Wed May 17 09:20:00 PDT 2006 T0292 Make started Wed May 17 15:36:24 PDT 2006 Running on camano.cse.ucsc.edu Wed May 17 19:08:13 PDT 2006 Kevin Karplus The t06 alignment has 232 PDB templates in it. The t04 alignment has 48 PDB templates in it. Wed May 17 21:04:40 PDT 2006 Kevin Karplus RPS-Blast identifies this protein as belonging to cd00180, S_TKc, Serine/Threonine protein kinases, catalytic domain. Phosphotransferases of the serine or threonine-specific kinase subfamily. The enzymatic activity of these protein kinases is controlled by phosphorylation of specific residues in the activation segment of the catalytic domain, sometimes combined with reversible conformational changes in the C-terminal autoregulatory tail.. NCBI-Blast (http://www.ncbi.nlm.nih.gov/BLAST/) finds the top hits in PDB to be 1zmwB =1y8gA 1zmuB =1zmuA 1zmvB =1zmvA 1y8gB =1y8gA 1mq4A 1muoA 1ol7A =1ol5A 2c6dA 2c6eB =2c6eA 2bmcF =2bmcA 1ol6A 2fh9A The best hit is 31% identity and 52% positives with 5 short gaps over 263 residues. This is not so close that it is an overwhelmingly right template---we probably want to use all the top templates in building a model. Wed May 17 21:57:52 PDT 2006 Kevin Karplus All three w0.5 models agree on 1zmuA as the top template, but the t2k HMM comes closest to the BLAST ordering (least drift from the target sequence?). Thu May 18 19:44:48 PDT 2006 Kevin Karplus Running blast locally on dunbrack-pdbaa gets top hits T0292 1y8gA 31.18 263 167 6 5 266 15 264 7.2e-33 136.0 T0292 1zmuA 30.80 263 168 6 5 266 15 264 1.6e-32 134.8 T0292 1zmvA 30.42 263 169 6 5 266 15 264 3.6e-32 133.7 T0292 1ol5A 31.84 267 167 6 4 269 10 262 3.4e-30 127.1 T0292 1muoA 31.84 267 167 6 4 269 25 277 3.4e-30 127.1 T0292 1mq4A 31.84 267 167 6 4 269 12 264 3.4e-30 127.1 T0292 2c6eA 31.84 267 167 6 4 269 13 265 5.7e-30 126.3 T0292 2c6dA 31.84 267 167 6 4 269 8 260 5.7e-30 126.3 T0292 2bmcA 31.84 267 167 6 4 269 34 286 5.7e-30 126.3 T0292 1ol6A 31.46 267 168 6 4 269 10 262 1.3e-29 125.2 T0292 2fh9A 31.06 264 166 8 5 266 9 258 2.8e-29 124.0 The HMMs get a different set of top hits, though still apparently the same family: 1rdqE 343 9.5900e-62 5.4286e-40 d.144.1.7 97314 1gz8A 292 2.2300e-59 7.5520e-39 d.144.1.7 83405 1ia8A 273 2.0500e-52 5.6014e-37 1ia8A d.144.1.7 62112 1t4hA 271 1.0400e-51 6.1071e-35 d.144.1.7 106420 2bikB 274 2.5300e-48 4.5183e-34 2evaA 296 1.7400e-51 7.5855e-34 1j1bA 355 1.6300e-49 1.1065e-32 d.144.1.7 90758 1fgkA 279 3.3400e-50 3.9700e-31 1fgkA d.144.1.7 41693 1p4oA 309 8.3600e-51 9.7298e-31 d.144.1.7 87775 1o6lA 337 2.1300e-63 1.8070e-30 d.144.1.7 81092 1z57A 334 3.1600e-45 2.1613e-30 Make started Thu May 18 19:51:40 PDT 2006 Running on camano.cse.ucsc.edu The make on camano was slow, because muscle was thrashing. I turned of the muscle pairwise alignment and restarted the job. Fri May 19 08:58:43 PDT 2006 Kevin Karplus Despite the number of templates, I'm not impressed with try1-opt2 (which seems to be based on 2bfxA). There are a lot residues exposed that are predicted to be buried. Perhaps I need to look at the templates more. Thu May 25 15:07:30 PDT 2006 Kevin Karplus Need to look at templates and try to figure out what to do on this one. Sat Jun 3 09:33:55 PDT 2006 Kevin Karplus try1-opt2 tore apart the N-terminal domain, probably to reduce clashes. Perhaps a couple of N-terminal-domain and between-domain constraints from the alignments would fix the problem. Perhaps H62.NE2 ad C125.SG at 6.3 Angstrom? Won't help much. V111.CB L207.CB 4.3 would though. Sat Jun 3 15:10:19 PDT 2006 Kevin Karplus I started another run from alignments (try2 on orcas) with the two constraints above, and with sheet and helix constraints from try1-opt2, which seems to have gotten the pieches right, just unpacked them in a couple of places. Sat Jun 3 15:28:28 PDT 2006 Kevin Karplus I tried scoring the server models, but I got a torsion angle assertion failure for karypis.srv.4_TS1-scwrl Sat Jun 3 16:39:52 PDT 2006 Kevin Karplus I modified undertaker to mark duplicate atoms as missing (even if *not* adjacent in the PDB file), so that the identical location ATOM 752 O LEU 92 20.778 -9.159 -9.815 1.00 0.00 ... ATOM 757 N ALA 93 20.778 -9.159 -9.815 1.00 0.00 in karypis.srv.4_TS1 no longer crashes undertaker. The best models with try2.costfcn are *not* try1-opt2, but SAM_T06_server_TS1, ROBETTA_TS4, SAM_T06_server_TS1-scwrl, ROBETTA_TS3, ROBETTA_TS1, .. Sun Jun 4 05:13:11 PDT 2006 Kevin Karplus The try2-opt2 model is ok, but SAM_T06_server_TS1 scores better, even with the try2-opt2 costfcn. The SAM_T06_server_TS1 model is also closer to the alignment to 1rdqE (the first alignment in undertaker-align). I think I'll do a polishing run that includes all the try2 models, the SAM_T06_server model and the best few models from the servers. With high crossover, we may be able to pick up a hint here and there. For try3 (on lopez), I'll start from ReadConformPDB T0292.try2-opt1.pdb.gz ReadConformPDB T0292.try2-opt1-scwrl.pdb.gz ReadConformPDB T0292.try2-opt2.gromacs0.pdb.gz ReadConformPDB T0292.try2-opt2.pdb.gz ReadConformPDB T0292.try2-opt2.repack-nonPC.pdb.gz ReadConformPDB servers/SAM_T06_server_TS1.pdb.gz ReadConformPDB servers/ROBETTA_TS4.pdb.gz ReadConformPDB servers/ROBETTA_TS3.pdb.gz ReadConformPDB servers/ROBETTA_TS1.pdb.gz ReadConformPDB servers/HHpred1_TS1.pdb.gz ReadConformPDB servers/FOLDpro_TS1.pdb.gz ReadConformPDB servers/RAPTOR-ACE_TS3.pdb.gz ReadConformPDB servers/SP4_TS1.pdb.gz I'll also SCWRL each of the server submissions, since the scwrled versions are sometimes much better than the unscwrled ones. Sun Jun 4 06:15:45 PDT 2006 Kevin Karplus I just noticed that all-align has 1966 alignments in it---this probably too many and is slowing things down. For future runs on this target, I should probably be more selective about which templates I load alignments from. Sun Jun 4 06:48:04 PDT 2006 Kevin Karplus Interestingly, try3 seems to be working mainly on optimizing try2-opt2.repack-nonPC, perhaps because the soft_clashes weight is so high. Sun Jun 4 10:40:24 PDT 2006 Kevin Karplus The top scorers with try3 costfcn are ReadConformPDB T0292.try3-opt2.pdb ReadConformPDB T0292.try2-opt2.pdb ReadConformPDB servers/ROBETTA_TS4.pdb ReadConformPDB servers/ROBETTA_TS3.pdb ReadConformPDB servers/SAM_T06_server_TS1.pdb ReadConformPDB T0292.try1-opt2.repack-nonPC.pdb (taking just the best scoring of the tryX-opt* models for each X). The try1-opt2 model is clearly damaged, so we won't use it. The other models are all very close. In fact, if we superimpose ReadConformPDB T0292.try3-opt2.pdb ReadConformPDB T0292.try2-opt2.pdb ReadConformPDB servers/ROBETTA_TS4.pdb ReadConformPDB servers/ROBETTA_TS3.pdb ReadConformPDB servers/SAM_T06_server_TS1.pdb ReadConformPDB servers/ROBETTA_TS1.pdb ReadConformPDB T0292.undertaker-align.pdb model 1 ReadConformPDB T0292.undertaker-align.pdb model 2 ReadConformPDB T0292.undertaker-align.pdb model 3 They are all within 2.75 Ang RMSD of the average of the superimposed models. The cores are within 0.8 Ang, with the ROBETTA models being closest to the core (probably because the ROBETTA models are so much like each other. The try3-opt2 and try2-opt2 seem to have spread things apart a little bit, probably to reduce clashes. I'd have to raise phobic_ft very high (from 3 to around 24) to make the ROBETTA models score better than the try3 models, though. Raising break, lowering sidechain, and removing constraints would make the Robetta models look realtively better. Sun Jun 4 12:03:47 PDT 2006 Kevin Karplus The try4 run looks like it is sometimes polishing ROBETTA_TS3 and sometimes try2-opt2 (though that *may* have been from a crossover with a ROBETTA model). It looks like both breaks and clashes can be reduced a fair amount. Sun Jun 4 12:33:35 PDT 2006 Kevin Karplus Looking at try4-opt1, I think that the crossover occured near E46, with the N-terminus coming from try2-opt2 and the C-terminus from ROBETTA_TS3. The model looks quite promising, and I suspect I'll be willing to submit try4-opt2. Sun Jun 4 15:01:09 PDT 2006 Kevin Karplus Rosetta likes repacking try4-opt2 best of any so far, and it scores best with the try4, try3, and unconstrained costfcns. There is still one bad break: T0292.try4-opt2.pdb.gz breaks before (T0292)N165 with cost 2.36091 Actualy, this isn't a break so much as a twisted peptide plane. If I rotated L164 in a bit, I could straighten out the L164-N165 peptide plane and bury L164 better. I might run into L136 if I do that. I wonder what tool would be best for this manipulation? ProteinShop? SwissPDBviewer? I downloaded Deep View to my Mac, but getting it to zoom seemed rather challenging (they assume a 3-button mouse, and the usual shift and control modifier to the 1-button mouse don't seem to work). Selecting and modifying also seemed a bit challenging, though the program *does* seem to have the ability to change torsion angles. I want to change the omega torsion angle between L164 and N165 and make a compensating correction in the phi angle of L164. Sun Jun 4 17:17:24 PDT 2006 Kevin Karplus Firas has agreed to look at this with ProteinShop. It might still be worthwhile for me to add an operator to undertaker though, as the problem is likely to recur. Sun Jun 4 17:50:36 PDT 2006 Kevin Karplus I tried creating a FixOmega operator, similar to the Tweak operators, but making omega be exactly 180 degrees, rather than making a random tweak. It may be worthwhile to add a tiny random term, though (probably up to about +- 10 degrees, but no greater than the current error in omega). I'm testing the FixOmega operator in try5, by giving it a very high probability of being applied. Sun Jun 4 18:39:52 PDT 2006 Kevin Karplus Well, try5 certainly applies FixOmega often, and it often helps, but by try5-opt1, it still hadn't fixed the peptide bond before N165. Some of the other breaks were reduced and two were cleaned up enough to disappear, but the overall effect was less than I had hoped for. I'll try again after making the angle change be between 0.4 and 1.6 of the optimal correction, and with soft_clash weight turned down, to let L164 move even though it will clash. Sun Jun 4 20:07:26 PDT 2006 Kevin Karplus try5 finished and did not fix the break before N165. I'm running try6 to try6 to try again with a somewhat more random FixOmega operator. Looking at try4 and try5 compared to other loops in that area---I'm not so sure I like what they are trying to do. I think I may want to cut-and-paste in the loop from SAM_T06_server_TS1 or even from undertaker-align model 1 [OOPS, that one is incomplete here]. The loop to copy is from A161 to T177. I should probably superimpose focusing on those two residues, to minimize the wobble. Sun Jun 4 20:20:25 PDT 2006 Kevin Karplus I made such a chimera (decoys/chimera-try5-server.pdb). I will kill try6, redo the costfcn, and redo the optimization to start from the chimera. Sun Jun 4 21:24:22 PDT 2006 Kevin Karplus The chimera introduced a lot of breaks and clashes, but most of these are fixed by try6-opt1. At least the three worst breaks (before V175, P178 and A173) are gone or greatly reduced. None of the remaining breaks are big enough to worry about. Sun Jun 4 22:02:37 PDT 2006 Kevin Karplus try6-opt2 scores almost as well as try5-opt2, and has much smaller breaks. Its clashes are a bit worse, so I should do a polishing run (say with the try4 cost function) starting from just the try6 models. Mon Jun 5 07:41:27 PDT 2006 Kevin Karplus try7-opt2 scores best with the try7 (hence try4) costfcn, also with try6. Rosetta still prefers repacking try5-opt2. I will submit ReadConformPDB T0292.try7-opt2.pdb ReadConformPDB T0292.try5-opt2.repack-nonPC.pdb ReadConformPDB T0292.try3-opt2.pdb ReadConformPDB servers/SAM_T06_server_TS1.pdb InFilePrefix ReadConformPDB T0292.undertaker-align.pdb model 1 (based on 1rdqE) Fri Jul 14 11:26:06 PDT 2006 Kevin Karplus With the evaluate.unconstrained.pretty, our try3-opt1-scwrl model was the best model (beating all the servers), and our model3 was our best submitted model (again beating the servers). Our models were in order: model3, model4, model1, model2, model5 The ordering may change a lot if one looks at the two domains separately (splitting somewhere between L164 and G176, since that part is disordered in the crystal). The SAM_T06 server is 18th of the servers on the TS1 models---not too great. Fri Jul 14 15:32:58 PDT 2006 Kevin Karplus Splitting into the two domains: evaluate_1.pretty best server: Frankenstein_TS2 -0.99 our best: try1-opt2.gromacs0 -0.99 best-submitted: model3 -0.95 our server: SAM_T06_server_TS1 -0.82 5th of 53 TS1 evaluate_2.pretty best_server: Zhang-Server_TS1 -1.23 our best try3-opt2 -1.13 best submitted: model3 -1.12 our server SAM_T06_server_TS -1.09 7th of 53 TS1 OK, on separate domains we are in the top crowd. Sun Sep 10 08:25:03 PDT 2006 Kevin Karplus I redid the evaluation function last week adding the GDT, smooth_GDT, and real_Hbonds, and reweighting components to have roughly the same scale. Our models are now the best (compared with the servers). Whole chain: our best: try2-opt2.gromacs0 -111.80 best submitted: model3=try3-opt2 -110.62 best server: LOOPP_TS3 -94.79 SAM_T06_server: -70.61 (6th of 52) Domain 1: our best try1-opt2.gromacs0 -122.16 best server: ROBETTA_TS2 -121.28 best submitted: model3=try3-opt2 -106.84 SAM_T06_server: -101.09 (4th of 52) Domain 2: best server: Pcons6_TS1 -142.85 our best: try3-opt1 -141.17 best submitted: model3=try3-opt2 -140.63 SAM_T06_server: -136.46 (3rd of 52) (Note: the incorrect count of 53 before was due to counting both the GeneSilicoMetaServer model and its scwrled version.) Mon Mar 19 11:12:50 PDT 2007 Kevin Karplus The domains I split into above (1-164 and 176-277) do not agree with the assessors domains. They split 1-86, 87-277, so I need to redo the assessment. Also, the June version of 2cl1 was replaced on 4 Dec 2006, and has been replaced again with 2javA. Mon Mar 19 14:10:04 PDT 2007 Kevin Karplus For the complete model, the best we have is try3-opt1-scwrl and the best submitted was model3, beating any of the servers. Whole chain: our best: try3-opt1-scwrl -156.88 best submitted model3=try3-opt2 -151.34 best server: Zhang-Server_TS4 -142.38 SAM_T06_server: -109.25 (6 TS1 of 52) domain 1 (1-86): our best: model4=al1 to 1rdqE -218.81 best submitted: model4=al1 to 1rdqE -218.81 best server: RAPTOR_TS5 -204.08 SAM_T06_server: -218.81 (3 TS1 of 52) domain 2: (87-277) our best: try7-opt2 -163.05 best submitted model1=try7-opt2 -163.05 best server: Zhang-Server_TS4 -169.63 SAM_T06_server: -140.40 (6 TS1 of 52)