Mon Jun 23 09:29:52 PDT 2008 T0466 Make started Mon Jun 23 09:30:40 PDT 2008 Running on cheep.cse.ucsc.edu Mon Jun 23 11:29:51 PDT 2008 Kevin Karplus T0466 may be a template-free modeling problem, as the best e-value from the HMMs is 33.5. I kind of like the barrel that is based on 2vbuA (the top-scorer in the t04 searches, and 2nd in t06 searches, though it does not score well with t2k). try1 seems to be leaning towards a flatter sheet. Mon Jun 23 13:19:36 PDT 2008 Kevin Karplus try1-opt3 doesn't look too terrible, but I think I'll try try2 with just the 2vbuA alignments Mon Jun 23 17:27:42 PDT 2008 Kevin Karplus try2-opt3 looks pretty good to me, though there is a break that needs closing. Rather than polishing try2, though, I'll try creating another model from alignments, but using a costfch that includes the try2 sheets and helices. Mon Jun 23 20:02:09 PDT 2008 Kevin Karplus try3-opt3.gromacs0 doesn't score quite as well as try2-opt3.gromacs0.repack-nonPC, but comes pretty close. I think it is time for a polishing run including all the models. Tue Jun 24 10:20:57 PDT 2008 Kevin Karplus try4 polishes try2-opt3.gromacs0, getting it to score better than try3-opt3.gromacs0, despite much larger breaks. I think I might want to do another run (from alignments) with the sheets extended a bit. Perhaps SheetConstraint M2 A5 I100 D97 hbond Y99 Tue Jun 24 17:32:02 PDT 2008 Kevin Karplus The try5 costfcn prefers try4-opt3 to try5-opt3, The barrel in try5-opt3 is interesting, but the strand R26-N32 is crossing the barrel with no Hbonds. This is not so very different from try4. Perhaps I should do a polishing run with the try5 costfcn. Tue Jun 24 20:32:10 PDT 2008 Kevin Karplus try6 starts out polishing try5-opt2, then switches to try5-opt3, then to try4-opt3, as the break and clash penalties increase. It looks like I have 4 main models now: try6 < try4 < try2 < align (2vbuA) try5 < align (1nvmA) try3 < align (3kinB) try1 < align (1doiA) I don't really know how to choose among them---let me see what comes in with the metaservers. Tue Jun 24 20:54:55 PDT 2008 Kevin Karplus Model4 (try1-opt3.gromacs0) seems to disagree with the others on the phase of the strand R44-G52 or I41-S50. The secondary structure prediction favors the other models. Fri Jun 27 09:23:14 PDT 2008 Kevin Karplus MQAC favors the Zhang-Server models and PSI models, with SAM-T08-server_TS1 in the 11th position. MQAU favors PSI, MULTICOm, MUProt, with Zhang-Server_TS2 in 14th and SAM-T08-server_TS1 in 15th. So the metaserver runs will be picking different models than SAM+undertaker, giving me another view on what this protein does. Fri Jun 27 09:36:00 PDT 2008 Kevin Karplus It looks like MQAU1 will be based on GS-KudlatyPred_TS3 Fri Jun 27 10:20:27 PDT 2008 Kevin Karplus It looks like MQAC1 will also be based on GS-KudlatyPred_TS3 Fri Jun 27 14:43:11 PDT 2008 Kevin Karplus I don't like the long sheets of MQAU1-opt3 and MQAUC1-opt3. THey are not nearly as good-looking as a barrel. I'll try doing meta-server predictions with the try6 costfcn, which has barrel constraints in it. I might also want to look at getting the anti-parallel constraints from the MQAU1/MQAC1 models but combining it into a barrel. Fri Jun 27 15:11:28 PDT 2008 Kevin Karplus MQAU6 and MQAC6 also favor GS-KudlatyPred_TS3. Perhaps I need a costfcn that asks for barrels, but doesn't care so much about breaks and clashes? The Zhang-Server_TS4 doesn't have good hbonds, but nearly forms the barrel. The PSI server models are not barrels. Sat Jun 28 06:45:18 PDT 2008 Kevin Karplus I set up try7.costfcn to try to match the sheets almost formed in Zhang-Server_TS4, then scored all the models with try7. Among the complete server models, the best sheet scores were for fais-server_TS3 Zhang-Server_TS4 RAPTOR_TS4 BioSerf_TS3 MULTICOM-REFINE_TS2 MULTICOM-RANK_TS1 MULTICOM-CMFR_TS1 Poing_TS1 some of which had really terrible clashes. I'll turn clashes and breaks way down and try optimizing from the servers with this costfcn. I woder if I ever fixed undertaker so that incomplete models would be skipped in OptConform, rather than causing a crash---I guess I'll find out! Note: fais-server_TS3 forms a sandwich rather than a barrel, which may be ok. RAPTOR_TS4 forms a single-sheet meander, which curves back a bit, but does not form a closed barrel. BioSerf_TS3 has some long 3-strand antiparallel sheets, but it not at all compact. MULTICOM-REFINE_TS2, MULTICOM-CMFR_TS1, and MULTICOM-RANK_TS1 form a different folded anti-parallel sheet (almost a jelly roll). Poing_TS1 forms a sandwich that could almost be a barrel. Sat Jun 28 07:10:45 PDT 2008 Kevin Karplus undertaker does seem to know to reject incomplete models in OptConform. Unfortunately, try7 seems to be concentrating on SAM-T08-server_TS1, which will probably not do what I was hoping for. Perhaps I should not scale down breaks quite so much, so as to allow some other server model to come to the top. Sat Jun 28 09:01:15 PDT 2008 Kevin Karplus As the clash and break costs scaled up, try7 switched from SAM-T08-server_TS1 to MULTICOM-CLUSTER_TS3 try8 started with fais-server_TS3 and seems to be still optimizing from that. Sat Jun 28 12:55:21 PDT 2008 Kevin Karplus try8.costfcn really likes try8-opt3, thanks mainly to almost getting the sheets. try7-opt3 does not do quite as well, but is just behind MQAU1, MQAC1, and MQAU6. The try6 costfcn likes try6-opt3 best, then try4-opt3. Sat Jun 28 13:07:13 PDT 2008 Kevin Karplus We have a lot of fairly good choices for this target, both our own try6 and several models from metaservers. I should make a costfcn that does not include specific sheet constraints, but which likes beta sheets in general, to see which models it likes. try9.costfcn likes best MQAU1-opt3 MQAC1-opt3 (very similar) try8-opt3 MQAU6-opt3 (similar to MQAU1-opt3) try6-opt3 try7-opt3 try5-opt3 But I don't like that order. I like best try6-opt3, then try8-opt3, try7-opt3 MQAU1-opt3, try5-opt3. I might want to do more optimization of try6 and try5, to improve the barrel. Tue Jul 1 15:57:23 PDT 2008 SAM-T08-MQAO hand QA T0466 Submitted Tue Jul 1 15:57:23 PDT 2008 SAM-T08-MQAU hand QA T0466 Submitted Tue Jul 1 15:57:23 PDT 2008 SAM-T08-MQAC hand QA T0466 Submitted Mon Jul 21 11:49:50 PDT 2008 Kevin Karplus try10 started to polish try6. Mon Jul 21 11:53:52 PDT 2008 Kevin Karplus try11 started to polish try5. Wed Nov 5 09:55:08 PST 2008 Kevin Karplus I hadn't copied in the comments from the "method" file about what I finally submitted. Here they are: For T0466, SAM was unable to find a strongly-predicted template. The strongest hit (2pwwA) had an E-value over 33. I ended up with 3 anti-parallel barrel models and 2 beta sandwiches. I've put the SAM+undertaker models first, because I personally like barrels better than sandwiches. Model 1 also has the best Rosetta energy of any of the models for which rosetta repacked sidechains, though this is largely a function of how well clashes are removed, since the second-best rosetta energy came from a model that was very loose foam. Model 1 T0466.try11-opt3.gromacs0.repack-nonPC.pdb # < try5-opt3 < align(1nvmA) # best Rosetta energy 2 T0466.try10-opt3.pdb # < try6-opt3 < try4-opt3 < try2-opt3 < align(2vbuA) 3 T0466.try8-opt3.pdb # < fais-server_TS3 4 T0466.try7-opt3.pdb # < MULTICOM-CLUSTER_TS3 5 T0466.MQAU1-opt3.pdb # < GS-KudlatyPred_TS3 Wed Nov 5 09:56:14 PST 2008 Kevin Karplus As it turned out, the best model I generated was MQAC1-opt3 and the best I submitted was MQAU1-opt3 (as model 5). The MQA measures were better than me at picking out good models! There were a few better server models (notably, pro-sp3-TASSER_TS2-scwrl, nFOLD3_TS5, and MUSTER_TS1), but the MQAC1-opt3 model is better than any of the server models that went into it. The SAM-T08 server did terribly on this model.