Tue May 20 01:24:17 PDT 2008 T0407 Make started Tue May 20 01:24:42 PDT 2008 Running on cheep.cse.ucsc.edu Tue May 20 01:26:56 PDT 2008 Kevin Karplus This may be a multi-domain protein. BLAST finds a good match for 39-171 to 2anuA, but the C-terminal domain may be harder to predict. Tue May 20 22:12:49 PDT 2008 Kevin Karplus strong hits to 2hnhA, 1m65A, 2yxoA, and 2anuA. These may even be full-length matches. Fri May 23 04:24:17 PDT 2008 Kevin Karplus I get homology modeling for about N26-L292, but the N and C-termini are not copied from templates. try1-opt3 wants to put the two termini together to form a domain, which seems reasonable, but what it is producing is badly packed foam, with some of the predicted-buried residues on the outside (M1,L12,F344,V346). Make started Mon May 26 12:13:20 PDT 2008 Running on orcas.cse.ucsc.edu It looks like John did a "make" here by mistake, but there doesn't seem to be any harm done. The MQA runs selected the Zhang-Server and the BAKER-ROBETTA models as best though the MQAU function likes SAM-T08-server also. I'll make meta-server runs for both MQAC and MQAU, just using the try1 costfcn. Mon May 26 17:17:11 PDT 2008 Kevin Karplus MQAU1-opt3 (from BAKER-ROBETTA_TS3) scores best, then MQAC1-opt3 (also from BAKER-ROBETTA_TS3). There is vague agreement for Y36-I219, but less agreement after that. I should proably make a subdomain prediction for I219-L363. I'll also make one for M1-H230. Mon May 26 21:16:57 PDT 2008 Kevin Karplus M1-H230 is getting good matches to c.6.3.1 domains (2yxoA, 2anuA, 1m65A), the 7-stranded beta barrel. I219-L363 gets no good matches (best e-value 28). Wed May 28 11:13:49 PDT 2008 SAM-T08-MQAO hand QA T0407 Submitted Wed May 28 11:13:49 PDT 2008 SAM-T08-MQAU hand QA T0407 Submitted Wed May 28 11:13:49 PDT 2008 SAM-T08-MQAC hand QA T0407 Submitted Sat May 31 19:23:28 PDT 2008 Kevin Karplus I'm getting pretty solid prediction for L39-H230, so I should probably do subdomain runs for P31-H260 and R241-L363. Sat May 31 19:34:14 PDT 2008 Kevin Karplus subdomain runs started. Question: are H44, H46, E113 a conserved metal-binding site? (D214 may also be involved.) I should look at 2anuA or 1m65A to see. Mon Jun 2 12:30:07 PDT 2008 Kevin Karplus Looking at P31-H260/best-models.pdb.gz The complete try1-opt3 model and the alignments are in pretty good agreement, but K227-I246 looks a bit messed up in try1-opt3. There is a bit of sheet involving M233-F235 and L39-D42 in some of the alignments that looks like it may be worth preserving, as well as a helix from S242 to R253. There are some horrendous breaks bfore M233 and V86, and some other bad ones (before G105, D51, T168, M183, ... ). I think that some gap closing and cleaning up will be needed even for this homology model. Looking at R241-L363, there is less clear evidence for homology, since this region of the protein appears to be an ORFan. The R241-L363/try1-opt3 run is marginally plausible, and we don't have much to go on. Tue Jun 3 19:33:00 PDT 2008 Kevin Karplus Started P31-H260/try2 on cheep, trying to make a cleaner homology model from alignments. I'll have to do a try3 polishing run when it is done. Maybe I should send the R241-L363 models to VAST, to see what templates to concentrate on. Tue Jun 3 22:37:34 PDT 2008 Kevin Karplus P31-H260/try2-opt3 scores better than P31-H260/try1-opt3, but still needs some break closing and clash removal. It does not fit the align_constraints and try1.sheets as well as try1 doesn, but it does get the extra strand I tried to force. One of the strands has been reversed between P31-H260/try2-opt3 and P31-H260/try1-opt3, and I think I like the try2-opt3 topology better. Let me do another run, with try2.sheets, instead of try1.sheets. Wed Jun 4 07:30:49 PDT 2008 Kevin Karplus P31-H260/try3-opt3 scores better than P31-H260/try2-opt3, even though the try2.sheets score is substantially worse. The breaks are bad, but not AS bad, and clashes are way down. Here are the P31-H260/try2 sheets SheetConstraint (T0407)T38 (T0407)C41 (T0407)F237 (T0407)T234 hbond (T0407)L39 1 SheetConstraint (T0407)D214 (T0407)H216 (T0407)H44 (T0407)H46 hbond (T0407)I215 1 SheetConstraint (T0407)T208 (T0407)I210 (T0407)G185 (T0407)E187 hbond (T0407)I210 1 SheetConstraint (T0407)F154 (T0407)N156 (T0407)G185 (T0407)E187 hbond (T0407)W155 1 SheetConstraint (T0407)H122 (T0407)F127 (T0407)N156 (T0407)A151 hbond (T0407)N124 1 SheetConstraint (T0407)K110 (T0407)A117 (T0407)F127 (T0407)P120 hbond (T0407)S112 1 SheetConstraint (T0407)L107 (T0407)I114 (T0407)A70 (T0407)I77 hbond (T0407)E113 1 SheetConstraint (T0407)K40 (T0407)F43 (T0407)D69 (T0407)S72 hbond (T0407)D42 1 and the P31-H260/try3 sheets SheetConstraint (T0407)Y36 (T0407)C41 (T0407)R231 (T0407)E226 hbond (T0407)T37 1 different SheetConstraint (T0407)D214 (T0407)I215 (T0407)H44 (T0407)M45 hbond (T0407)I215 1 shorter SheetConstraint (T0407)G211 (T0407)S213 (T0407)A189 (T0407)G191 hbond (T0407)T212 1 new SheetConstraint (T0407)T208 (T0407)I210 (T0407)G185 (T0407)E187 hbond (T0407)I210 1 same SheetConstraint (T0407)F154 (T0407)N156 (T0407)G185 (T0407)E187 hbond (T0407)W155 1 same SheetConstraint (T0407)H122 (T0407)F127 (T0407)N156 (T0407)A151 hbond (T0407)N124 1 same SheetConstraint (T0407)G111 (T0407)A117 (T0407)I126 (T0407)P120 hbond (T0407)S112 1 shorter SheetConstraint (T0407)L107 (T0407)I109 (T0407)A70 (T0407)S72 hbond (T0407)I109 1 shorter SheetConstraint (T0407)C41 (T0407)F43 (T0407)A70 (T0407)S72 hbond (T0407)D42 1 shorter The try3 sheets make more sense with the secondary structure prediction. Let's do another run from alignments with try3.sheets constraints (and a bit more weight for breaks and for secondary structure prediction). Wed Jun 4 12:14:17 PDT 2008 Kevin Karplus Doing a polishing run in P31-H260 (try5). I think that this domain is about as good as it is going to get. The hard part is for H260-L363. In R241-L363, the most convincing alignment is T0407-1dunA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5, which is fold b.85.4.1. The alignment starts at R285 Perhaps I should do another subdomain with a smaller range: say A256-L363. If it picks up b.85 folds, then I'll focus on them. Wed Jun 4 12:25:02 PDT 2008 Kevin Karplus Started on moai cluster. Wed Jun 4 17:42:15 PDT 2008 Kevin Karplus b.85.* does come up several times in A256-L363/best-scores, but do does b.1.* The three top scores (best having E-value 56) don't occur in SCOP. 2p22D pfam09452: Mvb12 no other structures (per VAST) 1y03A just a helix (peptide) 2ol1A dUTPase superfamily (most similar to b.85.4.1 domains) So it looks like b.85.4.1 is one of the top hits anyway. Thu Jun 5 10:13:54 PDT 2008 Kevin Karplus In A256-L363/Makefile, I added MANUAL_TOP_HITS:= 2ol1A 1dunA 2bsyA 1q5hA 1f7dA 1euwA \ 2ftsA 2nqrA 1g8lA \ 1vliA 1xuuA for all the b.85.* folds that came up in any of the searches with any of the subdomains. The first six are b.85.4.1 family, and are my current favorites. For A256-L363/try2, I'll start from just these alignments. Thu Jun 5 17:53:46 PDT 2008 Kevin Karplus The A256-L363/try2 run does not get the predicted helix, but the burial looks good. I'm tempted to try polishing it to close gaps and improve the sheets. Perhaps I should first look to see what happens if I try superimposing it on the best P31-H260 model (try5). Fri Jun 6 03:04:13 PDT 2008 Kevin Karplus The A256-L363/try3 run chose to polish the try1 rather than the try2 series. I'll do a try4 that uses only the try2 starting points, but then I'll have to put together more complete models using both possible domains. Fri Jun 6 04:23:30 PDT 2008 Kevin Karplus The A256-L363/try4 run does not really succeed in closing the gaps in the try2 model, so the try3-opt3 model actually still looks better. Next step will be to try to combine models. Fri Jun 6 05:03:09 PDT 2008 Kevin Karplus I made chimera1 and chimera2: chimera1: From MQAU1-opt3 (for M1-V30), P31-H260/try5-opt3 (for P31-Y258) A256-L363/try3-opt3 (for F259-L363) chimera2: From MQAU1-opt3 (for M1-V30), P31-H260/try5-opt3 (for P31-A256) A256-L363/try4-opt3 (for A257-L363) The try2 run will try to optimize chimera1, and the try3 run will try to optimize chimera2. Fri Jun 6 09:32:03 PDT 2008 Kevin Karplus Both the try2 and the try3 runs managed to reduce breaks and clashes to a reasonable level, but not as much as the MQAC1 and MQAU1 models. More polishing with bigger break and clash penalties would be needed to reach that level of cleanliness. The try2 and try3 costfcns have enough break and clash weight to prefer MQAC1-opt3 and MQAU1-opt3, but after that favor themselves, though the try2 cotfcn ends up with a lower cost for try2-opt3 than try3 does for try3-opt3. I definitely like try2 better than try3, not only for the lower breaks, but also for the better burial of conserved I350 and V346. So I'll do a polishing run of the try2 models. Fri Jun 6 14:20:26 PDT 2008 Kevin Karplus try4 reduces breaks a lot, but still doesn't eliminate them. Let's try once more, with the break weight doubled, and clashes turned up slightly. I think we can now do this from all models. try5 started on peep. Fri Jun 6 17:20:09 PDT 2008 Kevin Karplus try5 reduces breaks and clashes, but doesn't eliminate them. Fri Jun 6 18:15:29 PDT 2008 Kevin Karplus Currently, I have 4 distinct models: ReadConformPDB T0407.try5-opt3.pdb ReadConformPDB T0407.MQAU1-opt3.pdb ReadConformPDB T0407.try3-opt3.gromacs0.pdb ReadConformPDB T0407.try1-opt3.gromacs0.repack-nonPC.pdb If I can choose a 5th model, I'll submit, despite the remaining breaks and clashes. Sat Jun 7 08:58:58 PDT 2008 Kevin Karplus I think that for the 5th model, I'll try to tweak up BAKER-ROBETTA_TS5, which is the top-scoring model with both MQAU and MQAC. Sat Jun 7 12:05:57 PDT 2008 Kevin Karplus The try6-opt3 model, from BAKER-ROBETTA_TS5, now scores 3rd, after MQAC1-opt3 and MQAU1-opt3. Since this was originally a Rosetta model, I'll submit the try6-opt3.gromacs0.repack-nonPC version, which rosetta likes best of the models it has repacked. The main effect of the try6 optimization was to change the relationship between the two domains, making the protein a bit more compact than in the BAKER-ROBETTA_TS5 model. Before submitting, I'll do one more optimization from all models, with soft-clashes turned up. That will probably tweak up the try6 model some more. Sat Jun 7 12:29:19 PDT 2008 Kevin Karplus I also started a try8 run with constraints for the try5 helices and sheets, to tweak up the try5 models. Sat Jun 7 14:27:14 PDT 2008 Kevin Karplus try7-opt3 seems to have been derived from MQAC1-opt3, rather than from try6. I may want to do another run starting from all models, excluding MQA models and try7 models. I now have try8 from try5 < try4 < try2 < chimera1 try9 from try6 from BAKER_ROBETTA_TS5 try7 from MQAC1-opt3 from BAKER-ROBETTA_TS3 try3 from chimera2 try1 full auto I don't think I'll get much more out of tweaking these models. The first domain is probably right and all the second domans are probably wrong. I'll submit ReadConformPDB T0407.try8-opt3.pdb ReadConformPDB T0407.try9-opt3.pdb ReadConformPDB T0407.try7-opt3.gromacs0.repack-nonPC.pdb ReadConformPDB T0407.try3-opt3.gromacs0.pdb ReadConformPDB T0407.try1-opt3.gromacs0.repack-nonPC.pdb