Mon May 12 09:13:34 PDT 2008 T0397 Make started Mon May 12 09:14:12 PDT 2008 Running on cheep.cse.ucsc.edu Mon May 12 09:15:21 PDT 2008 Kevin Karplus BLAST does not find an obvious template. There is no indication from the crystallographers about the multimeric state of the protein. Mon May 12 09:54:01 PDT 2008 Kevin Karplus The T06.w0.5 HMM also doesn't find any good hits (best E-value 5.6) and the top hits are all from different folds, so the signal is definitely weak. This is probably going to be a template-free prediction. Make started Mon May 12 10:22:53 PDT 2008 Running on cheep.cse.ucsc.edu Mon May 12 10:23:14 PDT 2008 Kevin Karplus I was editing the Make.main file and temporarily damaged it, so I killed the make and restarted it. I may have to go back and clean up some damaged files that were in the process of being made, but I think that the make will catch them all. Make started Mon May 12 11:39:12 PDT 2008 Running on cheep.cse.ucsc.edu Mon May 12 11:39:26 PDT 2008 Kevin Karplus The rm_tiny make target had a bug in it, so I'm restarting make yet again. There are no great hits, but all the decent hits seem to be for b.84.1.1: # Fold: Barrel-sandwich hybrid [51229] sandwich of half-barrel shaped beta-sheets # Superfamily: Single hybrid motif [51230] 7 to 8 strands in 2 beta-sheets # Family: Biotinyl/lipoyl-carrier proteins and domains There are only 17 sequences in the t06 alignment (all archeal, I think), 17 in t04, and 16 in t2k. The top 5 alignments all suggest a similar model. Mon May 12 14:05:03 PDT 2008 Kevin Karplus The try1-opt3 model agrees with the models from alignment on the core, but there are several N-terminal strands and a helix that none of the alignments have. I think that two of the strands could be tacked on the edge of the half barrel---I'm less sure about the helix and the hairpin. Wed May 14 13:59:06 PDT 2008 Kevin Karplus The alignments agreed less well than I thought---there are some strands in the template that we could line up with. In particular there is a meander in best-models MODEL 5 which comes from ReadConformPDB T0397.undertaker-align.pdb model 3 which is an alignment to 1zkoA REMARK 44 model 3 is called T0397-1zkoA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:1zkoA I don't like that alignment much, but I think that we could get a better alignment to that template. Mon May 19 22:36:16 PDT 2008 SAM-T08-MQAO hand QA T0397 Submitted Mon May 19 22:36:16 PDT 2008 SAM-T08-MQAU hand QA T0397 Submitted Mon May 19 22:36:16 PDT 2008 SAM-T08-MQAC hand QA T0397 Submitted Wed May 21 20:38:31 PDT 2008 Kevin Karplus I'm starting metaserver runs with the try1 costfcn, but I think I need to come up with a better costfcn that tacks on the extra strands. Mon May 26 12:24:44 PDT 2008 Kevin Karplus MQAC1-opt3 scores best, followed by try1-opt3. MQAU1-opt3 scores substantially worse. MQAC1 is based mainly on Pcons_dot_net_TS1, while MQAU1 is based mainly on SAM-T08-server_TS1. The models are mostly in agreement for T81 on, but the N-terminal region has very different fold predictions. Let's do subdomains M1-K82 and G80-E150. Mon May 26 12:36:52 PDT 2008 Kevin Karplus subdomain predictions started on moai cluster. Mon May 26 14:04:27 PDT 2008 Kevin Karplus Actually, they didn't start until now, since a bug had crept into Make.main that made the subdomains try (and fail) to download documentation files from the casp8 directory). Mon May 26 17:36:09 PDT 2008 Kevin Karplus G80-E150 picks up hits to b.84.1.1 domains, though weakly (evalue 0.19) M1-K82 doesn't really pick up anything (no consistency, and best E-value 6.8). The G80-E150/try1-opt3 model looks ok to me. The M1-K82/try1-opt3 model is nasty, threading the first strand through another sheet, rather than pairing it with the second strand. This may be fixable. Tue May 27 16:03:17 PDT 2008 Kevin Karplus I added a bunch of sheet constraints to try to make M1-K82 form a simple meander, and started it as try2. Tue May 27 17:43:22 PDT 2008 Kevin Karplus I think I have the sheet constraints all wrong for M1-K82/try2, but despite that, try2-opt3 looks like it is making progress. I'll need to look at the model closely and pick out new sheet constraints to try for try3. Wed May 28 21:26:56 PDT 2008 Kevin Karplus I sent the M1-K82/try2-opt3 model to VAST, and it found similarity to 1jicA not in t2k 2fb7A in t2k 2eheA not in t2k 2q41C=2q41A not in t2k 2q0eA=2q0dA not in t2k 2pt6A not in t2k 2pwpC=2pwpA not in t2k 1y96A in t2k Wed May 28 21:44:53 PDT 2008 Kevin Karplus I made alignments to those templates by listing them in MANUAL_TOP_HITS for M1-K82/Makefile and making extra_alignments and read_alignments. I set up M1-K82/try3 to work primarily from those alignments. I also modifies the M1-K82/try3 costfcn from the M1-K82/try1 one by decreasing the align_constraints and breaks while increasing the ped_nb11_back, pred_alpha_back, hbond_geom_beta, and hbond_geom_beta_pair. (I got rid of maybe_metal and maybe_ssbond_geom, since there are no cys residues in this domain.) Wed May 28 22:39:48 PDT 2008 Kevin Karplus Aside from the N-terminal 8 residues and a few bad breaks, M1-K82/try3-opt3 looks pretty good. I'll turn up breaks and clashes and reoptimize. Wed May 28 22:54:23 PDT 2008 Kevin Karplus M1-K82/try4 started on moai cluster Thu May 29 07:34:07 PDT 2008 Kevin Karplus M1-K82/try4 looks pretty good. Now I need to close the gaps in G80-E150, then paste the models together. Thu May 29 09:31:43 PDT 2008 Kevin Karplus G80-E150/try2 has a bad break on either side of P95, which is unexpectedly appearing in the middle of a strand. Actually, it isn't so surprising, since it pairs with P137 in the next strand, which has a sharp bend. Perhaps I need to try pairing I96-F99 with I136-M133, but that would just force the break elsewhere. Maybe I should extend V103-G128, R102-T129, ... strand pair back to M97-N134 But the strand at F132 and M133 would have to flip over. Maybe I should just do another run with the try1 costfcn, and see what comes up this time. Perhaps I can find a set of strands that has few breaks and is more consistent. Thu May 29 10:31:11 PDT 2008 Kevin Karplus I'm doing G80-E150/try4 as another stab at starting from alignments, skipping try3 by mistake. Thu May 29 11:07:58 PDT 2008 Kevin Karplus The G80-E150/try4-opt3 model has the same problem as earlier models, with an unclosable break between P95 and I96. But I think I can cut-and-paste G80-P95 from the first alignment (to 2evbA) and make a chimera that works. Let's try that. Thu May 29 12:17:01 PDT 2008 Kevin Karplus The G80-E150/try5-opt3 almost closed the gap, but I don't like the burial of K92. Sigh, I don't seem to be able to get a model I like here. The Zhang-Server_TS3, which scores best with the MQAC costfcn, buries the K92 residue. If it made a salt-bridge with Q67, then maybe the burial would not be so bad. Sat May 31 07:57:56 PDT 2008 Kevin Karplus Turning breaks and soft_clashes weights way up in G80-E150 seems to be eliminating breaks and clashes, but the question is "at what cost to the overall fold?" G80-E150/try6-opt3 actually looks fairly good. There may be a little foaminess, but not much, and the G80-E150/try6-opt3 model scores best even with the try5 costfcn. Rosetta thinks that G80-E150/try6-opt3.gromacs0.repack-nonPC is the best model. The G80-E150/try6-opt3.gromacs0.repack-nonPC model picked a few different rotamers on surface residues, but is otherwise substantially the same. Sat May 31 08:27:16 PDT 2008 Kevin Karplus I'm now trying to do the same clash and break removal for M1-K82, where try4-opt3 still has a few breaks. I put in the sheet constraints and helix constraints from M1-K82/try4-opt3, since that structure looks much more convincing than the try2-opt3 one, which has fewer breaks. Sat May 31 09:37:12 PDT 2008 Kevin Karplus OK, both the subdomains look pretty good now. Rather than trying to paste them together and optimize, I'm first going to try to make a whole model using templates that VAST thinks are similar to the subdomain models, and a cost function that includes sheet and helix constraints from my best subdomain models. I'm trying that as try2. When it is done, I'll superimpose the best subdomain models on it, and try doing cut-and-paste to make things pack tigher or have fewer breaks. Sat May 31 09:45:24 PDT 2008 Kevin Karplus try2 failed with an assertion failure: undertaker: XYZpoint.h:57: void XYZpoint::unitize(): Assertion `mag>0' failed. I think these failures may be due to duplicated locations of points from aligning differently to the same template, but I'm not sure. I'll try running this under gdb and seeign if I can figure out where the assertion failure is coming from. Sat May 31 10:39:31 PDT 2008 Kevin Karplus By the time I got the program to run under the debugger, I couldn't duplicate the error. Sat May 31 10:57:59 PDT 2008 Kevin Karplus Trying try2 again. I may want to do it once more, but with the try2 costfcn for selecting the initial alignments also. Sat May 31 12:28:39 PDT 2008 Kevin Karplus The C-terminal domain of try2 looks fine, but the N-terminal domain is messed up. I'll try again as try3, but using the costfcn with the sheet and helix constraints in the initial TryAllAlign calls. For try4, I'll do the same optimization, but with the alignments from the subdomains added, and with no alignments *other* than to the templates listed in MANUAL_TOP_HITS in the Makefile. Sat May 31 13:35:11 PDT 2008 Kevin Karplus try4 seems to have settled on 2evbA and 2fb7A as the two templates, while try3 used 1mwwA and 2aalA heavily. Sat May 31 14:37:40 PDT 2008 Kevin Karplus The try3-opt3 model looks like junk for the N-terminus, but the try4-opt3 model is marginally plausible. I made two chimeras from the two individual domains, taking M1-G80 from M1-K82/try5-opt3 and T81-E150 from G80-E150/try6-opt3. The two chimeras differed only in how the two parts were superposed. For chimera-5-6, the superposition was done on residues G80-K82, while for chimeraB-5-6 the superposition was done together with the try4-opt3 model. I'll try optimizing the chimeric models in try5, using the same costfcn as for try4. Sat May 31 15:42:40 PDT 2008 Kevin Karplus try5-opt3 puts the two domains together, but still has a terrible break between the domains. I think I need to move M1-E11 out of the way, and pack the domains tighter, then put the strand back in. Sat May 31 16:00:25 PDT 2008 Kevin Karplus I made decoys/exploded-try5 by hand-editing some of the coordinates of the residues to move them a long way from the main body of the protein. I'm reoptimizing this model in try6, along with some packing constraints that I hope will cause the domains to settle together a bit. Sat May 31 17:13:37 PDT 2008 Kevin Karplus try6-opt3 looks plausible, though I think I may want to move the M1-E9 residues out of the way again and try squeezing things tighter. Sat May 31 19:17:30 PDT 2008 Kevin Karplus I'm trying to tighten the packing in try7, but the first 10 residues are likely to get in the way again, unless I can figure out where I want to put them. They look like they should be another strand, but on the edge of which sheet? Sat May 31 20:22:28 PDT 2008 Kevin Karplus try7 looks nearly identical to try6, just cleaning upt the sheets a tiny bit and reducing clashes. It certainly wasn't worth the time it took, and it didn't do anything about fixing M1-E9. Let's try adding sheet constraints, pairing L12 and N13 with V8 and Y7. SheetConstraint E11 D14 S37 E34 hbond Y36 SheetConstraint L12 N13 V8 Y7 hbond N13 This may well be wrong, but it should help keep the initial strand out of trouble. Sun Jun 1 08:05:13 PDT 2008 Kevin Karplus try8-opt3 looks terrible to me. For try9, I'll do a polishing run on all models, with no packing constraints or initial-strand contraints. I suspect that this will favor try5-opt3, since that scores best with the try9 costfcn initially. I'll do another, similar run for try10, but excluding try5 models. Sun Jun 1 09:29:41 PDT 2008 Kevin Karplus OK, try9-opt3 is basically a slight cleanup of try5-opt3, and try10-opt3 is a slight cleanup of try7-opt3. Both need more squeezing to remove breaks, so I'll do another polishing round for each. I think that I'm unlikely to make more progress on these models, so try11 and try12 will be my top two submissions (in some order). Now I need to decide what my other 3 submissions will be. My choices seem to be MQAC1, try3, try1, try8, and MQAU1. I'll only do one of the meta-server ones, and MQAC1 scores and looks better. It can be my 3rd choice. My 4th and 5th choices will be try1-opt3 and try8-opt3. I'll also have to decide between my models and the gromacs or gromacs+rosetta repackings. Sun Jun 1 10:53:57 PDT 2008 Kevin Karplus submitted: ReadConformPDB T0397.try11-opt3.pdb ReadConformPDB T0397.try12-opt3.pdb ReadConformPDB T0397.MQAC1-opt3.pdb ReadConformPDB T0397.try1-opt3.gromacs0.pdb ReadConformPDB T0397.try8-opt3.gromacs0.pdb Mon Jul 14 16:26:43 PDT 2008 Kevin Karplus PDB=3d4rA Our best model is chimera-5-6 with GDT 36.8% Or best submitted is model 2=try12-opt3 GDT=34.5% Model 1 has GDT 34.3% best server model is nFOLD3_TS3 (GDT=39.8%) SAM-T08-server had GDT 30.8%