Tue Jun 27 10:37:29 PDT 2006 T0348 Make started Tue Jun 27 10:38:40 PDT 2006 Running on lopez.cse.ucsc.edu BLAST gets only very weak hits in PDB: 1tzna, 1t6bY, 1shuX, 1shtX, 36% id over 28 residues, E-value 12). The target is only 68 residues, so may be a feasible ab-initio target. Tue Jun 27 14:34:22 PDT 2006 Kevin Karplus Top score with the HMMs is 2ct7A E-value 0.23. There are also several hits to g.41.3.*, starting with 1pft E-value 0.43, a zinc beta-ribbon. We should definitely look for metal-binding sites, probably involving C11, C14, maybe C29. Tue Jun 27 20:44:49 PDT 2006 Kevin Karplus The try1-opt2 model and the alignments are in good agreement, and there does appear to be a metal binding site for the three cys residues. D32 seem to be part of the same site. We should turn off "maybe_ssbond" and turn up "maybe_metal", perhaps adding a contraint or two for D32.OD1 Fri Jun 30 17:11:17 PDT 2006 Cynthia Hsu Following Professor Karplus's advice,"maybe_ssbond" value was set to 0 and "maybe_metal" was increased to 4. Examining the near and burial scripts seemed indicated that hydrophobic residues in the helices (Phe5, Leu9, Ile42, Leu46) were exposed in the current try1-opt2 model. To adjust for this, "phobic_fit" cost was increased to 5. The near script also indicated that Ile38, located on the third beta strand in the trough, was more exposed than it should be. My hypothesis is that if we bring the end of this strand closer to the adjacent beta sheet, it will not only solve the problem of exposure but also bring the D32 residue closer to the three cys residues at the metal bonding site. I'm a bit unsure as to how to do this, but I've increased the "hbond_geom_backbone" to 15 and the "hbond_geom_beta" to 80 in hopes of getting it to work. No changes were made to the try2.under file; I'm not sure what would have been appropriate in this case. Sat Jul 1 12:39:14 PDT 2006 Cynthia Hsu I've run the make file on lopez; I've been having trouble with make files for my targets so hopefully this one will be okay. Mon Jul 3 12:44:36 PDT 2006 Crissan Harris, Cynthia Hsu try2-opt2 looks like a sick snake, so we decided to start over using the original costfcn in try1.costfcn. However, there were the following differences: As before, we set "phobic_fit" to 5, "maybe_metal" to 4, and "maybe_ssbond" to 0. Under Grant's direction, we also included SheetConstraint and Helix constraints. In addition, we followed Prof. Karplus's advice in the earlier README and added several constraints to the Aspartate Cysteine metal-binding regions: Constraint D32.OD1 C11.CB -10. 7.0 14.0 1.2 Constraint C14.CB C29.CB -10. 7.0 14.0 1 Hopefully this will bring the two loops closer together. We are still concerned about the exposure of the three helices, but we wanted to see how these changes affected the packing before experimenting with that. try3 is currently running on orcas. Mon Jul 3 13:57:33 PDT 2006 Crissan Harris, Cynthia Hsu try3 looks even worse than try2. We set the HelixConstraints to 2 for all helices, to prevent them from being garbled up. For the individual residues in the metal-bonding region, we also increased the constraints as follows: Constraint D32.OD1 C11.CB -10. 7.0 14.0 4 Constraint C14.CB C29.CB -10. 7.0 14.0 2 try4 is currently running on Whidbey. Wed Jul 5 08:05:27 PDT 2006 Kevin Karplus I picked up the server models and scored them with unconstrained.costfcn Best scoring are our try4-opt2 and try1-opt2, then our SAM_T06_server_TS1 Best non-SAM models are ROBETTA_TS1, ROBETTA_TS3, ... Thu Jul 6 14:15:27 PDT 2006 Crissan Harris, Cynthia Hsu We've decided to start performing polishing runs on try4, as packing will hopefully decrease foaminess and cover some of the exposed regions. "dry6.5" was increased to 25 and "dry8" was increased to 20. Breaks were small, but we still increased it to 100. "soft_clashes" were increased to 60 try5 is currently running on shaw. Thu Jul 6 16:27:18 PDT 2006 Crissan Harris, Cynthia Hsu We had problems, as the undertaker couldn't seem to find the files we wanted to polish. We've modified try6.under so that it would specifically read input from T0348.try4-opt2.pdb. try6 is currently running on shaw. Fri Jul 7 12:02:35 PDT 2006 Crissan Harris, Cynthia Hsu try6-opt2 scored the best in an unconstrained environment, followed by try6-opt1. Looking at four best models superimposed in RasMol, we've decided to continue making improvements on try6-opt2, but to change try7.under so that it is no longer a polishing run. We copied the try6.costfcn file to try7.costfcn, and then, based on the rr script's results, added the following constraints: Constraint D25.CB L53.CB -10. 7.0 14.0 1 Constraint F36.CB L46.CB -10. 7.0 14.0 5 Constraint F20.CB M45.CB -10. 7.0 14.0 3 try7 is currently running on shaw. Fri Jul 7 13:36:28 PDT 2006 Crissan Harris, Cynthia Hsu The added constraints to try7 did not seem to have any of the desired effects. We increased the weights on the individual constraints as listed below, added an additional constraint to I38 and E37, and increased the overall weight of the constraint from 10 to 20. Constraint D25.CB L53.CB -10. 7.0 14.0 1 Constraint F36.CB L46.CB -10. 7.0 14.0 8 Constraint F20.CB M45.CB -10. 7.0 14.0 5 Constraint I38.CB E37.CB -10. 7.0 14.0 5 Fri Jul 7 14:50:50 PDT 2006 Crissan Harris, Cynthia Hsu Both try7 and try8 appeared very convoluted, with the helices from 44 to 68 protruding into the environment. We've decided to modify the try6 model for try9. We hoped that by lengthening the fourth sheet in the half-barrel, those two helices would be drawn closer together, so we modified the sheet constraint on the fourth sheet by changing the parameters to the following: SheetConstraint (T0348)A35 (T0348)K39 (T0348)L46 (T0348)I42 hbond (T0348)K39 1 try9 is currently running on Whidbey. Fri Jul 7 16:23:39 PDT 2006 Crissan Harris, Cynthia Hsu Instead of forming a sheet, residues 35 to 39 actually turned into a helix. We made the following changes to the weights of the sheet constraints: SheetConstraint (T0348)L18 (T0348)K22 (T0348)C29 (T0348)D25 hbond (T0348)V19 2 SheetConstraint (T0348)D25 (T0#ReadConformPDB T0348.try10-opt2.pdb //similar to try12348)C29 (T0348)I38 (T0348)L34 hbond (T0348)L27 8 SheetConstraint (T0348)A35 (T0348)K39 (T0348)L46 (T0348)I42 hbond (T0348)K39 6 We also shortened the helix constraint from P43 to L53 to only span from residues E47 to L53. Then we raised the weight on the constraints overall to 15. try10 is currently running on camano. Sat Jul 8 14:26:18 PDT 2006 Cynthia Hsu try10 did successfully move the helix from residues 54-68 closer to the protein. However, the helix from residues M44 to 54 seemed to unravel itself slightly, and as a result protruded even further from the environment. The burial script indicated that one side of the trough was also exposed, so I thought it might be possible to bring the unwound portion of the helix (residues 44-46) to pack by the trough. I've decided to increase sidechain packing to 8 in hopes of accomplishing this; we'll see how it turns out. The burial script also seemed to indicate that there were still some regions that were more exposed than they should have been, mostly in the region of two loops (C14-G16 and C29-L34). I'm hoping that by raising the constraints between C14-C29 to 4, the gaps between these might be closed off. Sat Jul 8 20:41:46 PDT 2006 Cynthia Hsu Apparently when I was editing try11, I had originally planned on restricting reisdues 44-47 into a helix, and had forgotten to return this constraint to the original (a HelixConstraint from E47 to L53 instead of from M44) way after I had changed my mind. With this in mind, try12.costfcn is identical try11.costfcn, except with the helix constraints actually the way they were actually intended in try11 (unchanged from try10). try12 is currently running on lopez. Mon Jul 10 22:37:22 PDT 2006 Cynthia Hsu Both Rosetta and Undertaker seem to score try11 as relatively high; try11 is the first choice model by Rosetta while in Undertaker try6 scores better than try11 only in the matter of the "wet6.5", the "near_backbone", "phobic_fit", and the dry weights. In view of this, the following changes were made. SetCost wet6.5 17 near_backbone 10 way_back 5 dry5 20 dry6.5 30 dry8 22 dry12 6 phobic_fit 7 try13 is currently running on Whidbey. Tue Jul 11 11:25:59 PDT 2006 Crissan Harris, Cynthia Hsu try12 was removed from our best models, given the large uncoiled strand from G41-L46, which left the helices from 46 to 68 protruding into the environment. We also added try8 to best models, given its high scores. After examining our models and reviewing the README, we decided that it would be in our best interest to attempt to polish try11 and try6. For try14, which is a polishing run of try11, we copied the try13.costfcn, since the differences between this one and try11 made it a feasible cost function for a polishing run of the try11 model. We returned "wet6.5" to its original value of 15, and "phobc_fit" to its original value to 5, and lowered "near_backbone" to 7. try14.under was turned into a polishing run to accept try11-opt2 as an input. try14 started on orcas Tue Jul 11 11:58:55 PDT 2006 for try15, which is a polishing run of try6, we copied the try6.costfcn. We then raised "n_ca_c" to 7, "bad_peptide" to 12, "soft_clashes" to 80, and "break" to 200. try15 was started on whidbey Tue Jul 11 12:07:25 PDT 2006. Tue Jul 11 14:05:16 PDT 2006 Crissan Harris, Cynthia Hsu try14 had unwound helices, very very large breaks, and unconnected beta strands. The entire protein has disintegrated beyond recognition. For the try16.costfcn, we copied try14.costfcn and returned "near_backbone" to 5, "dry12" to 5, and raised "break" to 150. Like try14, try16 read try11-opt2 as an input. try16 is currently running on camano. try15-opt1 is the second highest scoring model in undertaker, but performs slightly worse than try6 (0.23), particularly in the "sidechain" and "dry8" costs (generally, however, it performs better in other factors such as "breaks" and "soft_clashes"). In view of this, we copied try15.costfcn to try17.costfcn and raised dry weights as follows: dry5 20 dry6.5 30 dry8 23 We then modified try17.under to accept try15-opt1 as its input. try17 is currently running on squawk. Wed Jul 12 18:23:21 PDT 2006 Crissan Harris, Cynthia Hsu try16 scored very very poorly. try17, on the other hand, did the best out of all our current models, although try6-opt2 still scores the second best. It still looks very foamy, but we're not sure how to modify it so we've decided to wait until someone gives us a suggestion as to how to decrease the foaminess. Wed Jul 12 18:50:05 PDT 2006 Crissan Harris, Cynthia Hsu On second thought, we decided it might be worth polishing Rosetta's favorite model, try11-opt2.gromacs0.repack-nonPC.pdb.gz. In view of this, we copied try11.costfcn to try18.costfcn, then changed the try18.under file to accept the appropriate gromacs model. try18 is currently running on lopez. Thu Jul 13 10:22:51 PDT 2006 Crissan Harris, Cynthia Hsu We reexamined the scores for our try17 models, and observed that try17 did significantly worse than try18 in "n_ca_c" cost. Because of this, we raised, the weight of it from 7 to 10. We also decided to polish opt1, instead of opt2, since the difference in scores did not seem very significant and opt1 was much better in terms of breaks. try19 is currently running on shaw. try18 performed the best overall in Undertaker, but it scored worse in all the dry weights, "wet6.5", "near_backbone", "way_back", "bad_peptide", and "break". For try20.costfcn, we set "dry5" to 20, "dry6.5" to 30, "dry8" to 23, and "break" to 160. We also raised "bad_peptide" to 13. try20 is currently running on lopez. Thu Jul 13 11:16:29 PDT 2006 Crissan Harris, Cynthia Hsu We've updated the superimpose-best.under file to represent the best of a diverse selection of folds: ReadConformPDB T0348.try19-opt1.pdb //ReadConformPDB T0348.try6-opt2.pdb //very similar to try19 ReadConformPDB T0348.try20-opt1.pdb //ReadConformPDB T0348.try11-opt2.pdb //very similar to try20 ReadConformPDB T0348.try8-opt2.pdb ReadConformPDB T0348.try3-opt2.pdb ReadConformPDB T0348.try1-opt2.pdb We've decided to wait for further instruction before proceeding with new tries. Sat Jul 15 16:16:19 PDT 2006 Kevin Karplus If try20-opt1 and try11-opt2 are so similar, it might be better to include try11-opt2.gromacs0.repack-nonPC in the list, since it is Rosetta's favorite, though try20-opt2 is the favorite of the unconstrained costfcn. (I'm not that fond of either one---I wonder why they are scoring so well.) Do we really want such a large weight on maybe_metal in the later costfcns? The HIS tag seems to be getting in the way of folding for try20 and try11, so I'm starting a run on the subdomain M1-K60 on lopez. I'll superimpose it on the best whole-chain models and see if there is agreement with any of them. I may have to tack on a HIS tag and reoptimize. Sat Jul 15 20:32:25 PDT 2006 Kevin Karplus The M1-K60 run has several alignments that agree on the structure for V10-P37, but M1-L9 and I38-K60 are going to have to come from fragments. C11, C14, and C29 *do* seem to form a metal-binding site, with D32 probably also participating. In addition the maybe_metal weight, we might want to put some constraints on D32 (either OD1 and OD2 or CG) and the SG atoms. I superimposed M1-K60/try1-opt2 on the existing models, and it is in good agreement with try19-opt1 up to about M44. If I do decide to paste on a HIS tag, I could take it from try8-opt2, which has a roughly compatible orientation. Perhaps I should do another run from alignments on the protein without the his tag, but with a costfcn that reflects the metal-binding site and the structure through M44. Sat Jul 15 21:24:02 PDT 2006 Kevin Karplus M1-K60/try2 started on cheep, with longer than usual optimization run. Starting from alignments, but with strong constraints on SG atoms. I also made a metal.costfcn in the main directory, which prefers try20-opt2, try18-opt2, try19-opt1, try17-opt1, ... Sat Jul 15 22:10:04 PDT 2006 Kevin Karplus In M1-K60, both try2.costfcn and Rosetta agree that try2 is better than try1. Sat Jul 15 22:28:20 PDT 2006 Kevin Karplus I made a chimera of M1-K60/try2-opt2 and a badly placed HIS tag (from try8-opt2). (chimera-M1-K60-2-8) and am optimizing it (with constraints) as try21. Interestingly, it has the first strand upside down from our previous predictions. I don't think this is *as* good, based on the "near-backbone" burial predictions, but it does make for a plausible dimerization interface. Sat Jul 15 23:14:16 PDT 2006 Kevin Karplus try21, though not as highly polished as some of our other models, still comes up to 4th place in rosetta's estimation, (after try11, try9, and try12). With the try21 costfcn, the order is try20-opt2, try18-opt2, try19-opt1, try17-opt2, try6-opt2, try21-opt2 The amazing thing is that try21-opt2 almost exactly duplicates try1-opt2, but with better spacing of the SG atoms (if maybe_metal is right about the spacing). I'll replace try1 with try21. I will submit try20-opt2 best-scoring with "metal" costfcn try11-opt2.gromacs0.repack-nonPC rosetta's favorite, very similar NOT try18-opt2 (too similar) NOT try9-opt2.gromacs0.repack-nonPC (too similar) try19-opt1 somewhat different, still high scoring try21-opt2 made initially without HIS tag. Adding his tag produced model very similar to the automatic try1 model, but with better scores. NOT try8-opt2 too similar to try21 NOT try12-opt2.gromacs0.repack-nonPC UGLY try3-opt2 Sun Jul 16 00:04:16 PDT 2006 Kevin Karplus Submission done with comment For T0348, we mostly just ran optimizations from the set of possible alignments to templates, with different variations on the cost function. We did also try doing a couple of optimizations without the HIS tag, but when we stuck the HIS tag back on and reoptimized, we got models almost identical to models built with the HIS tag in the first place. Model 1 is try20-opt2, the best-scoring with several of the cost functions we tried. Model 2 is try11-opt2.gromacs0.repack-nonPC, the rather similar model that scored best with rosetta's cost function, after rosetta repacked sidechains. Model 3 is try19-opt1, a somewhat different model that is still high scoring. Model 4 is try21-opt2, the model optimized without the HIS tag (just M1-K60), then reoptimized with the HIS tag added. It comes out surprisingly close to the automatic try1-opt2 model, but both undertaker and rosetta like it better. Model 5 is try3-opt2, a rather different model that does not score all that well but gives us some diversity in our set of submissions. Sun Jul 16 00:04:53 PDT 2006 Kevin Karplus If anyone comes up with new models, or wants to replace one of the submitted models with a different one, send me e-mail. Mon Jul 17 13:11:47 PDT 2006 Crissan Harris, Cynthia Hsu We looked over the models, and with the exception of try3, which is only in the list for the purpose of diversity, we approve of all of the models that were used. Sun Sep 3 13:31:14 PDT 2006 Kevin Karplus This is one of the few targets for which we did much better than the Zhang server, though we did only a tiny bit better than the Robetta server. We did not submit our best models (try1-opt1-scwrl by GDT or try10-opt2.gromacs0 by our evaluation function), but our model4 (try21-opt2) is not far behind. ROBETTA_TS1 does best by our evaluation function, though we beat it slightly on GDT. We did quite well on P12-M44, but messed up on the C-terminus, having overdone the helices and missed the final strand. The CYS residues at the metal-binding site show very good agreement, though the D32 sidechain is a bit too far away. Model 5 (try3-opt2) was indeed the worst of the ones we submitted, but model 4 (try21-opt2) was the best we submitted. We would have done better subbmitting try1-opt2 than any we did submit. Our best alignment of the top 5 was align3, a t04 alignment to 1vk6A. Sun Mar 2 19:58:21 PST 2008 Kevin Karplus The CASP programs rated our model1 best.