Wed Jun 14 09:11:10 PDT 2006 T0331 Make started Wed Jun 14 09:12:42 PDT 2006 Running on orcas.cse.ucsc.edu Wed Jun 14 09:15:01 PDT 2006 Kevin Karplus BLAST does not get any good hits in PDB. Best are 1g79A and 1dnlA, both with 31% id over 55 residues (E-value 0.071) This may be enough to get picked up in the iterated alignments, but maybe not also. Wed Jun 14 10:40:56 PDT 2006 Kevin Karplus The 1g79A and 1dnlA templates are indeed scoring well with the HMMs, as are other b.45.1.1 templates. Wed Jun 14 15:03:42 PDT 2006 Kevin Karplus THere is general agreement on b.45.1.1, and the alignments are mostly very similar. We will have to optimize this as a dimer. Sun Jul 2 12:25:21 PDT 2006 Kevin Karplus Getting near the deadline, and no one has looked much at this. I'll do another run before I try to create the dimer. The biggest choice seems to be where the C-terminus goes---there are two choiced for S109 on in the different alignments, and try1-opt2 seems to have picked the one that has the poorer match to secondary structure and the poorer scores with the HMMs (based on 2fhqA instead of 1g79A). Sun Jul 2 12:45:29 PDT 2006 Kevin Karplus I put all the hits into MANUAL_TOP_HITS in the Makefile, and did make extra_alignments make read_alignments Now I need to select which templates to try in try2. Sun Jul 2 12:58:50 PDT 2006 Kevin Karplus I started try2 with templates 1g79A 1dnlA 1nrgA 1ty9A 1ci0A on cheep. It may not have been necessary to restrict the template set, as the try2 cost function (with fewer sheet constraints than the try1 cost function) already scores the alignments better than it scores try1-opt2. It looks like try2 will be working mainly on a 1dnlA alignment. Sun Jul 2 13:56:18 PDT 2006 Kevin Karplus On looking at the templates that try2 is based on, I'm not convinced that the helical insertion really does belong in this target---the all have the final two strands that are missing if we include the helical region. The multiple alignments are poor in the C-terminal region---we don't have good matches for t2k key residues: key actual P138 I138 WYFLH143 H143 YFLVW146 V146 and t04 has no key residues (finds only itself). The t06 alignment avoids the misalignments of the t2k alignment, but does find several sequences. Looking at just the t06.best-scores file, we get a different ordering of top hits, with 2fhqA as the top template (followed by 1dnlA, 1g79A, ...) I'm going to make try3 focus on the 2fhqA template, by including the constraints from its t06 alignment, and by using the t06.dssp-ehl2.constraints Sun Jul 2 14:15:21 PDT 2006 Kevin Karplus try3 started on cheep. The try3 costfn greatly prefers try1-opt2 to try2-opt2. Sun Jul 2 14:44:07 PDT 2006 Kevin Karplus I scored all the server models with the try2 and try3 costfcns. scoring with try3 likes our try1-opt2 best, then ROBETTA_TS4 (and Pmodeller6_TS2, that copies it). Next is RAPTOR_TS1-scwrl, RAPTOR_ACE_TS4-scwrl, BayesHH_TS1-scwrl, ... scoring with try2 crashed with an assertion failure trying to compute an Hbond undertaker: HbondCostFcns.h:21: double Lennard_Jones(double, double): Assertion `dist2>0' failed. probably on CaspITa-FOX_TS3-scwrl Sun Jul 2 14:56:11 PDT 2006 Kevin Karplus I made a tiny change to undertaker, so that H-bond geometry calculation would not crash on distance 0 between donor and acceptor, though this is not the *right* fix, since the Hbond should not be being scored in this case. Sun Jul 2 15:17:12 PDT 2006 Kevin Karplus The try2 costfcn likes SAM_T06_server_TS1 best (even better than try2_opt2), followed by PROTINFO_TS3, PROTINFO_TS1, FUGMOD_TS1-scwrl, PROTINFO_TS2, ... None of the ROBETTA models chose the try2 alignment. Sun Jul 2 15:24:36 PDT 2006 Kevin Karplus try3 finished and produced a much nicer C-terminus than try1-opt2, but I like the N-terminus of try1-opt2 better. Sun Jul 2 15:29:15 PDT 2006 Kevin Karplus I made a chimera-try3-try1, that has residues 1-14 of try1-opt2 and 15-149 from try3-opt2. There is a bit of a break added before V15, but it is not as bad as some of the existing breaks: Conformation[15] chimera-try3-try1.pdb.gz has 14 breaks chimera-try3-try1.pdb.gz breaks before (T0331)G131 with cost 5.71389 chimera-try3-try1.pdb.gz breaks before (T0331)Q130 with cost 5.32385 chimera-try3-try1.pdb.gz breaks before (T0331)V15 with cost 1.9208 chimera-try3-try1.pdb.gz breaks before (T0331)I74 with cost 0.441491 Because of the mismatch, it does not score quite as well as either try3 or try1, but I'll try optimizing it for try4. Sun Jul 2 15:45:34 PDT 2006 Kevin Karplus try4 (using constraints from try3-opt2 and chimera-try3-try1 as the starting point) started on lopez. Sun Jul 2 17:56:07 PDT 2006 Kevin Karplus I rather like try4-opt2, though it could be packed a bit more tightly. I'm tempted to do a polishing run starting from the gromacs models, because they are close to good-scoring model, but with lower clashes breaks, but score badly enough that undertaker can improve them fairly easily. Sun Jul 2 18:02:40 PDT 2006 Kevin Karplus try5 (polishing the try3 and try4 gromacs models) started on lopez. Sun Jul 2 19:31:00 PDT 2006 Kevin Karplus try5-opt2 is new best for try5 costfcn and rosetta likes repacking it best. In fact, that worked well enough that I'll try the same trick again, of starting from the gromacs models. Sun Jul 2 22:15:57 PDT 2006 Kevin Karplus Not so successful this time---try6-opt2 is not significantly better scoring than try5-opt2. Maybe I should do one more polishing run starting from all models. Nahh---the differences between try5 and try6 are so small when they are superimposed that we are probalby just looking at noise in the cost function at this point. I'll submit try6-opt2 best-scoring try2-opt2 alternative threading, using helical insert try5-opt2 similar to model 1, different clash/break tradeoff align1 t2k alignment to 1g79A align5 t06 alignment to 2fhqA Sun Jul 2 22:48:41 PDT 2006 Kevin Karplus Preliminary submission done, but I really should optimize in a dimer context. Mon Jul 3 16:57:32 PDT 2006 Grant Thiltgen I am working on making a dimer for this target. I used 2fhqA as a template for try1 in the dimer. I used the make-dimer.under file. I made a dimer of try6-opt2 for try1. I also decreased the num_gen, gen_size, and super_num_gen in the try1.under file in order to decrease run time. try1 for the dimer started on camano. Mon Jul 3 17:22:00 PDT 2006 Grant Thiltgen Oops! I forgot to change constraints in the try1.costfcn. I'm adding the sheet and helix constraints from the try6 model. I also removed the pred_alpha from the try1.costfcn. I restarted try1 on camano. Mon Jul 3 21:49:12 PDT 2006 Grant Thiltgen try1 looks alright. I'm going to work on refining it now that I have the basic one made. I am also creating a dimer of try5-opt2 and try2-opt2. try2 will be a run for try5 keeping the try5 constraints and try3 will be a run for try2-opt2 keeping the try2 constraints. Both of the dimers were using 2fhqA for a template. I'm not sure the try2-opt2 dimer looks right, but I'm going to give it a try. I used make-dimer2.under to make the try5-opt2 dimer and I used make-dimer3.under to make the try2-opt2 dimer. Both try2 and try3 are started on whidbey. Tue Jul 4 07:04:58 PDT 2006 Grant Thiltgen The dimer based on try6-opt2 scores the best, but just slightly above the try5-opt2 dimer. I am setting up try4-try6. Try4 is a polish of try1, try5 is a polish of try2, and try6 is a polish of try3. I am increasing the weights of break to 100, soft clashes to 40, dry 6.5 to 25 and dry8 to 20. try4 and try5 started on whidbey. try6 started on shaw. Tue Jul 4 09:58:56 PDT 2006 Kevin Karplus I superimposed the 6 dimer models into dimer/best-models.pdb.gz (not trying to figure out what is best yet, just in order of creation). I chose the key residues of chain A to initiate the superposition, so that the chain As would superimpose well. All the dimers are roughly the same *except* try5, which has a totally different location. Note that try5 is based on try2, but has move the second monomer extremely. try6 is similar to try3, and tries 1,2, and 4 are similar. Tue Jul 4 10:28:04 PDT 2006 Grant Thiltgen I used the gromacs0 packing for try4-try6. They didn't seem to improve much over the previous, so I'm running try7-try9. Basically try7 is the same as try4, but using the try1-opt2 instead of the gromacs packing. Try8 is using try2-opt2 and try9 is using try3-opt2. I'm using the same costfcns as before. Try 7 and try8 are started on whidbey try9 on shaw. Tue Jul 4 10:57:46 PDT 2006 Kevin Karplus You may want to turn constraints off (or way down) for polishing runs. I was looking for Hbonds that might form between the monomers, but I did not find any convincing enough to be worth forcing. Tue Jul 4 17:45:54 PDT 2006 Grant Thiltgen All the new polishing runs score better than the originals for the unconstrained costfcn. I am not sure how they score in comparison to the original models. I'm not sure if I did this right, but I manually removed one of the monomers from each dimer. I copied them into the T0331/decoys/ directory in order to score them. The two monomers, try7-opt2-monomer and try8-opt2-monomer score better (slightly) than the two monomers. try7-opt2-monomer is the monomer from a polishing run of try1 in the dimer directory. This is based on the try6-opt2 model for the monomer. This scores better than the monomer for try6-opt2. Try8-opt2-monomer is the monomer from the polishing run of try2 in the dimer directory. This is based on the try5-opt2 model for the monomer. This also scores better than the monomer for try5-opt2. Try9-opt2-monomer is the monomer from the polishing run of try3 in the dimer directory. This is based on the try2-opt2 model for the monomer. This one also scores better than the try2-opt2 model (slightly). These models may or may not be better than the original. If you want to exchange them for the originals, I'm sure they haven't changed that much. Softclashes are a bit higher, but the pred_alpha values seem a bit lower. If you want to change them out, the order should be: try7-opt2-monomer try9-opt2-monomer try8-opt2-monomer undertaker-align.pdb model 1 undertaker-align.pdb model 5 This is just substituting the new monomers optimized from the dimer for the old monomers. Tue Jul 4 18:17:24 PDT 2006 Kevin Karplus You took a hard approach for extracting the monomers, then left them in the wrong place. There is no need for ProteinShop for this task! 1) You can do make decoys/T0332.try9-opt2.unpack.pdb.gz to create a multimer with the different chain ids for each monomer. 2) If you do "make -k T0332.mult9" instead of "make -k T0332.do9" in the dimer directory, then the unpacked version will be made automatically. 3) You should never leave a monomer in the dimer/decoys directory. Move them directly to the decoys/ directory. (You should probably eliminate the dimer/decoys/Proteinshop directory, unless you are planning to use Proteinshop for real on the dimer.) Rosetta's favorite dimer is decoys/T0331.try9-opt2.unpack.gromacs0.repack-nonPC.pdb.gz decoys/T0331.try7-opt2.unpack.gromacs0.repack-nonPC.pdb.gz decoys/T0331.try8-opt2.unpack.gromacs0.repack-nonPC.pdb.gz With the unconstrained costfcn (adding in the KnownBreak M150 to the default unconstrained.costfcn), the favorite dimers are try7-opt2, try1-opt2, try8-opt2 Of the dimers, I dislike try3, try6, try9 (which all have a clumsy N-terminus) and try5, which is in the wrong place. The rest are very similar, though I prefer try1, try2, try7, try8 (not try4). So overall, my current favorite monomer is try7-opt2-monomer = dimer/decoys/T0331.try7-opt2.unpack.pdb.gz chain A. The order that you have given to the models try7-opt2-monomer try9-opt2-monomer try8-opt2-monomer undertaker-align.pdb model 1 undertaker-align.pdb model 5 is what I'll use for the preliminary submission. Tue Jul 4 18:44:15 PDT 2006 Kevin Karplus Preliminary submission done, with comment For T0331, our biggest decision was not between templates, but between alignments. Did we leave out a strand or a helix? The t2k multiple alignment appears to have some misalignment at the C-terminus (key residues in alignment doesn't match with target sequence), so we preferred the t06 multiple alignment. We also did some optimization in a dimeric context (based on dimer 2fhq[AB]). Model 1 is try7-opt2-monomer, our best-scoring model, optimized as a dimer. Model 2 is try9-opt2-monomer, a model from the alternative alignment, which keeps the helix but discards the C-terminal strands. It was also optimized as a monomer, and rosetta prefers repacking the sidechains on it (probably because the less-packed model allows more freedom for the sidechains). Model 3 is try8-opt2-monomer, which is very similar to Model 1. It was also optimized as a dimer. Model 4 is sidechain replacement by SCWRL for a t2k alignment to 1g79A. Model 5 is sidechain replacement by SCWRL for a t06 alignment to 2fhqA. Mon Jul 24 14:25:16 PDT 2006 Grant Thiltgen I took a look at things, and I'm not really sure there's much more we can do to this model. The preliminary submission is probably good.