Tue May 27 09:45:47 PDT 2008 T0419 Make started Tue May 27 09:46:38 PDT 2008 Running on cheep.cse.ucsc.edu Tue May 27 13:11:06 PDT 2008 Kevin Karplus I seem to be getting very weak hits to c.37.1.* Tue May 27 15:11:24 PDT 2008 Kevin Karplus The hits to c.37.1.1 (nuceotide and nucleoside kinases) seem pretty consistent, and I get an e-value of 2.7e-03 for 1p5zB, so this seems pretty reasonable. The PPK2 name is polyphosphate kinase, and Pfam doesn't know of any structures, but the c.37.1.1 family seems entirely appropriate for the function of the protein. The crystallographers say that this is a dimer, which may be slow to model with such a big protein. Wed May 28 09:23:30 PDT 2008 Kevin Karplus try1-opt3 is forming a bit of a beta-sheet core that looks ok, but there are a lot of floppy bits sitting around. Packing the whole core may be tricky. Note T0421 may be a related protein, so guidance from it may be useful. Sun Jun 8 11:06:26 PDT 2008 SAM-T08-MQAO hand QA T0419 Submitted Sun Jun 8 11:06:26 PDT 2008 SAM-T08-MQAU hand QA T0419 Submitted Sun Jun 8 11:06:26 PDT 2008 SAM-T08-MQAC hand QA T0419 Submitted Mon Jun 9 20:15:53 PDT 2008 Kevin Karplus The try1 costfcn prefers MQAU1-opt3.repack-nonPC MQAC1-opt3.gromacs0 try1-opt3.repack-nonPC None of these looks very convincing, but there seem to be a number of sheet relationships. I wonder how compatible they are, and whether we can paste together sheets out of sets of constraints. Fri Jun 13 18:16:41 PDT 2008 Kevin Karplus try2-opt3 is no better than previous models. I think I like MQAC1-opt3 best for the sheets, so I'll try a run from alignments just those sheets as constraints. I should also check to see if the MQAC1 model has clear templates using VAST, and perhaps try those. VAST Request ID: 478741695492769515 I should also do three separate domain predictions: M1-R243 K238-H297 K292-K496 Fri Jun 13 18:29:12 PDT 2008 Kevin Karplus subdomains started on moai cluster. Fri Jun 13 18:52:30 PDT 2008 Kevin Karplus The whole chain for MQAC1-opt3 does not get any VAST hits, but I do get hits on 2 domains: 1-270: 2zi3A 1oe0A 1g3uA 2a2zB 2ocpA 1nmyA 2zi4A ... 292-496: 2zi3A 1oe0A 1g3uA 2ocpA 2a2zB 2zi4A 1p60A 1nmyA ... 1oe0A=2vp4A (our 13th hit) 1g3uA=1gtvA (our 4th hit) 2a2zB=2a30A 2zi4A=2zi6A 1p60A=1p5zB (our top hit) So it looks like MQAC1-opt3 is finidng the same templates we are, but is aligning them twice, which I missed. I'll revisit this target after the subdomains have been run. Sat Jun 14 10:22:17 PDT 2008 Kevin Karplus M1-R243 gets good hits to the c.37.1.1 domains, as expected. The N-terminal helix (up to A38) is not taken from the top few alignments, so its location is a bit dubious. K292-K496 also has good hits to c.37.1.1 domains. K238-H297 has no good hits and really has little more than secondary structure prediction to guide it. For try4, I'll try making predictions from the alignments in each of the subdomains (as well as alignments for the whole chain), using constraints taken from M1-R243 and K292-K496. With the try4 costfcn, MQAC1-opt3.gromacs0 is scoring the best currently---I hope that try4 manages to construct a model that is at least as good. Sat Jun 14 20:52:27 PDT 2008 Kevin Karplus try4-opt3.gromacs0.repack-nonPC is not scoring quite as well as MQAC1-opt3.gromacs0, but is pretty close. Reducing breaks and clashes would probably help. Note: phobic_fit is quite large for the try4-opt3 models, but the K292-K496 sheets and helices seem good. The linker domain and the N-terminal helix are sticking out in awkward ways. I should probably look for a dimeric template, to get constraints for packing the two domains. The first one I came to in the M1-R243 best-scores list is 1p5zB. I could load the two monomers into the template library with different names, and use alignments to the M1-R243 domain and the K292-K496 domain to make an incomplete model of the two domains. I'm not sure how best to convert that incomplete model into a complete one---perhaps grab a number of constraints from it? Insert it (with TryAllAlign) into other models?? Sun Jun 15 09:19:04 PDT 2008 Kevin Karplus George was trying to fix the rr constraints for subdomains in K292-K496 and did a "make" there that did not complete, probably because his ssh setup is not quite right, but possibly because I had run some programs without doing a fixmode. In any case, the K292-K496/summary.html file was left incomplete. I'll remove the rr predictions for K292-K496 and redo the make for George. Thu Jun 19 16:44:33 PDT 2008 Kevin Karplus I've been having some trouble figuring out how to paste together separate domains on top of a dimeric template. It is obviously something useful, but undertaker does not support it well. I can do something like reading in alignments for a dimer ReadTemplatePDB 1p5z.mmol chain A name 1p5zA ReadTemplatePDB 1p5z.mmol chain B name 1p5zB Then I can make an a2m file that aligns only the first domain to 1p5zA and only the second domain to 1p5zB. Making a dimer of some model that sort of has both domains PrintMultimerPDB 2 try4-try4-1p5z.pdb would put the two domains in roughly the right orientation, and I could cut and past to make a try4-try4 chimera. I could then superimpose the subdomain models on this chimera, and paste them into a new chimera. Thu Jun 19 17:08:57 PDT 2008 Kevin Karplus Well, I got as far as making try4-try4-1p5z.pdb, but it turns out that the dimerization interface in 1p5z consists of a lot of helices that just don't seem to be present in these domains, so the two domains end up a long way apart---not very useful. Thu Jun 19 17:11:43 PDT 2008 Kevin Karplus My next random thing to try is to superimpose the subdomain try1-opt3 models with the try4 model and try making a chimera. I'll take M1-T237 from M1-R243/try1 K238-P249 from try4 L250-D291 from K238-H297/try1 K292-K496 from K292-K496/try1 Thu Jun 19 17:34:11 PDT 2008 Kevin Karplus This chimera-N1-try4-M1-C1.pdb has terrible clashes, but I'll try optimizing it with a copy of the try4 costfcn anyway, to see what happens. Fri Jun 20 11:19:37 PDT 2008 Kevin Karplus The try5 model scores better than MQAC1 with the try5 costfcn, but phobic fit is still large, because of A241-D291 and (to a lesser extent) M1-A38 sticking out. Actually M1-A38 seems to stick out less in try5 than in other models I'm considering (MQAC1, MQAU1, ...), but A241-D291 is perhaps worst in try5. Sat Jun 21 10:04:39 PDT 2008 Kevin Karplus For try6, I "exploded" try5-opt3, moving the two domains far apart and breaking the linker in between into two pieces. I'll optimize starting from just try5-opt3.exploded to see if I can assemble a more compact pair of domains. Sat Jun 21 14:01:27 PDT 2008 Kevin Karplus try6-opt2 has unclosable gaps at either end of the helix at R24-P249. Even moving that fragment to a better location would not help much. Sat Jun 21 17:05:45 PDT 2008 Kevin Karplus try6-opt3 still has the R240-P249 bit out in the middle of nowhere, despite the huge break cost it causes. I wonder why the gap is never closed? For that matter, there are other (not quite so bad) breaks that aren't being closed either. Why not? I think I'll try doing a shorter polishing run on the try6 models, with a costfcn that turns down the constraints and turns up the breaks. Sat Jun 21 21:26:27 PDT 2008 Kevin Karplus Something is really wrong with undertaker---try7-opt3 has the subdomains way far apart. Undertaker should have been able to bring them somewhat closer. What are possible problems? 1) spanning tree is buggy so subtrees are not compact, making it impossible to move a domain as a unit? 2) bug in Transform operator? 3) OptSubtree has stupid idea about "ideal" locations when one subtree is badly oriented. Oops, it is none of the above---the try7-opt3 model is not so loosely packed---gromacs pushed the domains far apart. There *are* still terrible breaks in try7-op3, but these look more like the sort that can't be closed except by overshooting and coming back from a different angle. Sat Jun 21 21:52:29 PDT 2008 Kevin Karplus I tried tweaking try7 by moving the stray fragment R240-P249 somewhat closer to where it needed to go. Sun Jun 22 08:22:23 PDT 2008 Kevin Karplus try8 comes closer to closing breaks, but still has some bad ones. It scores almost as well as MQAC1 with the try8 costfcn. For try9, I'll do a polishing run from all the models that have 2 domains (except the MQAC models, which should be separately polished). Sun Jun 22 08:29:20 PDT 2008 Kevin Karplus try9 started as a polishing run for try8 (with try8 sheet constraints) try10 started as a polishing run for MQAC1 (with MQAC1 sheet constraints) Perhaps I should also do polishing runs for the subdomains K292-K496 and M1-R243, to try to get better-closed models of each subdomain seperately. Sun Jun 22 08:39:55 PDT 2008 Kevin Karplus M1-R243/try2 and K292-K496/try2 started on moai cluster. Sun Jun 22 10:33:44 PDT 2008 Kevin Karplus try9 and try10 both seem to do a decent job of polishing their respective models, though even the try9 costfcn prefers try10 to try9, mainly because try9 still has some bad breaks, particularly before W444 and D261. try10 still has a bad break before L76. When the subdomain try2 models have finished, I'll try superimposing them on both try9 and try10, and see if I can make a better chimeric model using them. Sun Jun 22 10:48:36 PDT 2008 Kevin Karplus It looks like the try2 subdomain models close gaps better than the try1 models, though W181-V186 remains a problem in the M1-R243/try2 models. I think I should use M1-R243/try2-opt3.gromacs0.repack-nonPC as the breaks are somewhat less in that model, though it might be better to try cutting P170-E185 from some other model entirely, and try to redo this loop. The K292-K496/try2 models still have bad breaks before W444, K473, and K443, but the gromacs0 version reduces the breaks somewhat (by stretching the backbone). It might be good to find a model in which W438-Q446 is better, and patch it in. For the try9-N2-C2 chimera, I'll take M1-H172 V186-L235 from M1-R243/try2-opt3.gromacs0.repack-nonPC S298-K496 from K292-K496/try2-opt3.gromacs0.repack-nonPC the rest from try9-opt3 I'll then try patching in a fragment from try4-opt3 or try10-opt3 near P170-E185, and a fragment from try10-opt3 around E424-K443. The try10 chimera will be harder, since the models are not closely related. I'l take M1-F40 fom try10-opt3, P41-D169 from M1-R243/try2-opt3.gromacs0.repack-nonPC P170-S184 from try10-opt3 E185-A236 from M1-R243/try2-opt3.gromacs0.repack-nonPC T237-Q296 from try10-opt3 H297-R423 from K292-K496/try2-opt3.gromacs0.repack-nonPC E424-K443 from try10-opt3 W444-K496 from K292-K496/try2-opt3.gromacs0.repack-nonPC I think that the try10-M2-C2 chimera will be hard to optimize, because of bad clashes between the domains. For chimera-try9-M2-C2-try10, I'll take chimera-M2-C2 and patch in K168-V186 and E424-D445 from try10-opt3 Sun Jun 22 12:17:17 PDT 2008 Kevin Karplus try11 will try to fix up try9-M2-C2-try10 try12 will try to fix up try10-M2-C2 If I had a lot more time (and no other targets to work on), I'd try to do more fixing of the individual domains, since I think I can see additional strands that should be attached to the sheets: P207-E212 and P464-V468 Sun Jun 22 13:51:15 PDT 2008 Kevin Karplus try11-opt3.repack-nonPC (based on try9-M2-C2-try10) is now the top scoring model with the try11=try12 costfcn, finally beating out try10. try12 has not finished running yet, but I don't expect it to be very successful. Sun Jun 22 19:32:40 PDT 2008 Kevin Karplus try12-opt3.gromacs0.repack-nonPC scores better than MQAC1-opt3.gromacs0, though not as well as try11-opt3 Now I need to pick out 5 models that are worth submitting. costfcn top 6 distinct models try1 MQAU1-opt3.repack-nonPC try10-opt3 MQAC1-opt3 try12-opt3.gromacs0.repack-nonPC try1-opt3.repack-nonPC try5-opt3.repack-nonPC try2 try10-opt3 MQAC1-opt3.gromacs0.repack-nonPC MQAU1-opt3.repack-nonPC try11-opt3.gromacs0.repack-nonPC try12-opt3.gromacs0.repack-nonPC try9-opt3.gromacs0.repack-nonPC try3 try10-opt3 MQAC1-opt3.gromacs0.repack-nonPC try11-opt3.gromacs0.repack-nonPC try12-opt3.gromacs0.repack-nonPC MQAU1-opt3.repack-nonPC try9-opt3.gromacs0.repack-nonPC try4 try11-opt3.repack-nonPC try12-opt3.gromacs0.repack-nonPC try9-opt3.gromacs0.repack-nonPC try5-opt3.repack-nonPC try8-opt3.gromacs0.repack-nonPC try10-opt3 try5 try11-opt3.repack-nonPC try12-opt3.gromacs0.repack-nonPC try9-opt3.gromacs0.repack-nonPC try5-opt3.repack-nonPC try8-opt3.gromacs0.repack-nonPC try10-opt3 try6 try11-opt3.repack-nonPC try9-opt3.gromacs0.repack-nonPC try12-opt3.gromacs0.repack-nonPC try8-opt3.gromacs0.repack-nonPC try10-opt3 MQAC1-opt3.gromacs0.repack try7 try11-opt3.repack-nonPC try10-opt3 try9.opt3.groamcs0.repack try12-opt3.gromacs0.repack-nonPC MQAC1-opt3.gromacs0 try8.gromacs0.repack-nonPC try8 try11-opt3.repack-nonPC try10-opt3 try9-opt3 try12-opt3.gromacs0.repack-nonPC MQAC1-opt3.gromacs0 try8.gromacs0.repack-nonPC try9 try11-opt3.repack-nonPC try10-opt3 try12-opt3.gromacs0.repack-nonPC try9-opt3.gromacs0.repack-nonPC MQAC1-opt3.gromacs0 try8.gromacs0.repack-nonPC try10 try10-opt3 MQAC1-opt3.gromacs0 try11-opt3.repack-nonPC try12-opt3.gromacs0.repack-nonPC try9-opt3.gromacs0.repack-nonPC try8-opt3.gromacs0.repack-nonPC try11=try12 try11-opt3.repack-nonPC try10-opt3 try12-opt3.gromacs0.repack-nonPC MQAC1-opt3.gromacs0 try9-opt3.gromacs0.repack-nonPC try8.gromacs0.repack-nonPC rosetta try10 try12 MQAC1 try11 try7 try2 (all .gromacs0.repack-nonPC) I'll submit ReadConformPDB T0419.try11-opt3.repack-nonPC.pdb # similar to try9-opt3 # < chimera-try9-M2-C2-try10 ReadConformPDB T0419.try10-opt3.pdb # replaces MQAC1-opt3 # < MQAC1-opt3.gromacs0 < BAKER-ROBETTA_TS1 ReadConformPDB T0419.try12-opt3.gromacs0.repack-nonPC.pdb # same overall organization as try10-opt3 # < chimera-try10-M2-C2 ReadConformPDB T0419.try9-opt3.pdb # replaces try8-opt3 # < try8-opt3.repack-nonPC # < tweaked3-try7-opt3 # try7 < try6-opt3.repack-nonPC < try5-opt3.exploded # try5 < chimera-N1-try4-M1-C1 ReadConformPDB T0419.try4-opt3.gromacs0.repack-nonPC.pdb # < alignments (2plrA) Wed Nov 5 12:25:19 PST 2008 Kevin Karplus T0419 will have to be broken into two domains for evaluation. I think I beat the servers, though, on this one.