Tue Jun 24 09:23:56 PDT 2008 T0469 Make started Tue Jun 24 09:25:10 PDT 2008 Running on moai05.kilokluster.ucsc.edu Tue Jun 24 11:30:26 PDT 2008 Kevin Karplus modest hit (E-value 0.07) to 2fi0A. The try1-opt3 model looks pretty good, so I think I'll polish it to reduce clashes and breaks and try to get tighter packing. There appears to be a disulfide for the conserved C32-C35, though there is a possibility that they form a metal-binding site with H50. Tue Jun 24 16:59:11 PDT 2008 Kevin Karplus try2-opt3 looks pretty good. I think I'll wait on any more polishing until I see the meta-server results. Sat Jun 28 13:26:40 PDT 2008 Kevin Karplus T0473 looks like it is based on the same templates as T0469, and it does not seem to form the disulfide. Instead it forms a CH pair with the earlier CYS and a CRQ triple with the later one. Let's try re-optimizing this model *without* a disulfide. For try3, I'll start again from alignments, but without a bonus for forming disulfides. Sat Jun 28 14:32:35 PDT 2008 Kevin Karplus try2-opt3 scores better than try3-opt3, mainly because of smaller breaks, but Rosetta likes try3-opt3.gromacs0.repack-nonPC best. I think that a polishing run on try3 is likely to move it up to be at least as good as try2. Sat Jun 28 14:37:49 PDT 2008 Kevin Karplus try3-opt3 does have C32.SG near H50.NE2 Neither of these residues is the same as in 2fi0A, so I can't measure that template to get a constraint. I wonder if H50, C32, E41, and maybe C35 form a metal-binding site or other active site. Sat Jun 28 15:32:26 PDT 2008 Kevin Karplus try4-opt3 from try3-opt3.gromacs0 scores best with try4 costfcn, but rosetta still slightly prefers try3-opt3.gromacs0.repack-nonPC. So I now have two models without the disulfide and 2 with. I suspect that the disulfide is *not* formed, but I can't guarantee it. I wonder if it is worth aligning T0469 and T0473 and using that as a seed for finding homologs. The current t06 alignment has over 100 sequences, so it probably won't find a lot more. In any event, the two sequences already find each other and align gaplessly in the t06 alignments. Note: looking through the names of the homologs in the t06 alignment, I see that most are of unknown function, but three are sulphite reductases. Since the organism is Geobacter sulfurreducens PCA, I suspect that this protein was chosen as a putative sulfite reductase. Tue Jul 1 15:14:46 PDT 2008 Kevin Karplus I started MQAU1 and MQAC1 metaserver runs. I may also want to start MQAU4 and MQAC4 runs, unless these already do well with the try4 costfcn. The MQAU list has some weird servers that I haven't seen near the top before, but SAM-T08-server and BAKER-ROBETTA also come out in the top 10, so I guess it is ok. MQAC also has the unusual ones at the top (FFASflextemplate, FFASstandard, FFASsuboptimal, pipe_int), but also has more usual ones like RAPTOR and MUProt. The SAM server and the BAKER-ROBETTA server are not in the to 10 here, but the differences are small: SAM-T08-server is 79th with predicted GDT of 85.3% and the top prediction is for 86.9%. Tue Jul 1 15:43:09 PDT 2008 Kevin Karplus MQAU1 chose to optimize SAM-T08-server_TS1 MQAC1 chose to optimize MULTICOM-CMFR_TS3 Tue Jul 1 15:57:26 PDT 2008 SAM-T08-MQAO hand QA T0469 Submitted Tue Jul 1 15:57:26 PDT 2008 SAM-T08-MQAU hand QA T0469 Submitted Tue Jul 1 15:57:26 PDT 2008 SAM-T08-MQAC hand QA T0469 Submitted Tue Jul 1 15:59:54 PDT 2008 Kevin Karplus All the models are essentially the same, except for the loop L30-A38 and some wiggling of the N-terminus M1-T6 The loop includes C32 and C35, and some of the variation may come from trying to make them interact or to make C32 interact with H50. Fri Jul 4 10:46:55 PDT 2008 Kevin Karplus The big decision seems to be what to do with the CYS residues. If they form a disulfide, then they probably are *not* 3 residues apart on a helix. A 3-10 helix would make the disulfide even harder to form than an alpha helix. A pi-bulge would work. Could we have a metal ion coordinated by E41 H50 C32 and C35? Fri Jul 4 12:03:10 PDT 2008 Kevin Karplus I looked through MDB http://metallo.scripps.edu and found some sites that involve E,H,C,C, like a copper-binding site in 1ar1A (that turns out not to really involve the glutamic acid---just its backbone). Or one one of iron-binding sites in 1cc1 (with 3 cys and a his, and a glu sort of nearby). The copper-binding side in 1cc1 and 1cyx is 2cys, 2 his, a MET and 2 backbone O atoms coordinating a pair of copper atoms. There are others found by MDB, but the search doesn't seem to be selective enough to pull out interesting sites. Sun Jul 6 20:57:53 PDT 2008 Kevin Karplus Let me try pulling H50.NE2, C32.SG, C35.SG all within 3.6 of each other. try5 will try doing that from alignments try6 will try doing it from existing models. MQAX5 will try doing it from server models. Mon Jul 7 03:39:04 PDT 2008 Kevin Karplus All three runs managed to get low costs for the added constraints without greatly increasing costs elsewhere. Now I need to look at them to see what damage was really done. If they look ok, I'll have to do a polishing run to remove the breaks and clashes which have crept in. Mon Jul 7 03:51:46 PDT 2008 Kevin Karplus The try5 and try6 models seem to have accomodated the clustering without significant damage to the rest of the model, though with rather bad clashes between H50 and C35 for try6 and between C32 and C35 for try5. I'll see if the clashes can be resolved without losing the clustering. Hmm, just loosening the consraints a little and increasing the weight of the clashes and breaks moves try5-opt3.gromacs0 to the top of the list, so I think it is possible to fix things up. Mon Jul 7 04:33:59 PDT 2008 Kevin Karplus try7 reduced clashes a lot, so I'll try again with try8 turning up clashes and breaks further, and starting just from models that have been run through gromacs. Mon Jul 7 05:27:10 PDT 2008 Kevin Karplus try8-opt3 saisfies the constraints, and clashes are not bad, but I'd like to make the potential metal-binding site more convincing, by having the CB atoms of the CYS residues at least 4, and preferably 5.4 Angstroms from the NE2 atom of the HIS, 5.7 from the CE1, and 6 from the CD2. (That is, I like where C32 is coming out, but C35 needs to be pushed back a bit.) Mon Jul 7 06:53:39 PDT 2008 Kevin Karplus try9 does not fix the problem. Perhaps I should try again exluding try7 and try8, which seem to be trapped in a different local minimum. Mon Jul 7 07:56:01 PDT 2008 Kevin Karplus try10, from MQAX5-opt3, gets a better metal-binding site, but has some bad clashes and breaks. I'll turn up clashes and breaks and try to polish it. Mon Jul 7 08:44:10 PDT 2008 Kevin Karplus try11 gets the clashes down to a reasonable level but has bad breaks before G35, C35, M36, G37, and A38. I'll try to close those in try12. Mon Jul 7 09:29:22 PDT 2008 Kevin Karplus try12 cuts the breaks in half, but still has trouble. Mon Jul 7 09:39:59 PDT 2008 Kevin Karplus I'll try a run with the HHC constraints turned down a bit and and starting from just try10,11,12 models that have been run through gromacs, which will change the breaks. Mon Jul 7 11:04:28 PDT 2008 Kevin Karplus try13-opt3 manages to have no breaks and only small clashes, though gromacs wants to break it to relieve those clashes. The SG atoms are a little close, but I think this is about as good as I'm going to get without being able to model a metal ion directly. Mon Jul 7 11:17:50 PDT 2008 Kevin Karplus OK, I'll submit 5 different models to get a little variety: # attempt to form CCH site ReadConformPDB T0469.try13-opt3.pdb # < try12-opt3.gromacs0 < try11-opt3 < try10-opt3 < MQAX5-opt3 < Zhang-Server_TS4 # attempt to form CCH site ReadConformPDB T0469.try9-opt3.pdb # < try8-opt3 < try7-opt3.gromacs0 < try6-opt3 < MQAU1-opt3 < SAM-T08-server_TS1 ReadConformPDB T0469.try4-opt3.pdb # < try3-opt3.gromacs0 < align(2fi0A) ReadConformPDB T0469.MQAC1-opt3.gromacs0.pdb # < MULTICOM-CMFR_TS3 # attempt to form CCH site ReadConformPDB T0469.try5-opt3.pdb # < align(2fi0A+2cs8A) Wed Nov 5 10:13:32 PST 2008 Kevin Karplus I was wrong about C32, C35 and H50 clustering, which messed up all my models. The Zhang-Server models were the best by GDT, pro-sp3-TASSER_TS4 by real_cost. The best model I submitted was model4 (which was MQAC1-opt3.gromacs0) but model5-scwrl was better. This looks like a target that I would have done better (by GDT) with a metaserver MQAC1-opt.