Thu Jul 13 09:17:57 PDT 2006 T0376 Make started Thu Jul 13 09:20:13 PDT 2006 Running on shaw.cse.ucsc.edu Thu Jul 13 15:33:21 PDT 2006 Kevin Karplus BLAST gets strong hit on 1o5kA (26% over 225 residues, e-value 1.4e-14) HMMS get strong hits on c.1.10.1 TIM barrels. Sun Jul 30 15:12:38 PDT 2006 George Shackelford T0376 - TIM barrel? I hope so. I'm going to adjust the rr.constraints, double check the barrel and do another run. However this is looking pretty good already though there are some breaks to heal. In score-all+servers.unconstrained, we score at the top just ahead of ROBETTA. As for how good we are really, well... I need to check conserved and near predictions to see how stuff is lining up. The barrel actually looks good. The near and ehl2 agree almost perfectly. There is one small strand prediction that shows up as a helix. The rr.constraints are almost all in the barrel and in agreement. for this strongly predicted barrel I'll be taking them out for future tries; they could only confuse things. The conserved_t06 is odd; the conserved seemed to be in odd places, e.g. L153, Q59, D51. I'm not sure what that means. The next run will be a repeat of try1 with rr.constraints removed and including t06.str2.constraints Sun Jul 30 23:50:15 PDT 2006 George Shackelford I found that try1 <- 1hl2A. What will we get this time? Same one? try2 running on vashon Tue Aug 1 14:57:03 PDT 2006 George Shackelford Really need to catch up on this one, even if it is lookin' good. The results on try2 are in, and it scores almost the same as try1. It gets the first negative score value from Rosetta. Both don't do well in breaks and soft clashes. try1 has breaks>1.0 about I72, E275, V269, A115, Y266. Try2 has breaks>1.0 at V186, P73, E275, K271 A115. I could push for an early polishing and see what I can get. Wed Aug 2 10:19:55 PDT 2006 George Shackelford Time flies and so I'm doing a polishing on what we have by using read-pdb.under. Soft C set to 40, breaks to 150. try3 running on peep Wed Aug 2 12:21:42 PDT 2006 Grant Thiltgen It appears that most of the top fold recognition hits are in the same vicinity as the top pdb hits, but they do get out of order. For try4 I'm going to try to run the top pdb hits instead of using all the alignments to see how that works out. Try4 is ready. I used the constrants for helix and strand from the str2 alphabets and added some sheet constraints and some hbond constraints (who knows it might give us an edge) Try4 started on lopez. Wed Aug 2 15:06:55 PDT 2006 Grant Thiltgen Try4 crashed because I typoed one of the pdb files. I'm restarting try4, and sending it to the cluster. Thu Aug 3 00:30:41 PDT 2006 George Shackelford Well, try4 is different, but scoring unconstrained it's actually worse than the best of the other three. Interestingly it scores best using Rosetta, however Rosetta isn't necessarily trying to match an actual structure. Try4 is definitely worth including. (Its phobic fit is not good but then it is rather spread out in some of the structure.) try3 still has a few breaks. I am going to try and polish it some more. try5 running on lopez Thu Aug 3 12:27:46 PDT 2006 Grant Thiltgen Try5 now scores best on unconstrained and with rosetta. We may have a good model here. Try4 scores worst with the unconstrained costfcn, but second best for rosetta, so it could be worth including. Polishing it up may help I am also going to work on a few new models from alignments. I'll look at what try4 chose for alignments and comment those out. I have updated the superimpose-best.under and best-models.pdb.gz file to correspond with our current best models. Try3 seems to be polished off of try1. So currently I'd go with try5, try2, and try4, with a few alignments for our best models. try6 (polish of try4) started on whidbey. Try4 seemed to use 1xkyA mostly for alignments, so I'm commenting that out and starting over with alignments for try7. Try7 started on camano. Thu Aug 3 13:37:40 PDT 2006 George Shackelford I'm rather satisfied with try5. It polishes try3 well; almost no breaks or soft clashes. The alpha predictions suffer as well as sidechains and bystroff. It's enough to cost try5 the lead in unconstrained, but it scores tops in Rosetta. I think I'm going to quit while I'm ahead. I'll take some of this afternoon to wrap things up (if Grant doesn't beat me to it). Thu Aug 3 15:48:48 PDT 2006 Grant Thiltgen I just rescored everything with unconstrained, and try3 actually beats try5 in unconstrained, but try5 beats it in rosetta. It might be a better model to go with, but I might try one more polish on it. Try6 (a polish of try4) scores better than try4 is both unconstrained and in rosetta, and is the current top score with rosetta. Try7 which is a new model based on alignments scores better than try4 did. So currently we have try5, try2, try6, and try7, and maybe one alignment for submissions. I am running one more polish of try5 which is try8. Try9 is a polish of try6, Try10 is a polish of try2, Try11 is a polish of try7. Try8 and try9 started on whidbey. Try10 and try11 started on shaw. Thu Aug 3 17:18:09 PDT 2006 Grant Thiltgen Try3 scores better than try5 on the unconstrained costfcn. I scored the servers with the unconstrained as well, as the SAM_T06 server model 1 scores the best of the servers, then Robetta, which socres worse than most of our models. I am going to start try12 and try13. Try12 is a polish of the best SAM model, and try13 is a polish of the Robetta models TS1 and TS5 which score virtually the same. Try12 started on abyss Try13 started on lopez. Thu Aug 3 22:19:18 PDT 2006 Grant Thiltgen Okay, everything is done running, now let me take a look at everything for submissions. Try13 and try12 score best, which are polished off of the server models, try13 was polished off of the robetta models, and try12 off of the top SAM model. Next is try3. Try5 which is a polish of try3 scores better on rosetta, but worse on the undertaker costfcn. It was started from gromacs, so it's hard to say. The best scoring model by rosetta is try8. Try8 was a polish of try5. It could be worth putting in, but it is very similar to try3 and try5. Try10 is the next highest scoring model in undertaker. It is a polish of try2, which George said was close to the original model. OF the models I built from different alignments, try11 which is a polish of try7 scores best. Including a different model from the original tries I think I should include this one, because it is much more protein-like than the models from try4. So this gives me try13, try12, try3, try8-gromacs0.non-PC, and try7. Should I include the server models before our new model? Probably not. I'm thinking I should go with this order: try3-opt2 try13-opt2 try12-opt2 try7-opt2 try8-opt2.gromacs0.repack-nonPC. Try9 does score second best with rosetta, but since it's based off of try4, I think I am going to forget about that one too. Model 1 is try3-opt2 which is a polished run off of the original automatic undertaker run for this target. It seemed to use mostly 1hl2A for alignments. Model 2 is try13-opt2 which is a polished run off of the top two ROBETTA server models using the undertaker unconstrained costfcn. We polished the model a little bit, and it was the best scoring model we had based on server models. Model 3 is try12-opt2 which is a polished run off of the top SAM_T06 server model. This was the second best scoring model we had based on server models. Model 4 is try7-opt2. For this model we attempted to use alignments from the closest pdb files found by BLAST instead of the models found by fold recognition. This was done in order to give a bit of variety to submitted models. This model seemed to use mostly 2a6nA for alignments. Model 5 is try8-opt2.gromacs0.repack-nonPC. This model is actually a further polished model from try3, but scores worse than try3 in the undertaker costfcn. This model is the best scoring model using the rosetta scoring function, so we are including it here in our model. Submissions mailed at 10:40 pm on August 3rd, 2006.