Tue May 13 09:16:02 PDT 2008 T0399 Make started Tue May 13 09:16:52 PDT 2008 Running on cheep.cse.ucsc.edu Tue May 13 09:31:06 PDT 2008 Kevin Karplus This is probably a fold-recognition target, though pdb_blast has a poor best e-value of 1.8 (for 2d3nA). Tue May 13 15:36:05 PDT 2008 Kevin Karplus There seems to bee pretty good consensus on the C-terminal domain (after K59 or F67) being d.98.1.1 (2g2uB or 1jtgB), but the N-terminus seems to be ab-initio. (Adding 1xxmC and 1s0wC from w0.5 searches) Perhaps I should try folding it separately, though I suspect it is a bit too small. (Subdomain prediction started in M1-G69.) L10,L14,F18,L21,M22,L23 are all predicted to be buried, but are exposed in try1-opt3. The C-terminal domain does not show any buried patches, so perhaps the N-terminal helix is a dimerization domain? Maybe I should do a V60-E206 subdomain also, to avoid confusion from the N-terminal stuff. Tue May 13 16:55:59 PDT 2008 Kevin Karplus Looking at the conserved residues in try1-opt3, I wonder about W175---it seems to be on the wrong side of the sheet---perhaps that strand is flipped. There are only 2 CYS residues: C19 and C25. I'll have to wait for results on the N-terminal domain before deciding if these are significant. BLIP *is* a secreted protein, so it seems likely that they form a disulfide. Tue May 13 17:42:38 PDT 2008 Kevin Karplus M1-G69 didn't really find anything, so I should probably rerun it with SSBOND C19 C25. (OK, M1-G69 try2 started) Mon May 19 22:37:15 PDT 2008 SAM-T08-MQAO hand QA T0399 Submitted Mon May 19 22:37:15 PDT 2008 SAM-T08-MQAU hand QA T0399 Submitted Mon May 19 22:37:15 PDT 2008 SAM-T08-MQAC hand QA T0399 Submitted Fri May 23 06:13:16 PDT 2008 Kevin Karplus Both MQAU and MQAC like the SAM-T08-server for this one! I probably need to add SSBOND C19 C25 before running the metaserver scripts Fri May 23 06:30:45 PDT 2008 Kevin Karplus Oops, M1-G69 try2 was a bit of a waste, as the SSBOND command was not included in the scoring (due to a typo in the ConstraintSet name). Running it again as try3, with the costfcn corrected. Fri May 23 07:40:53 PDT 2008 Kevin Karplus M1-G69/try3-opt3 gets the ssbond, but has some bad breaks. I'll try running again, with the known_ssbond turned down a bit and breaks turned up. Fri May 23 08:40:10 PDT 2008 Kevin Karplus M1-69/try4-opt3 scores well and has an adequate ssbond (though the torsion angle is off), but the secondary structure is quite different from earlier attempts. I don't know what to think of this domain! Fri May 23 09:00:23 PDT 2008 Kevin Karplus Given that I have no idea what to do with the first domain, I think I'll run the metaservers with a try2 costfcn that is just try1 plus SSBOND C19 C25 and with maybe_metal and maybe_ssbond removed. Wed May 28 22:15:33 PDT 2008 Kevin Karplus MQAC1, apparently mainly from SAM-T08-server_TS1, looks like quite a reasonable sheet, though the packing of the N-terminal helices is a bit dubious. MQAU1, mainly from fais-server_TS1, is not as convincing. I like the placement of the initial helices in Zhang-Server_TS1, so maybe I should try mixing that with MQAC1-opt3. MQAC1, Zhang-Server_TS1, MQAU1, and try1-opt3 all agree pretty much after N84. I'll make a chimera of Zhang-Server_TS1 and MQAC1 and try optimizing that. Thu May 29 08:19:21 PDT 2008 Kevin Karplus The try2 run messed up the parts of the N-terminal domain that I had copied from the Zhang-Server model. It tore apart already in the try2-opt1 model. What tears it apart is the attempt to form the C19-C25 ssbond, which is not compatible with the Zhang-Server N-terminal fold. I'll try again without the ssbond for try3. Thu May 29 11:46:15 PDT 2008 Kevin Karplus try3-opt2 does not score as well even as try1-opt3. Though it has good hbonds, it loses on the predicted burial and on the align_constraints. I think I may want to go back to the 2g2uB structure, and get the inital helices and strand from that. Thu May 29 12:00:28 PDT 2008 Kevin Karplus The V60-E206/try1 model has a decent structure, but needs to have a gap closed, so I'll polish up that. Actually, before polishing, I'll try doing an optimization from the alignments again (though just the 4 hits from the right family). Thu May 29 14:14:01 PDT 2008 Kevin Karplus The V60-E206/try2 model has one bad break, but overall looks a little better than the V60-E206/try1 model. I'll do a polishing run fom the two of them, then try pasting on some possible N-terminal domains. Thu May 29 18:17:48 PDT 2008 Kevin Karplus The V60-E206/try3 model has eliminated the bad breaks. I think it is about as polished as it it going to get. Now I have to stick on an N-terminal domain. I don't really like any of the N-terminal domains I've generated---they all look like random junk structures. I get the most beta structure from M1-G69/try3-opt3.gromacs0, so maybe I should try that. Thu May 29 18:58:11 PDT 2008 Kevin Karplus I put together a two-domains.pdb file containing M1-T62 from M1-G69/try3-opt3.gromacs0 (moved 100 in x to avoid clashes) and D63-E206 from V60-E206/try3-opt3. I'll try optimizing them to close the huge break between domains and see what we come up with. Thu May 29 20:54:11 PDT 2008 Kevin Karplus try4-opt3 is the best-scoring model so far, but it did unwind the short helix at the beginning of the second domain. Maybe I should do some cut-and-paste to restore that? Also L14-C25 are predicted to be buried but are rather highly exposed. I wonder if I should also make a two-domain model using M1-G69/try5-opt3 I think I'll make 2 new models---one from try4-opt3 with F55-G69 replaced by the fragment from the first alignment, then reoptimize adding a helix constraint for F55-F67. The other new model will be that model, but with M1-S54 taken from M1-G69/try5-opt3. Thu May 29 21:28:11 PDT 2008 Kevin Karplus Oops---that is nonsense, since the first alignment omits T61-N65! I can copy T62-N65 from the V60-E206/try-opt3 though. I'll make that chimera-try4-62-65 and optimize it with try5. Thu May 29 21:41:31 PDT 2008 Kevin Karplus I took chimera-try4-62-65 and replaced M1-T61 with M1-G69/try5-opt3 and callid it two-domain-try5-try4. I'll optimize it in try6. I'll also redo try4 as try7, but with a different costfcn---one that tries to retain the sheet of the N-terminal prediction. Fri May 30 08:21:20 PDT 2008 Kevin Karplus try5 got the helix back. I don't particularly like the N-terminal domain for it. try6 doesn't have the little helix, but it packs the N-terminal domain a bit better. Perhaps I should copy the helix from try5 and reoptimize with the pred_nb11_back and dry weights turned up. try7 never rejoined the domains! Perhaps clashes were too high to allow the break to be fixed. Fri May 30 09:21:48 PDT 2008 Kevin Karplus I started a polishing run in try8. I think it will probably rely most heavily on try6, and not move it much. Fri May 30 15:56:37 PDT 2008 Kevin Karplus try8-opt3 scores slightly better than try6-opt3, but the try8 costfcn does not include any ssbond terms (either known_ssbond or maybe_ssbond). I should do another polishing run with the known_ssbond. Fri May 30 19:08:49 PDT 2008 Kevin Karplus try9-opt3 looks ok, and sort of forms the SS bond (the SGs are ok, but the CBs are too close). Mon Jul 14 16:19:09 PDT 2008 Kevin Karplus PDB=3d4eA Our best model is MQAU1-opt3 (GDT 50.46%), beaten in real_cost only by BAKER-ROBETTA_TS4 and HHpred2_TS1 (and scwrled versions of each). My intuition is bad, as I thought this model was less convincing than MQAC1, with a GDT of 46.5%. Our best submitted model was model2 = try5-opt2, with GDT=47.2%). model1 has GDT 46.8% The SAM-T08-server got only GDT=44.6%