Tue Aug 13 12:45:33 PDT 2002 t0181 No clear homology---fold-recognition or ab initio. Mon Sep 2 14:47:51 PDT 2002 Rachel Karchin The best scoring decoy from our first try, try1-opt-scwrl.pdb is not at all shabby. Like rosetta, undertaker is doing better on small proteins, which of course, makes sense. The STR prediction for this target shows a hairpin at the N-terminus, T5-K10 D15-F19 which is folding up very nicely in the decoy. A weakly predicted helix-turn-helix motif E24-L30 K31-S34 I35-L44 helix turn helix also shows up in the decoy. A predicted (antiparallel) strand F45-T49 is curling up into a piece of coil and a small piece of helix. This should be straightened with a length constraint. Where would the strand attach? Is it part of an antiparallel sheet with T5-K10 and D15-F19? Or does it pair with another tiny piece of predicted (antiparallel) strand L98-K99 which is curling into a helix in this decoy? We also have two more strongly predicted helices: K65-F72 I78-R87 K101-Y110 is predicted as coil and turn but is curling into a helix in the decoy. If it is a helix, it lacks the hydrophobic-polar alternating patterns observed in the other helices in this decoy. For a first try, the try1-opt-scwrl decoy is remarkably compact. Although Kevin writes above that there are no clear homologies, we do have some weakly good hits to this target. 1g4yR has E-value .17 to the STR model FSSP-rep is 1cll (calmodulin biological unit?) SCOP domain a.39.15 which is all-alpha FOLD is EF-hand-like -- motif of two helices connected with calcium binding loops. 1g4y is a structure that was complexed with calcium and sulfur. The R chain has two domains separated by a coil linker region and one of the domains has an anti-parallel beta sheet plus four helices arranged like this: H6 H4 E5 E2 H1 H3 The OTHER domain does the calcium-binding. This may be a clue to how to link up the putative anti-parallel strands in our target. with H1 = I35-L44 E2 = F45-T49 H3 = K65-F72 H4 = I78-R37 E5 = L98-K99 H6 = K101-Y110 ?? Ideas: generate a strand-heavy alignment for this target, generate an alignment only for the domain we're interested in. It is feasible that the target folds into a similar arrangment, Looking at the STR global alignments, 1g4yR I63-Y99 aligns strongly to E71-N111 of T0181 this piece is a helix in the OTHER domain, the linker coil, plus helix and short strand in the domain of interest. Aargh. In the STR heavy local alignments, the alignment is to G59-F89 in 1g4yR which is just the two helices adjacent to the linker plus the linker -- a helix-turn-helix piece. We're not picking up alignment to the sheet very well. Want an alignment for the region between 35-110 in the target. 1g4yR-2track-EBGHSTL-local-adpstyle1 aligns I35-I85 of the target to T79-I125 of 1g4yR This is alot more promising because it only aligns to the domain we're interested in and covers what I describe as H1 E2 H3 H4 above. More hits to look into . . . 1ehyA has E-value .25 " " 1fuA has E-value .38 " " More alignments. I've added STR-heavy to the basic build and ran this target over the webserver. It's favorite alignment is a 2-track STR heavy global alignment to 1sfe. 1sfe is SCOP domain c.55.71, specified as the domain from residues 12-92 "methylated DNA-protein cystein methyltransferase domain", a domain with a three-stranded anti-parallel sheet surrounded by two adjacent helices: H2 E3 E4 E5 H1 The entire thing is aligned to the target in our STR heavy global alignment. Alot of the global str-heavy alignments are in the best list, so rather than analyzing each one let's try remaking T0181 alignments, getting ALL the str-heavy alignments, and rerunning undertaker. (before adding constraints). Let's add a bunch more alignments, for many of the best templates used in the STR-heavy global alignments: 1sfe 1fjgI 1hruA 1ehyA 1a6q try2 is started. Tue Sep 3 10:19:56 PDT 2002 Kevin Karplus try2-opt currently best score, outperforming robetta1, which was based on 1e8aA (a.39.1.2), same superfamily as several of our top hits, but a different family. Maybe we should toss 1e8aA or 1b4cA (its FSSP representative) into the pot of alignments to consider. Note: though Robetta was based on 1e8aA, it has added a long hairpin which is NOT from 1e8aA. Maybe we should try polishing up the robetta1.pdb model also. Tue Sep 3 12:42:44 PDT 2002 Rachel Karchin best scoring decoys: T0181.try2.20.60.pdb putative strands F45-T49 and L98-K99 are still curled up, but now in closer proximity to each other. T0181.try2-opt.pdb try3 alignments added for 1e8aA and 1b4cA starting conformations will include try2-opt and robetta1.pdb added straightness constraints for F45-T49 and L98-K99 Thu Sep 5 12:53:21 PDT 2002 Rachel Karchin The try3 decoys aren't scoring well overall, but try3-opt-scwrl succeeds in straightening out F45-T49. The piece at L98-K99 is not straight, but has broken off from the subsequent piece K101-D104, which is still winding up into a helix, although predicted as turn. This prediction is very weak. F45-T49 could form a mixed sheet with the hairpin T5-K10 and D15-F19, it could be parallel to D15-F19. Another possibility is that the helix from E24-L44 could twist across the sheet and then F45-T49 could attach to T5-K10. Let's try assembling a sheet with F45-T49 anti-parallel to T5-K10. 44< qpTFVEFl 4> vTKYFYKg 15< aFVILDt hairpin of 4> and 15< is from try3-opt-scwrl try4 begins with this conformation: try3-opt-scwrl try4 did not succeed in moving F45-T49. I'm going to try breaking off the entire strand-helix piece E23-T49 and move it to the other side of the hairpin. try5 is complete garbage. Looks like the undertaker script was missing a "\" and failed. I haven't succeeded in getting the n-terminal hairpin on the inside and the helix-turn-helix from E24-L44 on the outside, nor in finding a place for the F45-T49 strand to pair. try6 -- just attempt to get the N-terminal hairpin to pack into the center of the protein and add some constraints, where? starting conformation will be our current best scoring conformatin T0181.try2-opt.pdb The hairpin has charge around the turn area and hydrophobics around its base. The base could possibly pack this way: F8 and L17 near the hydrophobic face of helix I35-L44 perhaps V41 and F45 (use weak constraints) requires moving out the coil piece from 49-61, so starting conformation for try6 is T0181.try2-opt-moved.pdb Sat Sep 7 10:18:08 PDT 2002 Rachel Karchin try6 failed because of typo in define-score.script rerunning it now Sat Sep 7 19:23:13 PDT 2002 Rachel Karchin try6 results: best scoring decoy with current constraints (try6-constraints) is T0181.try2.20.60.pdb and T0181.try2-opt.pdb These are also best with no constraints (makes sense because try6-constraints are very weak). try6-opt is way down in the list and has NOT succeeded in packing the hairpin against helix I35-44. I may be choosing a bogus orientation. best scoring with try3-constraints (straightness constraints for predicted strands) is the same T0181.try2.20.60.pdb and T0181.try2-opt.pdb. if we crank up constraints to 0.1 (from 0.05) best scoring with try3-constraints (same as above). . . . with try3-constraints AND try6-constraints (same). These two favorite decoys don't look very good to me, they have breaks and are not compact. From: Rachel Karchin Date: Sat, 7 Sep 2002 19:34:02 -0700 To: karplus@soe.ucsc.edu Subject: T0181 not making progress Please feel welcome to look at T0181. It has a hairpin sticking out into (what I assume is) solvent, and I've been trying to figure out how to navigate it back into the core of the protein, by packing it against the hydrophobic face of one of the helices. I am getting nowhere -- perhaps my constraints are poorly chosen. Details of what I've been trying to do are in the README. -------------------------------------------------- 7 Sept 2002 Kevin Karplus 21:50 Took a quick look at the -opt files. It looks like try3-opt and try4-opt are currently the most promising. I have to do some debugging in undertaker now (it keeps crashing), but I'll look at the constraints in more detail in a while. 22:38 (still debugging, but I have to wait for the program to die again) I think that the pair of helices E24-Y29 and L38-E43 have a turn between them, so you end up near where you started. This suggests to me that strand F45-T50 is actually antiparallel to D15-A20. It could end up parallel to T5-Y10, but we have a suggestion in the str predictions that D15-A20 is a middle strand rather than an edge strand, while the other two strands both favor being edge strands. Let me try this alignment: 1> mstvTKYFYKge 24< eesaAFVILDtn 40> evvelFEVFTpqd try7 will try using all alignments, try8 will start with just a random conformation. I'll take the best model(s) and do a longer run with them afterwards. The first few iterations of try7 are having no trouble forming the sheet, but are having trouble getting the two helices to meet up. I'll probably have to turn break up when I try to improve on the results of try7 and try8. 8 Sept 2002 00:40 Kevin Karplus try8-opt-scwrl looks pretty good, except for a bad break. try7-opt-scwrl doesn't look as good, but has some potential. I'll start try9 to try to clean them up. 8 Sept 2002 08:12 Kevin Karplus try9-opt scores best, but still has the bad break of try8. Perhaps if I moved A20-Y29 out (moving A21 away from V41), it would be able to reform and repack A20-V41, fixing the bad break at V41-V42 I'll try that for try10. Other than the bad break, try9 looks pretty good, so I hope it is fixable. 9:42 try10 fixed up the break, but unwound the helices more than I'd like. try11 will try the same thing as try10, but with more superiterations, to sample a larger space. try12 will try to polish try10-opt. 11:49. try12 worked better than try11, but neither did a great job---perhaps I need some helix length constraints for the helices that are having trouble. I set this up in try13.constraints. 13:17 try13 is beginning to get the two helices wound back up, but has E28 stuck in the wrong place---perhaps we should add the Hbond constraints for the center of the helix. 15:27 try14-opt has introduced a bad break between v41 and v42. 18:25 Sigh, try16 has removed strand F45-T49 from the sheet. Maybe it wants it on the other side of the sheet. Let's try that for try 17 // 40> evvelFEVFTpqd // 1> mstvTKYFYKge // 24< eesaAFVILDtn 20:34 No try17 seems to make an even worse mess than the earlier ones. Let's go back to try14.constraints and reoptimize. 22:00 Even with the try14 constraints, the score function likes making the strand cross the other two at right angles (try18-opt,try16-opt). The score function may like this foam, but I don't. If we go back to the try13 constraints, which did not have the big benefits for getting the helix just right, we move try12-opt to the top. Let's re-optimize with the try13 constraints. 9 Sept 2002 Kevin Karplus Made try20.constraints with Hbonds added to try13 constraints. Made try21.constraints with extra fourth strand. Best score with try21 constraints is try20-opt, not try21-opt---what improvement was made got lost in re-reading the input. Trying try22 with more rearrangment and less OptSubtree. 15:17 try22-opt looks slightly better, but still has problems with L25-V42. Let's break out just Y29-P33, moving Y29 a couple of Ang closer to A63. 17:04 Hmm, try23-opt has put the broken-off pieces back where they came from, so the best score is still for try22-opt. With try21.constraints: try22-opt, try20-opt, try23-opt, try19-opt With no constraints: try22-opt, try20-opt, try19-opt, try23-opt, try12-opt With try13.constraints: try22-opt, try20-opt, try19-opt, try23-opt, try12-opt With try14.constraints: try18-opt, try16-opt, try22-opt, try20-opt, try19-opt With try17.constraints: try18-opt, try16-opt, try17-opt, try22-opt with try3-constraints: try22-opt, try20-opt, try19-opt, try23-opt, try12-opt submit: model 1: try22-opt model 2: try18-opt