Wed Aug 14 14:48:13 PDT 2002 t0187 3 Sept 2002 Jenny Draper link from CASP5 doc: http://bioinfo-core.jcsg.org/cgi-bin/psat/analyzer.cgi?acc=TM1585&start=1&end=417 8 Sept 2002 Kevin Karplus The link Jenny provided seems to be dead---perhaps they have disabled it to keep CASP participants from seeing the structure, which is claimed to have been solved on the results of the search for protein name TM1585 on http://targetdb.pdb.org/apps/ TM1585 is also in COGS http://www.ncbi.nlm.nih.gov/cgi-bin/COG/palox?COG2379 So far, I haven't found much useful for our purposes on the COGS site---the BLAST alignment they show indicates that the center of the protein may be more conserved. try1-opt is an ugly mess, with predicted strands wound up into coils. We may want to add straightness constraints for the strands. The protein is large enough that it probably has multiple domains. I might as well try doing some arbitary subdomains. Let's try every 100 or so residues: t0187-1-150 t0187-91-240 t0187-201-350 t0187-301-417 8 Sept 2002 16:26 Kevin Karplus None of the try1 runs for any of the subdomains look worth anything. Probably all the "subdomains" are good for is to provide more fragments to the main script. I'll modify undertaker.script to include them along with the fragments, and do a short try2 run (without TryAllAlign). I don't expect them to come to much. If we do have an idea about TRUE domain boundaries, it would be worth creating a subdirectory for that subdomain. I think I'll leave this one to Jenny though, as I currently ahve no ideas about it. 8 Sept 2002 8pm Jenny Draper The link I have at the top of the page is fine (maybe the code generating it has problems displaying on linux?) It contains predictions of the structure, and links to homologous sequences (none solved yet) & various other info, like SWISS-PROT info. They describe the target as "glycerate kinase, putative", like the COGS site. Since I have no real idea how to start on this one, I'll try the same thing I ended up doing on the last two huge "I have no clue what to do with this" targets: use the robetta predictions. They don't seem very great, but at least robetta1 & 2 have strands where we predict strands, and they're pretty much straight, even though they haven't formed into sheets... I'll try to whip up some "hold these strands straight & kinda close to each other" constraints, and then try a run or two from the robetta1 & 2 models, hoping that undertaker won't curl up the strands in the process of moving the pieces together... 9 sept 2002 Kevin Karplus The like works fine for me today---perhaps the problem was with the site over the weekend, or with the new Nescape 7.0 I installed on my Mac, which seems to be a bit buggy. 9 sept 2002 Jenny Draper Re-starting try3, which died due to group permission problems: "T0187 10 sep Jenny is working on this one. Her last run died because I had left Template.atoms not group-writable. My apologies. Neither of us has any ideas what to do with this protein." 9 sept 2002 Jenny Draper 1pm Try3 overwrote try2.0.20, try2.4.20, and try2.5.20. I renamed them as "try3.*". They are new best-scoring with constraints, and look MUCH better than try1 or try2; they have two sheets starting to form. 9 sept 2002 Jenny Draper 1:30pm Starting try4 from try3.5, try3.4, and try3.0, using the same constraints. 9 Sept 2002 Jonathan Casper & Jenny Draper 4:57pm Adding a try5, which starts from a random conformation and attempts to apply our alignments. This script also uses some new constraints that Jenny Try5 is 30 generations, 20 super-iterations, with the best alignments included in doing try-all-align. Try6 is next, and will work entirely from the robetta models. The starting point for try6 will be try3.5.20, and the new constraints will be used. We're also adding a try7, which starts from try3.5.20, and just runs for a long time (80 gen, 5 super) with Jenny's new constraints. Try7 doesn't include any robetta information. 10 Sept 2002 Jenny Draper1:30am Try7 is new best-scoring with try5.constraints; try6 is close behind. Try5 (from alignments) is somewhere much lower, and looks terrible. 10 Sept 2002 Jenny Draper 2:00am Starting try8 from try6 and try9 from try7 (I like the overall shape of try6 better than try7, but I like the sheets a little better in try7), using try8.constraints, which make the constraints for the helix that insists on being a hairpin more steep, and switch the order of strands 10 & 12, which are crossed in tries 6 & 7. Hopefully lots of cpu hours can help sort this puppy out! :) 10 Sept 2002 Jenny Draper 10:30am My undertaker runs were not finished, but the condor restart this morning had them starting from the beginning again, so I killed them. We have 8 super-iters for each, which should be a good-enough length run. 10 Sept 2002 Jonathan Casper 10:50am I'm re-making score-decoys to see how well the new conformations do. Apparently, try9 is doing better than anything else. All of the try9 conformations score better than anything else, with try8 following close behind. For some reason, a try9-try7.7 has just appeared, which didn't get into score-decoys. I'll re-make one more time. 10 Sept 2002 Jonathan Casper 11:40am I've been looking at try9-try7.5, which is the best scoring conformation. Overall, it looks pretty good. There are still some large breaks, though. I'm a bit worried about the register for I271 in the second sheet. Either undertaker isn't doing a great job of obeying the constraints, or else that residue needs to be shifted down a notch on each side. As it currently stands, there isn't quite enough strand on the 270 end, and the 273 end is peeking out of the sheet a bit. There's a little problem where the 347 strand and the 413 strand are getting in each other's way. A very small move residues might take care of this. Alternatively, they're clashing right now, so maybe additional CPU time will make the problem go away. Strand 5 is still off on it's lonesome. The helix constraints haven't had much effect, but at least 5 is no longer pairing with that section. Instead, it's pairing with some of the preceding coil. 10 Sept 2002 Jenny Draper 1:00pm I'm still in the same situation with try8-try9 as with try6-try7. While try7 is more compact (& scores a little better), I think this is mostly because it is derived from try3 and has had more run-time. I think try8 (derived from try5) is better, but not as far along; try8-try6.5.80 seems to have much better sheet formation & has some nice helix-packing along the sheets. I suggest continuing in the two directions: try3->try4->try6->try8->try10 try5->try7->try9->try11 This time I feel they should use different constraints. Try8 needs: * are strands 10 & 12 in the right order? right now it's trying 9>10>12>. should it be 9>10>12>, like it was originally? * helices fixed * it's really looking like strand 5 should be parallel to strand 6 (for try8)... Try9 needs: * are strands 10 & 12 in the right order? right now it's trying 9>10>12>. should it be 9>10>12>, like it was originally? * helices fixed * strand 3 tucked back into it's sheet (it looks like it's being pushed out to make the overall protein more compact, and that the 2->3 loop is on the wrong side of the sheet) 10 Sept 2002 Jenny & Johnathan 3:30pm Started a try10 run, starting from try8-try6.5.80: * added strand 5 to strand 6 in parallel Started a try11 run, starting from T0187.hand15: * created a "T0187.hand15" pdb file to start from, which straightens out the sheets with move-resiudes.perl & moves some stuff out of the way, to help undertaker from getting stuck; this file is derived from try9-try7.5.80 For both try10 & try11: * changed max dist on all sheet constraints to "6" (from 8) * added helix constraints for broken helices * running w/ break cost @ 5, w 40gens, 5 super_iters * will need to be run again to fix breaks 10 Sept 2002 Jenny Draper THE "tryX" RUN: the try10 & try11 runs initially got mixed up, and try10 was wrtiting to try11 at the same time try11 was running. Both jobs were killed and re-started, and the two decoys try10 had created were named T0187.tryX.* TryX should therefore be viewed as "another short version of try10". 10 Sept 2002 Jenny Draper 7:50 Well, try10 & try11 runs are still going, and before submitting anything from them, they would probably need another run to reduce breaks? Currently, try9-opt-scwrl is best-scoring without constraints; it has a bad overlap between two strands, though. Try8-opt-scwrl scores pretty well, too; I prefer it to try9-opt. TryX-try8.0.40 is my favorite of the runs this afternoon; it has too many breaks to score well w/o constraints. It is pulling strand 5 into the 6-7-8-9-10-12 sheet and has the helices pretty straight -- but desperately needs to be run with gap-closing high for awhile. The try11 runs seem to be pulling the sheet apart in favor of the helices. Tue Sep 10 20:08:21 PDT 2002 Kevin Karplus with try11: try11-hand15.0.40 try10-try8.4.40 try11-hand15.1.40 tryX-8.0.40 try11-hand15.2.40 with no: try9-opt try8-opt hand1 try1-opt-scwrl with try10: try10-try8.4.40 tryX-8.1.40 try10-try8.1.40 Looked at try11-hand15.2.40 sheets look pretty good, helices cluster well around bigger sheet tryX-try8.0.40 bit of a funny reverse curve in middle of larger sheet, but all 7 strands present try11-hand15.0.40 scores best with try11.constraints try10-try8.4.40 best with try10.constraints, but has bad clashing strands try8-opt seventh strand not in sheet, but helices not too scattered try9-opt one strand pulled out of place in larger sheet hand1 slightly modified try9, sheets look a bit better than try9 try8-try6.5.80 was basis for try10, doesn't seem to be any better. model 1: tryX-try8.0.40 model 2: try11-hand15.0.40 model 3: try9-opt model 4: try8-opt model 5: hand1 28 Nov 2002 Kevin Karplus Overall, model 2 seems to be the best, but for domain 1 (4-22, 250-417) models 3 and 5 seem to be better. None of the models look like particularly good fits to me.