. Let's run try16 from scratch. 10:06 Well, the try16 run DOES look much better than try15---the strands have no trouble forming, and the active site is pretty compact. I do think that E36 is pointing the wrong way though. So that strand has to be moved up or down one. I hand-edited the alignment to move E36 up 1 to line up D in 1jclA, which also lined up some Cys later on (T0183-1jclA.vast-hand2.a2m). I made a PDB file from in (T0183-1jclA.vast2.pdb) and extracted hbonds (T0183-1jcla-2.hbonds) Sigh---I don't know why it took me so long to get this alignment---the conserved E36 should have been a strong signal that is was the correct strand and how it should be aligned. 10:43 The try17-al12 initial model looks pretty good (all the charges near the barrel point into it), but we probably want to pack the helix R23-A34 against the barrel. 13:05 try17-opt looks pretty good. There are a few small breaks to patch, so let's do one more run, with the break penalty set higher and constraints set much lower. Then it'll be ready for submission, because I don't know what else to do with it. 15:40 Well, I would have LIKED one of the N-terminal helices to fold agains the barrel, but I'll take what I can get and submit try18-opt.