Fri Jun 30 09:42:11 PDT 2006 T0355 Make started Fri Jun 30 09:44:20 PDT 2006 Running on lopez.cse.ucsc.edu Fri Jun 30 09:50:09 PDT 2006 Kevin Karplus OOPS, accidentally stepped on the README file by rerunning new-target. Here is what was said before: Thu Jun 29 09:52:38 PDT 2006 T0355 Make started Thu Jun 29 09:54:19 PDT 2006 Running on shaw.cse.ucsc.edu Thu Jun 29 10:02:35 PDT 2006 Kevin Karplus BLAST finds no good hits in PDB, best is 1qddA or 1lit (46% for 26 residues, E-value 0.6) 288 residues. Thu Jun 29 10:25:22 PDT 2006 Kevin Karplus This seems to be a tryptophan synthase subunit alpha. (There are subunit beta structures c.79.1.1, but I don't know whether subunit alpha is likely to be the same.) Fri Jun 30 12:09:21 PDT 2006 Kevin Karplus Strong consensus among the HMMs for 1rd5A 5.e-23, 1qopA 8.e-23, 1kfcA 5.e-22 TIM barrels c.1.2.4. Unfortunately, the top alignments do *not* agree with the secondary structure prediction, so this model will need work---either pulling out a non-TIM-barrel fold or finding an alignment to the TIM barrle that works. Mon Jul 17 11:34:40 PDT 2006 Grant Thiltgen I looked at the model, and the strands in the TIM barrel didn't match the predicted structure. I'm assuming that the neural nets would be good at predicting these, since they are a common fold. I looked to see what was being used for the alignment, and it appears that undertaker was using 1tjpA. This was in our list of top alignments, but way down the list, to an evalue of e-04, where the top hits with e-23 were not being used. I am going to start a run that selects from the top few alignments instead of all of them. Mon Jul 17 11:48:17 PDT 2006 Grant Thiltgen I chose the top five fold recognition hits to make a model after and included the sheet constraints from the top two models. This is try2 and it is started on whidbey. Mon Jul 17 15:22:03 PDT 2006 Grant Thiltgen I'm not sure if try2 is any better than try1. I am going to look at the t2k alignments, because the conserved residues from the multiple alignment are closer to the conserved residues for the protein. The t2k fold-recognition comes up with c.1.2.1 as the top SCOP family as opposed to c.1.2.4 that the t06 alignments came up with. I am going to try to run off the t2k alignments instead of the t06 alignments to see if this will match something better. I made try3 with the top 3 alignments from the t2k alignments. I now need to make the sheet constraints for the TIM barrel in this family. try3 started on whidbey. Tue Jul 18 13:35:54 PDT 2006 Grant Thiltgen I am going to work on polishing the three models we have. Of the three models, so far try3 is best, then try2, then try1. Try4 is a polish of try1. Try5 is a polish of try2. Try6 is a polish of try3. All of these are started on the cluster. Tue Jul 18 14:01:51 PDT 2006 Grant Thiltgen Rosetta likes try3-opt2.gromacs0.repack-nonPC.pdb the best, try7 is going to be a polish of that model. Try7 is also started on the cluster. Wed Jul 19 11:27:40 PDT 2006 Grant Thiltgen Of the polished models, try6 looks the best, followed by try7, then try5, and try4. I need to work a little on try4 in order to get the strand in place in the TIM barrel, since the last strand is a bit off. I am also going to polish the tries again. Rosetta likes the try7-opt2.gromacs0.repack-nonPC model the best, so I will probably polish that one up as well. I have the sheet constraints in for the first undertaker model This is try8. Try9 is a polish of try5. Try10 is a polish of try6 and try11 is a polish of try7-opt2.gromacs0.repack-nonPC. Try8 and try9 started on whidbey. try10 started on camano. try11 started on orcas. Wed Jul 19 19:35:50 PDT 2006 (grant?) Currently, the best models are T0355.try10-opt2.pdb T0355.try11-opt2.gromacs0.repack-nonPC.pdb T0355.try9-opt2.pdb T0355.try8-opt2.pdb top alignment for 1rd5A T0355.try10-opt2.pdb is ultimately a polish model of try3. Try3 was built on the alignments of the top three fold recognition hits for the t2k alignments. I chose this because the sequence logo for the t2k alignments showed much better conservation of specific residues in the target. The top three hits for fold recognition were 1vh7A, 1h5yA, and 1thfD. T0355.try11-opt2.gromacs0.repack-nonPC.pdb is the best model of the rosetta cost function. I originally polished the try3-opt2.gromacs0.non-repackPC.pdb model, since that was the best rosetta model. It is very similary to try10, but the polish was done on the gromacs0.repack-nonPC models instead of the original undertaker output. T0355.try9-opt2.pdb is ultimately a polish of try2. Try2 was based on the top five consoldated alignments: 1rd5A, 1qopA, 1kfcA, 1geqA, and 1ujpA. T0355.try8-opt2.pdb is a polish of the original undertaker try1. There were some strands in the barrel misplaced, and try8 was an attempt to bring the strands back into the barrel. The alignment model is the top model in the alignments file. Wed Jul 19 21:43:15 PDT 2006 Kevin Karplus I submitted, but I moved try11-opt2.gromacs0.repack-nonPC ahead of try10-opt2, because it did a much better job of closing the gaps. try10-opt2 did not score particularly well in either undertaker or rosetta evaluations. ReadConformPDB T0355.try11-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0355.try10-opt2.pdb ReadConformPDB T0355.try9-opt2.pdb ReadConformPDB T0355.try8-opt2.pdb InFilePrefix ReadConformPDB T0355.undertaker-align.pdb model 1 1rd5A Thu Jul 20 22:29:37 PDT 2006 George Shackelford Kevin and I discussed differences between the t06.str2 and t2k.str2 predictions, and, after looking at t06.w0.5 and t2k.w0.5 sequences, we concluded that the predictions by t2k were better. With that in mind, I was checking over try11 this evening by using the t2k.near-backbone-11 and t2k.dssp-ehl2 rasmol scripts to see how well the latest version is doing. The ehl2 predictions line up very well in terms of the barrel though there are some differences away from the barrel. However the near-backbone predictions suggest that some adjustments are needed. I am also bothered by the lack of the usual "flap" structure that covers one end of the barrel. There is enough sequence to do this, but where it comes in is not clear. I have shifted the two strands, 102-106 and 162-166, up by 2 (i.e. 102-106 -> 104-108) because I feel these may be better matches based on t2k.near-backbone-11. The results are in try12.costfcn. In try12.under, I have restored the TryAllAligns for the specific templates and T0355.t2k.undertaker-align.under. I can hope that we get a bit of an option that has the breaks fixed. try12 running on peep Fri Jul 21 11:07:39 PDT 2006 George Shackelford try13 running on peep Try12 still had ReadConformPDB of the try7 in place so all it did was optimize that. I ran a try13 that had the ReadConformPDB commented out and the results were actually worse. At least try12 scores the best with Rosetta. I'm going back to the roots by using the try3.under along with my new sheet constraints to see if there is a better alignment. Try14 running on peep Fri Jul 21 15:45:33 PDT 2006 try14 forced the changes I was looking for but strand 104-108 broke badly. I have rerun the rr.constraints using the t2k data and I am getting predictions that are weak but agree strongly with what we have. The only changes I still see is to flip 104-108 over by moving it up 1. The other change is to include our new t2k.rr.constraints, and put in a constraint to see if we can bring up the loop that needs to be in the slot over I23 (I12-I23). I may need to flip I23 over as well. but my constraint will be loose enough. try15 running on vashon I did so well with my unintented polishing with try12, I've decided to repeat the whole thing but starting with try12. Such will be try16. try16 running on camano Fri Jul 21 17:14:39 PDT 2006 George Shackelford try15 has gotten its opt1. It is interesting to see what kind breaks it has. There is nothing that cannot be fixed as far as I can see. try16 is still running. I may find I need one unconstrained polishing run. The superimpose has been updated to include try12, and best-models.pdb is updated as well. Fri Jul 21 18:41:09 PDT 2006 George Shackelford Try16 has finished. It does not score as well as try11 or try12 in best-rosetta. We can include it but otherwise I think we've pushed it about as far as we can go. Try15 has finished and does not appear to have done well. It may have corrected some problems but it scores poorly by best-rosetta. I think I'll have to lay the effort to rest. Damn. My window manager froze and I lost notes. The final order I have selected is: try11 try12 try16 try9 server model 1 Try12 and try16 are based on try11 which comes from try7 which comes from try3 which uses the top three best alignments from t2k. Try10 comes from try6 which apparently also comes from try3 (I'm not sure what the difference is). Try12 and try16 included the top FIVE t2k models. try9 is a polishing of try2 which is the best models of t2k, t04, and t06. superimpose and best-models are updated. Fri Jul 21 20:55:16 PDT 2006 Kevin Karplus George has given me an ordering of 5 models, but no explanation for why he chose the ones he did. I checked some of the cost functions, to see how they ordered things: unconstrained: try10, try6, try11, try7, try12, try16 rosetta: try11, try12, try7, try16, try10, try6, try3 try10: try11, try16, try12, try7, try6, try3 try11: try11, try16, try12, try7, try6, try3 I created a new cost function, like unconstrained, but with strand and helix constraints from tryk.str2 and t2k.dssp-ehl2 t2k: try10, try6, try11, try7, try12, try16 Based on just this data (and no information about how the models actually differ), I would choose try10-opt2 try11-opt2.gromacs0.repack-nonPC try6-opt2 try7-opt2.gromacs0.repack-nonPC try12-opt2.gromacs0.repack-nonPC but if two of these were too similar, I might pull in try1-opt2 or align 1 (to 1rd5A). Perhaps I need to also look at try9 and try16, which George inclued in his list. Fri Jul 21 21:04:39 PDT 2006 George Shackelford There were basically only two lines of tries: one (try2) based on the templates from consensus best e-values, and one (try3) based on the three best templates from t2k best e-values. Only try9 is from the best consensus templates, try10, try11, try7 come from try3. try12 and try16 come from try11 with the next two highest templates included; I have included them because they did well on the two scorings of unconstrained and rosetta. They already did well on their own scoring which is based on constraints from t2k. Given a little time on my hands, I decided to start two polishing runs: try17 is polishing try15 (which appears to have two corrections to the strands) and try18 which is polishing try16. These polishings are unconstrained with the clashes and breaks turned up. If you want to see something completely different, you can look at try15. Fri Jul 21 21:29:08 PDT 2006 Kevin Karplus George stepped on the README file, ignoring the lock I had on the file. (IF he used emacs, he would have been warned of the lock.) I managed to recover my work and add his comments in. try10 and try11 differ mainly in a small movement of the loop D71-F87, but also in try11 having fewer and smaller breaks. Try10 compensates with better hbonds and sidechains, which may not be as good. try6 is almost identical to try10, so I'll drop it. try7 is almost identical to try11, so I'll drop it. try12 is almost identical to try11, so I'll drop it. try16 is slightly different try9 is wildly different align1 seems to be the basis for try9. So I'll submit ReadConformPDB T0355.try11-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0355.try10-opt2.pdb ReadConformPDB T0355.try16-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0355.try9-opt2.pdb InFilePrefix ReadConformPDB T0355.undertaker-align.pdb model 1 # 1rd5A Maybe I'd better look at try15 first, though George hasn't said anything about it except that it is different. It scores terribly, and does not have any features that jump out at me as recommending it, so I don't think I'll include it unless someone points out an interesting feature that is worth keeping. Fri Jul 21 21:38:32 PDT 2006 George Shackelford Ok, I'm using emacs now... Try15 was an effort to get certain sheet constraints to take place. Specifically it moves the strand at 161-167 to be 163-169, and flip the strand at 103-107 over by shifting it up 1 to 104-108 and changing the hbond. I also put a constraint in to try and move the loop up into the slot over I23 (the constraint was between I13 and I23). However this did not happen. The results would need a lot of polishing to close gaps. I thnk there is a way to force word wrap in emacs but after spending over an hour the other day looking for it I gave up. Fri Jul 21 21:46:27 PDT 2006 Kevin Karplus The command is auto-fill-mode. I turn it on in my .emacs file for text mode buffers with (setq text-mode-hook '(lambda() (turn-on-auto-fill))) I did the submission for T0355 at 21:40, but couldn't add to the README file at that time (because emacs quite properly locked the file while you were editing it). Sat Jul 22 10:00:23 PDT 2006 George Shackelford Try18 is an improved polish of try16. I suggest replacing try16 with try18. Sat Jul 22 16:48:54 PDT 2006 Kevin Karplus Sorry, I did not get this message until almost 5 pm, and the deadline was noon. If you wanted it changed that close to the deadline, you would have had to do it yourself.