Fri Jul 7 09:11:36 PDT 2006 T0364 Make started Fri Jul 7 09:12:40 PDT 2006 Running on lopez.cse.ucsc.edu Sat Jul 8 00:01:28 PDT 2006 Kevin Karplus No good hits with BLAST in PDB (best is 2gm3A, E-value 1.1). [ Mon Jul 24 19:42:45 PDT 2006 Kevin Karplus HMMs get excellent hit to 1s5uA, 1z54A, ... SCOP domain d.38.1.1 ] Mon Jul 17 16:14:59 PDT 2006 Navya swetha Davuluri, Crissan Harris We looked at 362 and it looks a lot like 364. It was aligned from the same template as 364 which is 1s5u. Before we dimerize this we will optimize it. Residues 13-41 are predicted to be part of a helix, but try1-opt2 has 13-18 as a strand and 19 , 20 as part of a turn. We are going to add a helix constraint to make them part of the helix. We increased the weight on constraints to 20 and the weight on added helix constraint to 5. try2 is running on abyss Tue Jul 18 11:14:56 PDT 2006 Crissan Harris try2 did not make the helix as we had hoped. However, in viewing this model further I'm not so sure making this helix is the right thing to do. The models in the best-models.pdb all show the same break in the helix and in looking at the pdb website for the top 5 models it seems that this may be involved in the binding site of a dimer, quatramer, or octamer. The structures of the top alignments all agree with our model to a very large extent. The only areas of question are E137-A156 (the C-terminus helix), and the loops from A117-S121 and E102-Q138, and some variations in the helix H126-L135. In viewing the helix from E137-A156 it seems difficult to determine exactly where it should go. Since it is predicted to be exposed and is currently on the opposite side of the dimer interface I will leave it as it is for now since I don't think it will interfere with the protein function. I decided to run another try using the alignments with the original sheet and helix constraints from try1-opt2. I have also increased the dry weights slightly just to improve the packing before dimerizing. The weights were altered as follows: dry5 20 dry6.5 25 dry8 17 Also since there is only one Cys I removed "maybe-ssbond". Since there are a large number of His but they do not seem to be localized in a general area I have left the "maybe-metal" at its default value. try3 started on whidbey at Tue Jul 18 12:36:25 PDT 2 [Mon Jul 24 19:45:05 PDT 2006 Kevin Karplus Maybe_metal only affects CYS residues.] Tue Jul 18 14:35:03 PDT 2006 Crissan Harris try3 scored better than try2 in the unconstrained scores. However, in viewing try3, I've decided that I like the placement of the helix from E137-A156 better in try2 since it gives the overall protein a better shape. In try 3 the helix is protruding into the environment more. I have decided to run one more try with the alignments before polishing. try4.costfcn was copied from try2.costfcn with removing the specific helix constrain from try2. try2 did worse in dry5, dry8, so these weights were altered as follows: dry5 17, dry8 17 I also increased "phobic_fit" to 3 in the hopes that it will keep a better shape than in try3. Also I again removed the "maybe-ssbond" constraint Tue Jul 18 16:27:09 PDT 2006 Crissan Harris Try4 looks better that try3 but still not as good as try2 (as far as the overall shape due to the placement of the 137-156 helix). Try4 also scores the best in the unconstrained scores. Rosetta liked try2 the best followed by try1 and try4. At this point I think I will try polishing try4 and try2, then attempt to dimerize. try5 is a polishing run of try2-opt2 I increased the dry weights as follows: dry5 20 dry6.5 25 dry8 17 I also increased the soft_clashes to 60 and the breaks to 150 as well as removing the maybe_ssbond. I also still included the sheet and helix constraints from try1-opt2. try5 is currently running on camano try6 is a polishing run of try4-opt2 I increased the dry weights as follows: dry5 20 dry6.5 25 I also increased the soft-clashes to 60 and the breaks to 180. try6 is currently running on whidbey Tue Jul 18 18:41:14 PDT 2006 Crissan Harris try5 and try6 came out pretty well and scored well in the unconstrained scores. I decided to go ahead and dimerize them to see how they turned out. make-dimer1.under makes a dimer from try5 and make-dimer2.under makes a dimer from try6. Both dimers seemed to look pretty good so I am goign to try optimizing them with a general costfcn. The try1 costfcn and .under file are for the dimer-1s5uA-from-try5-opt2.pdb And try2 costfcn and .under file are for the dimer-1s5uA-from-try6-opt2.pdb try1(mult) is currently running on whidbey try2(mult) is currently running on abyss Wed Jul 19 17:36:05 PDT 2006 Navya swetha Davuluri The dimer from try6 does not have the h-bonds between two monomers formed. I am going to add a sheet constraint between these two monomers to form a h-bond between these two monomers. try3(mult) is running on bark Thu Jul 20 11:57:06 PDT 2006 Crissan Harris try1 and try2 (of the dimer) came out terrible. I think that I am doing something fundamentally wrong with one of the files, I will re-read the README and see if I can find the problem. Also try3 did not complete its run. I am not sure what the error was, but in viewing the try3.under file I realized that the "try2"s had not been changed to "try3"s, so this may have been the problem, and in either case this means that try2 has been written over (I'm not sure that this is really that big of a deal since try2 was probably pretty bad anyway). So at this point I will try to figure out what I am doing wrong with the dimer optimzation and until then I will not do any more runs since I don't think they would do any good. Thu Jul 20 12:11:50 PDT 2006 Crissan Harris Okay, so I decided to look at the try1.costfcn and the try1.under files from the T0284 dimer directory (the example model in the README). I have decided to do another run modeling my files after these. thus in try4.costfcn I commented out the "CreatePredAlphaCost" and removed the "pred-alpha", "missing_atoms", and "maybe_ssbond" constraints. In the .under file I tried to model the one from T0284 so I made the following changes: I changed the weight of "TweakMultimer" to 0, the "num_gen" to 40, "gen_size" to 80, "super_iter" to 3, and "super_num_gen" to 200. In the InitMethodProbs I deleted "FixOmega" And I set it up to ReadConformPdb dimer-1s5uA-from-try5-opt2.pdb If this works then I will do the same thing to optimize the dimer-1s5uA-from-try6-opt2.pdb, but there is not point in running it before I know if it will work. try4(mult) is currently running on whidbey. Thu Jul 20 13:39:06 PDT 2006 Crissan Harris try4 is not quite done but I took a peek at the opt1 in rasmol and it seems to be working (that is it isn't losing all of its shape) so I will run try5 with the same costfcn and .under file as try4 but set it up to ReadConformPdb dimer-1s5uA-from-try6-opt2.pdb. try5(mult) is currently running on orcas Thu Jul 20 14:15:24 PDT 2006 Crissan Harris try4 did very well in both the unconstrained scores and under try4's constrained scores. Although it scores very well compared to the others (which of course are terrible) it still looks very foamy and needs to be packed tighter. So try6 will be a modification of try4 with increased packing. The dry weights were changed as follows: dry5 20 dry6.5 30 dry8 25 dry12 7 The "soft_clashes" were raised to 60 and the "break"s were raised to 200. try6(mult) is currently running on camano. Thu Jul 20 15:28:00 PDT 2006 Crissan Harris try5 also looks pretty foamy, and I am not sure that I like its shape as much as try4's shape. However try5 does score better than try4 in the unconstrained scores as well as the try5 constrained scores and the try4 constrained scores. I decided to try to pack try5 more. I changed the costfcn as follows: dry5 20 dry6.5 30 dry8 25 dry12 7 phobic_fit 3 soft_clashes 70 break 200 try7(mult) is currently running on orcas. Fri Jul 21 10:41:20 PDT 2006 Navya swetha Davuluri I tried to create hbonds between the two monomers on Wednesday and it did not work as I overwrote try2. So I am going to start a new run including the sheet constraint for forming hbonds between these two monomers. I am using try7.costfcn and adding the sheet constraint and try7.under try8 is running on shaw Fri Jul 21 12:46:00 PDT 2006 Navya swetha Davuluri try8 did form the hbonds between the two monomers, but it created 3 breaks in the backbone. I am starting a new run with increased weight on breaks and soft_clashes. I have increased the weight on breaks to 300 and on soft_clashes to 100. try9 running on lopez. Fri Jul 21 15:15:35 PDT 2006 Navya swetha Davuluri try9 did fix the breaks but did not form h-bond. And there is one huge break at 157. I realized that the known break was given at 156th residue when it has to be at 157th residue. This probably is causing all the problems. I am now going to correct it and start a new run again. I am also increasing the weight on the sheet constraint that forms the hbond between the two monomers. try10 is running on peep Mon Jul 24 13:50:56 PDT 2006 Crissan Harris I was going over the readme and realized that try 8-10(mult) are all modifacations from the try6 monomer, and that try6(mult) was the last try from the try5 monomer. Since I actually liked the shape of the try5 monomer better than the try6 monomer I have decided to go back and do some more optimizations starting from the try6(mult). I copied the cost fcn from try10 and made changes as follows: dry5 25 dry6.5 35 soft_clashes 80 break 200 I also left in the sheet constaint for adding the hbonds between the two monomers and the KnownBreak location was fixed to M157. try11.under was set up to take in T0364.try6-opt2.pdb (mult). try11(mult) is currently running on camano. I also looked at the try8, try9, and try10 .under files and realized that the ReadConformPDB was nevcer changed from try5(mult). In view of this I have decided to make a try12(mult) that will ReadConformPDB T0364.try7-opt2.pdb which is still the best scoring model which originates from the try6 monomer. I will use the same costfcn from try11(mult). try12(mult) is currently running on abyss. Mon Jul 24 16:27:02 PDT 2006 Crissan Harris try11 failed to create the hbonds between the sheets at the interface and it pulled some of the helices out of the correct locations so that they are now protruding into the environment. try12 looks pretty good. The hbonds were created and the only thing that seems to need improvement is that it is still a little foamy. Neither try11 nor try12 really had problems with breaks (besides a few very minor ones <0.5), Although try12 scored very poorly in breaks and try11 scored much better. try12 is now the highest scoring in the unconstrained scores and try11 scores very poorly. I would like to make another attempt to optimize the try6(mult) model and also try to increase the packing on the try12(mult) model. Mon Jul 24 19:48:49 PDT 2006 Kevin Karplus It can also be instructive to look at dimer/decoys/grep-best-rosetta. Rosetta like best decoys/T0364.try10-opt2.unpack.gromacs0.repack-nonPC.pdb decoys/T0364.try12-opt2.unpack.gromacs0.repack-nonPC.pdb It might be worth doing a polishing run from each of these, if they look reasonable (I've not looked at them yet---download is a bit slow here in the hotel). Tue Jul 25 12:37:03 PDT 2006 Crissan Harris As Professor Karplus suggested I will do a polishing run of both of Rosetta's favorites and also of try4-opt2. I decided to polish try4 instead of try6 since it scored better and obviously the last polishing that I did of try6 did not come out very good. Thus try13 will be a polishing run of try4-opt2.pdb try14 will be a polishing run of try10-opt2.unpack.gromacs0.repack-nonPC.pdb try15 will be a polishing run of try12-opt2.unpack.gromacs0.repack-nonPC.pdb try13(mult) is running on vashon try14(mult) is running on camano try15(mult) is running on peep Tue Jul 25 14:07:16 PDT 2006 Crissan Harris dimer/try13, try14 and try15 still did worse than the trys that they were set up to polish. However Rosetta now likes decoys/T0364.try14-opt2.unpack.gromacs0.repack-nonPC.pdb.gz the best soat least there was someimprovement there. And also try13 did manage to create the hbonds, so that too is an improvement. I am slightly concerned about the burial of try14 and try15. They both look p[retty good under the near script but there are a lot of regions near the interface that are predicted to be buried under the burial script but that are actually exposed. try13 also has the same problem but not quite as bad as try14 and try15. At this point I think that our top models are as follows: Undertaker's top 2: dimer/T0364.try12-opt2.pdb.gz // 1st best scoring, only has minor breaks dimer/T0364.try7-opt2.pdb.gz // 2nd best scoring, only has minor breaks Rosetta's top 2: dimer/T0364.try14-opt2.unpack.gromacs0.repack-nonPC.pdb.gz // other versions of try14 have 0 breaks where as this has 7 minor breaks (i.e. <0.6) dimer/T0364.try13-opt2.unpack.gromacs0.repack-nonPC.pdb.gz // has 2 breaks that are scored >1 My Favorites (as far as overall shape and score): dimer/T0364.try4-opt2.pdb.gz // has only minor breaks and is based off of the try5 monomer (which I prefer). Similar to try13, but scores better in undertaker I am not sure what else to do from here. I suppose that the next step is to retrieve the monomers from the dimers so that we can submit them. I will search through the README for some info on how to do that. Tue Jul 25 17:17:04 PDT 2006 Crissan Harris So I figured out how to add the monomers from the dimers into the best-models. I'm just wondering now if I need to somehow move copies of these into the main decoys directory. If so then I am not sure how to do that. I will look into this some more tomorrow when more help is available. But other than that, the best-models.pdb.gz is up to date. One other thing that I am concerned about is that a few of our models are very similar. We might want to look into generating a few more models. Wed Jul 26 11:08:35 PDT 2006 Crissan Harris I looked at some of the models that were further down than what was shown in the best-models top 5, and found that model 9 (1vpmA) had an interesting placement of the c-terminal helix. So I decided to do a run that reads in our top model(1s5uA) and this model(1vpmA) only in the hopes that undertaker will get a better idea of where to place the helix. The costfcn was copied from try1.costfcn. try7 is currently running on vashon. Wed Jul 26 13:41:57 PDT 2006 Kevin Karplus try7 failed, because 1s5uA was not in MANUAL_TOP_HITS and make read_alignments had not been run, so the read-alignments-scwrl file was not available. I'm not sure the try1 costfcn is the best choice once you've started picking templates, as it is trying to average out constraints over several possible templates. I'll start the try7 run anyway, so that we can see what it does. Meanwhile, Crissan might want to think about a more specific cost function. Wed Jul 26 13:46:09 PDT 2006 Kevin Karplus try7 started on cheep. I'm not sure that a monomer-only optimization is as useful as the dimer optimizations at this point. I agree that we seem to have converged on the models worth looking at. Wed Jul 26 19:28:11 PDT 2006 Kevin Karplus try7 does not score very well with the try7 costfcn, but try7-opt2.gromacs0.repack-nonPC is rosetta's second favorite, after try5-opt2.gromacs0.repack-nonPC. All the models are so close, whether optimized as a monomer or half a dimer, that I see little point to further polishing. (Exception: the c-terminal helix in try7-opt2 is in a different, possibly better, location) Wed Jul 26 19:41:51 PDT 2006 Kevin Karplus Of the 8 models currently being considered: InFilePrefix dimer/decoys/ ReadConformPDB T0364.try12-opt2.unpack.pdb chain A //Undertaker's #1 ReadConformPDB T0364.try14-opt2.unpack.gromacs0.repack-nonPC.pdb chain A //Rosetta's #1 ReadConformPDB T0364.try4-opt2.unpack.pdb chain A // Crissan's favorite ReadConformPDB T0364.try7-opt2.unpack.pdb chain A //Undertaker's #2 ReadConformPDB T0364.try13-opt2.unpack.gromacs0.repack-nonPC.pdb chain A //Rosetta's #2 InFilePrefix decoys/ ReadConformPDB T0364.try5-opt2.pdb //monomer from before dimerization ReadConformPDB T0364.try6-opt2.pdb ReadConformPDB T0364.try7-opt2.pdb //monomer from alignments (1s5uA) Models 1(dimer/try12) and 4 (dimer/try7) are too similar. Models 3(dimer/try4) and 5 (dimer/try13) are too similar. Model 6(try5) is also rather similar to model 3(dimer/try4) So I'll submit InFilePrefix dimer/decoys/ ReadConformPDB T0364.try12-opt2.unpack.pdb chain A //Undertaker's #1 ReadConformPDB T0364.try14-opt2.unpack.gromacs.repack-nonPC.pdb chain A //Rosetta's # ReadConformPDB T0364.try4-opt2.unpack.pdb chain A // Crissan's favorite InFilePrefix decoys/ ReadConformPDB T0364.try6-opt2.pdb // < try4-opt2 < alignments (1s5uA) ReadConformPDB T0364.try7-opt2.pdb //monomer from alignments (1s5uA) Wed Jul 26 19:58:05 PDT 2006 Kevin Karplus I just noticed that the unconstrained costfcn was missing the KnownBreak command in dimer/ so I'm rescoring with the unconstrained costfcn. OK, dimer/try12-opt2 still scores best. I have submitted with comment Aside from a couple of C-terminal helices we had excellent hits to 1s5uA. We did some optimization in a monomeric context, then dimerized by superimposingon 1s5u[AB] and did further optimization in the dimeric context. Very little movement of the backbone occured during the minimization. The models we are submitting are Model 1 is dimer/T0364.try12-opt2.unpack chain A, which is the dimeric model that scored best with the unconstrained costfcn. dimer/try12-opt2 < dimer/try7-opt2 < dimer/try5-opt2 < monomer try6-opt2 Model 2 is dimer/T0364.try14-opt2.unpack.gromacs0.repack-nonPC chain A which is the dimeric model that Rosetta liked best of all the ones that it repacked sidechains for. dimer/try14-opt2 < dimer/try10-opt2.unpack.gromacs0.repack-nonPC < dimer/try5-opt2 < monomer try6-opt2 Model 3 is dimer/T0364.try4-opt2.unpack chain A, which is Crissan's favorite of the dimeric models. It was optimized from the monomeric try5-opt2. Model 4 is T0364.try6-opt2, the monomer used for building a couple of the dimers above. try6-opt2 < try4-opt2 < alignments (mainly 1s5uA) Model 5 is T0364.try7-opt2, a monomer that has not been polished much, but which was chosen for its somewhat different placement of the C-terminal helices. try7-opt2 < alignments (mainly 1s5uA)