Mon Jun 24 09:19:56 PDT 2002 T0140 has strong hits to 1mjc, 3mefA, 1c9oA (all homologs of 1mjc) and to 1sro, 1g03E, .... all in SCOP domain b.40.4.5 This looks like a homology-modeling target. We should probably remake the "best-scores" with tighter thresholds, to eliminate noise. 25 June 2002 Kevin Karplus The T0140.try1-opt.pdb file does not look like a good structure---the secondary structure match is not so great. The alignments for 1mjc and 3mefA in T0140.undertaker-align.pdb look fine though, so I should try running undertaker again with a smaller set of starting structures. Some of the domains have indulged in domain swapping, which may be relevant, since the protein is believed to form a tetramer. Note: this protein is "a synthetic protein constructed by non-homologous recombination. The N-terminal part derives from cold shock protein A (CspA), while the C-terminal segment comes from the E.coli 30S ribosomal subunit protein S1." Several of the ribosomal subunits are in the cold-shock superfamily, but I don't know about S1. 26 June 2002 Kevin Karplus try3-opt scores slightly better than try1 and try2, but has really mangled the beta sheet. 29 July 2002 Jonathan Casper Adding SCWRLing and pred_alpha to the undertaker script. 3 Aug 2002 Kevin Karplus try4-opt is current best. Added dry5 and phobic_fit to score function, which still leaves try4-opt as best. The beta sheet is being pulled apart though, particularly at K70-T75, which are PAST the end of the beta sheet copied from 1mjc. But 1mjc forms a closed barrel, so we need to look at some more alignments. Let's try that for try5 (using the read-alignments.under scripts for the top few templates), before we try adding constraints to hold the barrel together. Hmm---after trying all the alignments, the best scorer did not have a complete barrel. Let's see what comes out at the end, then add what constraints we can to fix things up. It seems that the alignment to the barrel is not obvious. I think that there are two main choices: to widen the barrel by 2 strands, or to lengthen two of the strands to form a hairpin that sticks out of the barrel. I'm not sure which I favor right now. 4 Aug 2002 Kevin Karplus Best scorer is try5-opt but it has not put in new beta strands, and it has destroyed the Hbonds G31-F48, G33-V46, I35-V44. Adding back in the 5 Hbonds (G31-F48 only had one in try5-al6), the best scorer is still try5-opt. The topology of the cold-shock proteins is a 5-strand beta barrel: 1^ 2v 3^ 5v 4^ with a parallel hbond connection back to strand 1. Some (but not all) cold-shock proteins have one turn of a 3-10 helix at the end of strand3. We have 6 predicted strands: T0140 1mjc M19-A28 M5-N13 (A14) G33-T36 F18-P23 2 longer--one on each side K42-F48 (KD)V30-F34 R66-T75 K50-E56 Note---this is much shorter strand. G78-L83 A63-S69 D88-H92 F48 has the FSA signal for the 3-10 helix at the end of strand 3. The 4-1 strand Hbonds ought to be M19-G69 G21-V67 V23-D65 Instead of 5-3 parallel Hbonds, maybe we have a simple anti-parallel barrel: 1^ 2v 3^ 6v 5^ 4v Let's try adding the 4-1 Hbonds and constraints for strand 6-strand 3. Strand 3 has F45 and H47 available for Hbonding. Let's try F45 to H92. With these constraints added, the best-scoring decoy is T0140.try5-al4+1mjc-T0140-local-adpstyle5.pw.a2m.gz:1mjc.9.80.pdb Let's do the optimization, starting over with no initial conformation. 5 Aug 2002 Kevin Karplus try6-opt-scwrl is the new best score. It does not look bad, except that the last strand has coiled up into a helix. Let's try straightening it our a bit, and adding in a few more Hbonds, and see what happens. The best changes try6-al6.8.80, so we'll start try7 from that and from try6-opt-scwrl, which scores second best. 8 aug 2002 Kevin Karplus New low cost is try7-opt-scwrl. It is almost forming a 6-strand barrel, but I think that we need to slide some of the hbonding of strands a bit to get better register and better turns. 10 Aug 2002 Kevin Karplus Let's try redoing the strand 3-6 interaction as F45-L90, H47-D88. This should help close the gaps at the base of the strand2-3 hairpin. T0140.try7-opt-scwrl still scores best. 11 Aug 2002 Kevin Karplus New best score is not try8-opt but try8.29.90 The structure formed is not a barrel, as the hairpin centered around G85 does NOT form hbonds to the strand K42-F48, but tucks into the center of what would be the barrel. Maybe we should give up on the barrel formation, and allow the hairpin to tuck in. Removing those strand3-6 constraints (making try9.constraints) and tweaking parameters slightly makes try8.15.90 score best, though it still has a bad break. 11 Aug 2002 Kevin Karplus Reran "make" with the new template library. No change among the top hits. 12 Aug 2002 Kevin Karplus try9-opt is the new low cost, and looks pretty good. A bit foamy in places, but with the hydrophobic mostly in the center The hairpin D84-D88, G82-L90 could be tighter, to make real hbonds. Adding CB (or CA) constraints makes try9-opt score best still. 13 Aug 2002 Kevin Karplus try10-opt is new lowest cost. It is foamy and the hairpin is still not forming Hbonds, but the hydrophobics are all on the inside. I should probably increase the dryness weights, and increase the probability for jigglesubtree and closegap, while decreasing the break penalty. With the changed score function and try11.constraints, the best scoring decoys are now try9-try8-15.37.80 and try10-opt. 14 Aug 2002 Kevin Karplus Best score now try11-opt. There is one bad break mid-strand at 80-81. Maybe it's time to increase the break penalty again. Doing so makes try11.10.90 be the best score. 15 Aug 2002 Kevin Karplus Best Score now try12-try11-30.27.90.pdb (not try12-opt) 16 August 2002 Kevin Karplus Best score is now try13.7.80 (not try13-opt). This is still pretty foamy and has some breaks. Upping the break penalty and modifying some constraints (now try14.constraints) makes try10.28.80 score best. I'll start a new run from that and from try13.7.80. Fri Aug 16 17:56:48 PDT 2002 Kevin Karplus Best is now T0140.try14.13.80.pdb Changes are getting to be pretty small, and further optimization does not always survive the writing out to PDB and reading back in, so I think I'll give up and submit T0140.try14.13.80.pdb.