Wed Jul 5 10:50:59 PDT 2006 T0357 Make started Wed Jul 5 10:52:14 PDT 2006 Running on lopez.cse.ucsc.edu Wed Jul 5 11:24:22 PDT 2006 Kevin Karplus BLAST finds no good hits in PDB, best are only 40% over 14 residues (E-value 0.7) 141 residues. Wed Jul 5 19:04:27 PDT 2006 Kevin Karplus Modest HMM hit to 1v7lA (4e-02) Wed Jul 5 19:49:43 PDT 2006 Kevin Karplus For some reason the try1.log file is incomplete---perhaps we were running the buggy version of undertaker from this morning. Let's try again. Wed Jul 5 19:53:15 PDT 2006 Kevin Karplus try1 started on cheep. Wed Jul 5 22:15:23 PDT 2006 Kevin Karplus I had to fix a bug in undertaker (not introduced today, but last weekend), before undertaker could handle empty alignments again. try1 ran successfully and actually looks reasonably good. It seems to be based on 2gwhA, though not too closely. The result is a bit foamy, but not ludicrous. We may need to look at making a multimer, though 2gwh seems to be monomeric. Mon Jul 17 17:03:45 PDT 2006 Kevin Karplus I must have been asleep---2gwh[AB] is clearly a dimer in PDB. The interface is so small, though, it might as well be a monomer (and PQS does classify it as 2 monomers in the unit cell). Perhaps I should submit try1 to VAST, to see if it is more similar to some other template. Mon Jul 17 17:14:05 PDT 2006 Kevin Karplus VAST Request ID: 333154454919829087 reports "no. neighbors" as zero. So this is not coming from an obvious single template. Make started Tue Jul 18 12:30:07 PDT 2006 Running on vashon.cse.ucsc.edu Make started Thu Jul 20 10:46:07 PDT 2006 Running on cheep.cse.ucsc.edu I removed the old "best-scores" files (moved them to old/) and am re-running the make so that the best-scores files get updated this time. When the make is done (it shouldn't take *too* long as all the alignments were updated a couple of days ago), I'll try doing the equivalent of try1 over, to see what comes up this time. The multiple alignments are similar, with the t06 and t04 actually seeming to have better conservation than the t2k. There are some Hbond predictions of varying strength: from n_sep D32.N I37.O I37.N D32.O T35.N D32.O C42.N K39.O G76.N R73.O V77.N L72.O V114.N V121.O E130.N V121.O from o_sep E15.N A8.O T10.N R13.O K39.N G30.O D32.N I37.O C42.N K39.O L54.N E47.O R52.N V49.O V77.N L72.O G76.N R73.O I100.N M97.O V103.N I100.O V121.N V114.O E130.N V121.O N123.N Y128.O V121.N E130.O Putting these together gives us # hairpins SheetConstraint A8 T10 E15 R13 hbond A8 # weak SheetConstraint G30 D32 K39 I37 hbond I37 # strong SheetConstraint E47 V49 L54 R52 hbond R52 # weak SheetConstraint V121 N123 E130 Y128 hbond Y128 # weak # C-cap motif Hbond G76.N R73.O Hbond V77.N L72.O Thu Jul 20 11:56:54 PDT 2006 Kevin Karplus try2 started on cheep. I did not include the sheetconstraints from the alignments in try2, and only put in a few rr constraints (weakened). I *did* include rewriting Template.atoms, because the set of templates was expanded by redoing best-scores, though the top hits are the same as before. Thu Jul 20 13:18:41 PDT 2006 Kevin Karplus Rosetta likes try2 better than try1, as does the unconstrained costfcn. We're getting the hairpins and a parallel sheet, but I think we'll have to add more constraints to get sheet(s) that include all the bits. For example, it looks like L102-E104 may be parallel to V121-N123. On closer examination, it looks like shifting it 2 would be better: SheetConstraint L102 V105 R119 V122 Hbond V120 I'll add that constraint to try3.costfcn I also picked up a parallel sheet from try2-opt2: SheetConstraint I53 P57 P79 I83 hbond L54 1 SheetConstraint A81 K85 P101 V105 hbond I82 1 Thu Jul 20 13:42:08 PDT 2006 Kevin Karplus I'll start two new runs with the try3 costfcn: try3 on cheep, starting from alignments. try4 on abyss, starting form random using only fragments, not alignments. Thu Jul 20 14:53:10 PDT 2006 Martin Madera My comments on the tries so far: - try1 has several strands predicted by t06-str2 turned into helices - try2 is better, but still 92-97 should be a strand, not a helix (though the str2 prediction does have a small helix probability, and in fact ehl2 has it as a weak helix) - try3 and try4 still running! Let me do the n_sep and o_sep constraints independently (based on t06), to make sure there are no typos: n_sep: D32 R = +5 *strong* E34 C = ?? mediocre I37 I = -5 weak C42 G = -3 weak G76 G = -3 mediocre V77 I = -5 *strong* D125 C = ?? weak E130 M = -9 *strong* (I see, C = multiple) o_sep: I37 R = -5 *strong* K39 G = +3 *strong* V49 G = +3 weak R73 G = +3 weak Y128 R = -5 weak E130 V = -9 weak which gives: D32.N ... I37.O S I37.N ... D32.O w C42.N ... K39.O w G76.N ... R73.O M V77.N ... L72.O S E130.N ... V121.O S I37.O ... D32.N S K39.O ... C42.N S V49.O ... R52.N w R73.O ... G76.N w Y128.O ... N123.N w E130.O ... V121.N w or (using n_sep,o_sep), K39.O ... C42.N w,S D32.N ... I37.O S,S D32.O ... I37.N w,- V49.O ... R52.N -,w L72.O ... V77.N S,- R73.O ... G76.N M,w V121.N ... E130.O -,w V121.O ... E130.N S,- N123.N ... Y128.O -,w while Kevin has: # hairpins SheetConstraint A8 T10 E15 R13 hbond A8 # weak SheetConstraint G30 D32 K39 I37 hbond I37 # strong SheetConstraint E47 V49 L54 R52 hbond R52 # weak SheetConstraint V121 N123 E130 Y128 hbond Y128 # weak # C-cap motif Hbond G76.N R73.O Hbond V77.N L72.O -------------------------------------------------------------------- so we pretty much agree, only in the constraints Kevin seems to have missed the strongly-predicted K39.O ... C42.N (strongly in o_sep, weakly in n_sep), and SheetConstraint V121 N123 E130 Y128 hbond Y128 # weak actually seems pretty well established. -------------------------------------------------------------------- Thu Jul 20 15:56:23 PDT 2006 Kevin Karplus Some disagreements: I didn't miss K39.O C42.N, but it isn't compatible with SheetConstraint G30 D32 K39 I37 hbond I37 # strong I looked at several different predictions for the 92-97 region, and decided I prefered the helical predictions to the strand predictions. It could be either way, of course, so it is worth looking for different solutions. try3 and try4 are now done, and rosetta likes try4 as a base for repacking sidechains, but undertaker still prefers try2. I like some of the sheet fragments in try3, but the meander from K22 to E43 looks dubious (though it is almost suggested by some rr predictions). The first hairpin has an off-by-one aligment, as the predicted hydrophobicity alternation is out of phase. This is the phase we requested based on the Hbond predictions, but I now believe it is wrong. I think we want SheetConstraint F4 A8 A18 A14 hbond ?? or perhaps SheetConstraint F4 A8 V20 G16 hbond ?? (I'm not sure which strand needs to flip over). We could probably ProteinShop try4-opt2 after C42 into a pretty nice alpha-beta sheet, but what comes before C42 is pretty trashy in try4. Thu Jul 20 17:58:53 PDT 2006 Martin Madera To see the sheet constraints: # green and brown are weak select 10,13 color brown select 32,37 color red select 49,52 color green select 121,130 color blue select * Looking at secondary structure assignments, there seems to be COMPLETE disagreement between str2 for T06 and T2k between 90 and 115: - T06 has: strand - strand - helix - T2k has: helix - - - - strand Both HMMs are basically junk from ~100 onwards, though T06 actually has a strongly conserved proline at P101. The two HMMs look pretty similar for 90-100, so I don't understand why str2 has such a different prediction for that part -- maybe that proline? But if you look at the actual sequence, it has FOUR alanines... hmmm, that sure smells like a helix! Fri Jul 21 10:51:13 PDT 2006 Kevin Karplus I'll do a "subdomain" prediction for P79-E133, even though I don't expect it to be a domain, to see if we can get a cleaner secondary structure prediction for this region, not contaminated by over-extended alignments. (make started on cheep) Fri Jul 21 15:08:47 PDT 2006 Martin Madera I quite like 1v7l as a template (NB the last letter is L, not number 1) -- compare that to try3-opt2! 1v7l is the top hit in T0357.best-scores.rdb and it's a sandwich consisting of: - an anti-parallel sheet - a parallel sheet - helices But I actually prefer the C-terminal part of try4 to that of try3. Actually no, I take that back, it made the His tag into the third strand of that small anti-parallel sheet, but I like the longer third helix for the parallel sheet. So I'll try assembling a complicated chimera of the good bits of try3 and try4, trying to form a sandwich a la 1v7l. Fri Jul 21 16:08:31 PDT 2006 Kevin Karplus The "domain" P79-E133 actually got quite consistent fold prediction. It may be worth copying in either constraints from there, or using its prediction as part of a chimera. Fri Jul 21 17:05:43 PDT 2006 Kevin Karplus I superimposed P79-E133 with the other models in best-models.pdb.gz It is pretty similar to the same region in try2 and try3. The region for I90-M97 is predicted to be helix by both str2 predictionsthe t2k and the t06 alignments for the "subdomain", though the dssp-ehl2 consensus is weakly for strand. The secondary structure predictions for the fragment are probably not very good, as there is little diversity in the sequences aligned. I think we need to resign ourselves to not knowing the secondary structure there. Sat Jul 22 08:40:51 PDT 2006 Martin Madera I put my best shot at the anti-parallel/parallel/helices sandwich into decoys/ProteinShop/edit3.pdb.gz (based on try3), and I've started two runs on it (try5 and try6) using the try3 cost function. Both running on lopez. I also started one more run from alignments a la try3 and try4 -- this is try8, running on peep. Sat Jul 22 16:51:16 PDT 2006 Martin Madera Oops, noticed I skipped try7. Also noticed that for some reason I actually failed to copy edit3.pdb into decoys/ProteinShop, so try5 and try6 didn't really work. Try8 is OK, it looks similar to try3 and try4 but isn't as well packed. But it shouldn't be too difficult to ProteinShop into something useful -- in fact probably easier than try3. Try5, try6 restarted as try9,try10. Sat Jul 22 19:36:33 PDT 2006 Martin Madera Try9 is the new best-scoring model according to the unconstrained cost function (marginally beating try2), but it got torn apart. Same for try10. Not surprising really; so I need to add some constraints: SheetConstraint I100 E104 D118 V122 hbond I100 SheetConstraint E133 V129 T10 C6 hbond R7 SheetConstraint K39 C42 G51 L54 hbond D41 I'm not sure about where to put the hbonds, hopefully something will come out. Try11 and try12 running. Sat Jul 22 21:24:06 PDT 2006 Kevin Karplus Martin has not left me anything here that I can submit tonight. Either I put in some time on it myself (but I have to pack still and catch a plane early in the morning), or I have Martin do the first submission. Maybe I'd better do a submission, even if with the wrong models, just so that there is a complete submission template for Martin to follow. The decoys/grep-best-rosetta file seems to be damaged, because there are empty repack-nonPC files. I'll delete the empty files, remake them, and remake grep-best-rosetta. The problem most likely results from simultaneous rosetta runs stepping on each other, but I'm not certain. Rosetta clearly likes try9-opt2.gromacs0.repack-nonPC best. I'll submit try2-opt2 best with try4,try6,try9,try10 costfcn try9-opt2.gromacs0.repack-nonPC best with rosetta try4-opt2 best with try1 costfcn try1-opt2 full auto align 1 t06 alignment to 1v7lA Sat Jul 22 21:47:52 PDT 2006 Kevin Karplus Preliminary submission done with comment: For this preliminary submission I chose 5 models: Model 1 is try2-opt2, which scores best with several of the cost functions we tried. It was optimized from alignments, including 1i9aA, 1v7lA, 1uarA, 2gwhA, 1p3rA, 1jtkA. Several of these alignments may have ended up being just short fragments. Model 2 is try9-opt2.gromacs0.repack-nonPC, which is try9-opt2, reoptimized by gromacs (which mainly removes small clashes), then with sidechains repacked by rosetta. It is rosetta's favorite of the models for which it repacked sidechains. Model 3 is try4-opt2, which scores best with our initial costfcn. Model 4 is try1-opt2, the fully automatic model. Model 5 is sidechain replacement by scwrl on a t06 alignment to 1v7lA, our top-scoring template. ------------------------------------------------------------ Mon Jul 24 19:14:06 PDT 2006 Kevin Karplus try11-opt2 now scores best with unconstrained costfcn (as well as try11.costfcn). For some reason the rosetta runs for them were not included in grep-best-rosetta. Ah, my fault---the grep-best-rosetta file did not have write permission. I remade it and fixed the permissions. Rosetta does not care much for either try11 or try12, though try12 does much better than try11. Tue Jul 25 13:28:07 PDT 2006 Kevin Karplus try11-opt2 is almost forming a barrel, should I push it to become one? I could add parallel sheet constraint SheetConstraint A5 I9 E130 L134 hbond R7 or SheetConstraint C6 I10 V129 E133 hbond R7 SheetConstraint L102 E104 R119 V121 hbond V120 Sheeconstraint K39 C42 R52 V55 hbond D41 Tue Jul 25 14:05:34 PDT 2006 Kevin Karplus Started try13 with these constraints added to the try11 costfcn. It will attempt to polish existing models, so it may get a bit stuck on try11-opt2. If so, it may be worthwhile backing up to edit3 or to try11-opt2.gromacs0. Tue Jul 25 14:21:48 PDT 2006 Kevin Karplus Typo in try13.under, trying it again! Tue Jul 25 14:52:10 PDT 2006 Kevin Karplus try13 does seem to be doing just minor tweaks to try11-opt2, so I think I will start again with the same costfcn but only edit3 as a starting model. try14 started on vashon. Tue Jul 25 19:06:41 PDT 2006 Kevin Karplus The try13=try14 costfcn like try13-opt2 best, as does the unconstrained costfcn, but rosetta likes decoys/T0357.try14-opt2.gromacs0.repack-nonPC.pdb almost as well as it likes the try9 file (its favorite at the moment). Tue Jul 25 19:18:57 PDT 2006 Kevin Karplus I haven't seen anything new from Martin, so I'm going to submit ReadConformPDB T0357.try13-opt2.pdb best unconstrained ReadConformPDB T0357.try9-opt2.gromacs0.repack-nonPC.pdb best robetta ReadConformPDB T0357.try2-opt2.pdb another good unconstrained ReadConformPDB T0357.try14-opt2.gromacs0.repack-nonPC.pdb 2nd best robetta, good unconstrained ReadConformPDB T0357.try1-opt2.pdb full auto None of these succeeded in making the barrel. Martin can submit another if he manages to get a better one by tomorrow morning. Tue Jul 25 19:25:23 PDT 2006 Kevin Karplus So submitted. I think there is a possibility of getting a better barrel by tweaking the costfcn, but I don't have time to do it tonight, and this target has a soft deadline of noon tomorrow (hard deadline 27 Jul). I started scoring the servers with the try13 costfcn on the farm cluster. I'm curious to see if any of ther servers are trying for a similar barrel. Wed Jul 26 05:41:17 PDT 2006 Kevin Karplus No, none of the servers score well with the try13 costfcn. I hope that Martin has more things to try on this target, as it doesn't look done to me. Wed Jul 26 19:12:04 PDT 2006 Kevin Karplus Seeing no more work by Martin, and a deadline 16 hours away, I'm going to declare this target done. If Martin adds something, he'll have to submit it himself.