Thu May 8 09:04:09 PDT 2008 T0393 Make started Thu May 8 09:04:40 PDT 2008 Running on cheep.cse.ucsc.edu Thu May 8 09:05:41 PDT 2008 Kevin Karplus Yet another target with a BLAST-detectible homology. Thu May 8 13:27:30 PDT 2008 Kevin Karplus The HMMs like 2f1kA and 2izzA best, but there are a *lot* of templates. BLAST likes 2i76A best, which the t06 HMM only tolerates, but which comes 8th for the t2k.w0.5 HMM. Thu May 8 17:43:10 PDT 2008 Kevin Karplus The crystallographers report that this is a dimer. I should probably look for a dimeric template and build from that. I'm particularly wondering whether the C-terminal helices are really a long coiled-coil dimerization interface. The a.100 superfamily that comes up as a hit is indeed a dimerization domain with swapping of C-terminal helices. So I'll HAVE to build a dimer from a template. It isn't clear which template to use. I'm tempted to try 2f1kA, 2pgdA, and 2ahrA, each of which has the c-terminus from a different family. 2f1k has 4 chains, but 2f1kA and 2f1kB are the dimer. Thu May 8 20:03:39 PDT 2008 Kevin Karplus I build dimer models from both 2f1k and 2izz, using both try1-init and try1-opt3 models. The try1-init models has some atoms in the same places---undertaker removed the duplicates then crashed because of the missing atoms. I'm now using only the try1-opt3 models. I put a constraint on the M185.CB atoms into dimer/try1.costfcn, based on the dimer-try1-init-2f1k model, but I now think that was a mistake, as the initial model seems to be based on 2izz, and so that dimer is cleaner. I may have to make a monomer based *separately* on a single template, then dimerize it on that template to get different dimers. Fri May 9 10:50:43 PDT 2008 Kevin Karplus The dimer/try1 run destroyed the dimerization interface and crashed with an assertion failure (unitizing a zero-length vector somewhere). Maybe I should do just the dimerization interface as a subdomain, to pick out the closest variant to use. Sat May 10 04:54:12 PDT 2008 Kevin Karplus It looks like a good breaking point would be A163, so I'll do M1-D164 and Y161-Q263. Sat May 10 05:02:12 PDT 2008 Kevin Karplus Makes for subdomains started on peep and lopez. Sat May 10 08:42:01 PDT 2008 Kevin Karplus The dimerization subdomain Y161-Q263 definitely likes 2i76A best, but the score is not strong (Evalue 2e-03). Sat May 10 09:08:49 PDT 2008 Kevin Karplus The part of Y161-Q263 that matches 2i76A is dimerization domain similar to the others I've looked at for this protein, so I should build a dimer from 2i76.pdb Sat May 10 14:17:45 PDT 2008 Kevin Karplus Trying to build a Y161-Q263 dimer now. There may be some chain numbering problems with dimers of subdomains. The M1-D164 domain HMMs find 2f1kA best, though blast finds 1gpjA best, and 2f1kA is down around E-vvalue 29. The t2k w0.5 HMM also likes 1gpjA best, but the t06 w0.5 model finds 1vpdA first. Sat May 10 17:53:50 PDT 2008 Kevin Karplus The Y161-Q263/dimer/decoys/T0393.try1-opt3 file does not look like the 2i76 dimer, which has the first helices anti-parallel, then the second helices almost antiparallel, a short helix parallel to the first one but on the other side of the dimer-partner's first helix, then an anitparallel pair of helices almost at right angles to the first helix. The pair of first helices makes a core that the rest wrap around. The dimer-from-try1-opt3-2i76 is not bad, though there are some bad clashes between the dimers Apparently undertaker resolved these by unwinding the interlaced dimer. Tue May 13 17:32:40 PDT 2008 Kevin Karplus Both the MQAU and MQAC QA methods like the Zhang servers best, so we should try making a monomer from them, then making a dimer from that. Wed May 14 19:08:14 PDT 2008 Kevin Karplus started undertaker < metaserve-MQAC.under >& metaserve-MQAC.log and undertaker < metaserve-MQAU.under >& metaserve-MQAU.log Thu May 15 09:35:31 PDT 2008 Kevin Karplus MQAU1-opt3 scores best with the try1 costfcn. I'll need to make the dimer and see how it comes out, using the 2i76 template. I should also build a dimer from Zhang-Server_TS3 (which MQAU1 used), and try optimizing that. Mon May 19 10:51:44 PDT 2008 SAM-T08-MQAO hand QA T0393 Submitted Mon May 19 10:51:44 PDT 2008 SAM-T08-MQAU hand QA T0393 Submitted Mon May 19 10:51:44 PDT 2008 SAM-T08-MQAC hand QA T0393 Submitted Mon May 26 10:19:51 PDT 2008 Kevin Karplus Let's try superimposing our models with the server models. The N-terminal domains all agree, but the C-terminal ones arrange the helices differently, and need to be resolved by building dimeric models. Mon May 26 11:00:42 PDT 2008 Kevin Karplus I tried making a dimer of Zhang-Server_TS4 using the 2i76A alignments from the whole chain and from just the C-terminal domain. They have a bit of clash between the monomers, but it looks fixable without unwinding everything. I'll try optimizing the dimers in dimer/try2 (I just hope undertaker doesn't crash!) Mon May 26 15:38:13 PDT 2008 Kevin Karplus RATS! undertaker didn't crash, but it undid the nice dimer. try2-opt1 had already undone it, and I don't know which operator undid it. It was probably OptSubtree, which is too bad, since I was counting on that to do some of the more minor reshaping. For try3, I'll do a short run with a similar costfcn, but constrain A211.CB atoms to be close and a solt bridge between the monomers for D208-R212 Mon May 26 20:15:18 PDT 2008 Kevin Karplus The dimer/try3 run did keep the dimerization interface intact, but introduced some bad breaks---most notably before T182. I'll do another run with breaks turned up to close the gaps, and with the dimer constraints tweaked a bit. try4 will start with just the try3 models. If the try4 model looks ok, I'll want to take the best N-terminal domain model I have and patch it in. Mon May 26 20:54:12 PDT 2008 Kevin Karplus I started a try2 optimization in M1-D164 to optimize first domains from several sources (including metaserver results). Tue May 27 07:04:41 PDT 2008 Kevin Karplus The dimer/try4 model has put the broken helix back together, though with a bulge at T178-T182 that seems a little unlikely. The result is more curved than the original template, so I'm worried that this is going in the wrong direction. Tue May 27 07:17:42 PDT 2008 Kevin Karplus For the M1-D164 domain, M1-D164/try2-opt3 scores best at the moment, but it does not have many the Hbonds for the sheet. I actually like M1-D164/try1-opt3 better for the sheet. Perhaps I should rerun with the try1 sheet constraints. Tue May 27 13:22:34 PDT 2008 Kevin Karplus M1-D164/try3-opt3 gets good sheets. I tweaked the cost function for M1-D164/try4 to make the try3 and try2 runs almost equal (interpolating between the try3 and try2 costfcns) and am doing another optimization to try to get the best features of both models. Tue May 27 14:07:00 PDT 2008 Kevin Karplus For the dimer, I should probably make a chimera that uses the dimer-Zhang4-2i76-Cterm up to T182, then dimer/try4-opt3 for N183 on. I can reoptimize that chimera, then replace the N-terminal domain with the separately optimized M1-D164 domain. Tue May 27 14:55:55 PDT 2008 Kevin Karplus I'm trying to optimize the chimera (on peep) and still optimizing the N-terminal domain (on lopez). Tue May 27 17:01:59 PDT 2008 Kevin Karplus M1-D164/try5-opt3 looks pretty good to me, though Rosetta likes M1-D164/T0393.try4-opt3.gromacs0.repack-nonPC slightly better than M1-D164/T0393.try5-opt3.gromacs0.repack-nonPC The dimer/try5-opt3 also looks ok, so now I have to try to stick the M1-D164 model onto the dimer. Tue May 27 17:10:43 PDT 2008 Kevin Karplus Perhaps the easiest thing to do is to superimpose the M1-D164 model and the dimer model, and do cut-and-paste. I might want to save K50-D58 from the dimer model, since it makes an extra strand of the sheet. This will only get one copy of the model in place though. I can then make chimera-M1try5-try5.unpack.pdb.gz to get a 2-chain version of the dimer, swap the chains, and do it again. Tue May 27 17:39:11 PDT 2008 Kevin Karplus I made that dimer model as chimera-M1try5-try5-twice and am optimizing it as dimer/try6. If it doesn't get destroyed in the process, I'll probably make that be my #1 model. Tue May 27 20:13:39 PDT 2008 Kevin Karplus dimer/try6-opt3 does not score quite as well as dimer/try4-opt3, because the clashes and breaks are a bit worse. I'll turn up soft_clashes and breaks and reoptimize just the dimer/try6 models. I'm starting that as try7. When it is done, I should probably do another optimization wtih the constraints turned down (or even off). Wed May 28 08:00:30 PDT 2008 Kevin Karplus RATS! dimer/try7 crashed with undertaker: Spot.cc:172: XYZpoint Spot::get_spot(const ChainsResiduesAndAtoms*, int, const Pointlist&, const Conformation*, bool, const Spot*) const: Assertion `return_point.mag2() < 1.e12' failed. This is weird, since that is a genuine assertion---the Spot shouldn't be able to get so far from the origin, without something having previously gone badly wrong. I'll try running the optimization again, hoping that a different random seed avoids the problem. I'll save the broken try7.log, so that I can attempt to debug later. (This target is due tomorrow, and I don't have much debugging time today.) Wed May 28 18:01:12 PDT 2008 Kevin Karplus OK, this time try7 ran fine and seems to have produced a decent result. I might want to run again, but first let's look at how the various monomers compare. Wed May 28 18:16:59 PDT 2008 Kevin Karplus I like the C-terminal domain of try7-opt3A.gromacs0, but I have some problems with the N-terminal domain. I might want to take G16-R39 from MQAU1-opt3, in order to fix up the edge strand. Wed May 28 18:48:08 PDT 2008 Kevin Karplus Done as chimera-try7-MQAU1 I've made a dimeric model of this and am reoptimizing it for dimer/try8 Wed May 28 18:54:41 PDT 2008 Kevin Karplus I've submitted ReadConformPDB chimera-try7-MQAU1.pdb ReadConformPDB T0393.try7-opt3A.gromacs0.pdb ReadConformPDB T0393.MQAU1-opt3.pdb ReadConformPDB T0393.MQAC1-opt3.pdb ReadConformPDB T0393.try1-opt3.pdb but I'll replace the first model by try8-opt3, if it looks better. Wed May 28 20:42:43 PDT 2008 Kevin Karplus I liked try8-opt3, and don't see much to be gained with further fussing. I've submitted Model 1 try8-opt3 optimization of independently produced M1-D164 domain and dimerized version of Zhang-Server_TS4 2 try4-opt3 dimer-optimized version of Zhang-Server_TS4, before replacing M1-D164 3 MQAU1-opt3 metaserver optimization, mainly from Zhang-Server_TS3 This model scores best with the constraints from alignments. 4 MQAC1-opt3 metaserver optimization, mainly from Phyre_de_novo_TS1 5 try1-opt3 fully automatic prediction Mon Jul 14 17:16:58 PDT 2008 Kevin Karplus PDB=3d1lA Best server=Zhang-Server_TS3 (GDT=62.9%) best model=dimer/try4-opt3.unpack.gromcas0 (GDT=57.8%) best submitted=model2=try4-opt3 (GDT=57.68%) model 1 (GDT=54.6%) Note: MQAU1 GDT=48.8% is much worse than Zhang-Server_TS3 (GDT=62.9%) which it was optimized from! SAM-T08-server_TS GDT=43.8% is way down in the middle of the pack.