Thu Jul 6 09:10:11 PDT 2006 T0360 Make started Thu Jul 6 09:11:31 PDT 2006 Running on lopez.cse.ucsc.edu Fri Jul 7 23:54:51 PDT 2006 Kevin Karplus No good hits in PDB with BLAST (best is 2glfA, E-value 0.3) Strong hit to 1dvoA with HMMs. (E-value 3.6e-14) Fri Jul 14 13:01:29 PDT 2006 Crissan Harris, Cynthia Hsu We were examining try1-opt2 in Rasmol and noted that the ehl2 script suggested that the residues from L44-L46 could potentially form a sheet with residues G85-F88. We also hoped that this might bury the L44-L46 residues, which are colored brown in RasMol when viewed with the burial script. Because of this, we copied the sheet constraints from try1-opt2. sheets, and changed the relevant sheet to 2. Other than that, try2 was identical to try1. try2 is currently running on camano. Fri Jul 14 15:59:34 PDT 2006 Crissan Harris, Cynthia Hsu try2 did not form the third sheet, as desired. We decided to comment out the dssp-ehl2 constraints, the undertaker alignments, and the rr constraints. We also decided that we generally liked the position of the helices from try1 more than those from try2, except for the long C terminal helices that are protruding into the environment. Thus, we copied the terminal helix constraints from try2-opt2.helix and all other helices from try1-opt2.helix, as follows: //from try1-opt2.helices HelixConstraint (T0360)M1 (T0360)M18 1 HelixConstraint (T0360)T17 (T0360)I31 1 HelixConstraint (T0360)I48 (T0360)L56 1 HelixConstraint (T0360)D60 (T0360)R66 1 HelixConstraint (T0360)V67 (T0360)R73 1 HelixConstraint (T0360)R74 (T0360)L81 1 //from try2-opt2.helices HelixConstraint (T0360)T99 (T0360)H109 1 HelixConstraint (T0360)Q107 (T0360)Q114 1 HelixConstraint (T0360)V112 (T0360)E141 1 try3 is currently running on lopez. Sun Jul 16 17:35:00 PDT 2006 Cynthia Hsu I just observed that we had the order of the residues in the last sheet constraint reversed (P43-A45, instead of A45-P43). I made the appropriate change. I also set "maybe_ssbond" to 0, given that there is only one Cysteine residue. However, there are three His residues, all of them clustered near the beta sheets, away from the N and C termini of the helices. The conserved_t06 script also indicates this to be the location of the active site, and the rr script agrees that these should be brought closer together. I've raised "maybe_metal" to 1.5 in view of this. Hopefully, this will also rotate the helices so that the parts that should be buried will be buried. try4 is currently running on camano. Mon Jul 17 09:16:47 PDT 2006 Cynthia Hsu P43-A45 still refused to form a sheet, but other than that, I am really pleased with the way try4 turned out. It's produced an interesting shape with a kink in the long helix, but it appears as expected under both the near and the burial scripts. It still appears rather foamy, though, so I'll try raising the dry weights as follows: dry5 20 dry6.5 25 dry8 20 Other than that, I'm not sure what else to change. We'll see how it goes later on. try5 is currently running on camano. Mon Jul 17 11:32:24 PDT 2006 Crissan Harris, Cynthia Hsu try5 did not seem that much improved. However it did score better in the unconstrained scores and the try5 constrained scores. The superimpose-best.under was updated with our current four best distinct models. Mon Jul 17 15:51:49 PDT 2006 Cynthia Hsu Given that we only have four distinct models, I thought it might be in our best interest to pursue previous efforts of forcing residues P43-G47 to adopt some conformation. Although they appear to be attempting to form a parallel beta sheet, there is no indication of this from any of the prediction logos. It may make more sense for it to form an antiparallel sheet, although SAM_t04 seems to suggest that it could be a helix. try6.costfcn was copied from try5.costfcn. To adjust for the orienation of the P43-P45 sheet, we changed the hbond to start from A45. This SheetConstraint was also raised to 4. try6 is currently running on camano. For try7.costfcn, try5.costfcn was again copied. We added the additiona Helix Constraint as follows: //implied from SAM_T04 str2.logo HelixConstraint (T0360)P43 (T0360)G47 1.5 Because of this, the strand constraints in the region were also removed. try7 is currently running on vashon. Mon Jul 17 18:03:24 PDT 2006 Cynthia Hsu In order of unconstrained scores, the preferred models by Undertaker are try5, try4, try7, and try6. However, Rosetta favored try7, followed by try1 and try6. I decided that it might be worth it to polish try1. For try8.costfcn, I set "dry5" to 25, "dry6.5" to 30, and "dry8" to 20. "soft_clashes" was raised to 60 and "break" was raised to 80. "bad_peptide" was raised to 12 and "n_ca_c" to 7. I also commented out the initial constraints. The Undertaker script read the PB results from the gromacs repacked version of the try1 model. try8 is currently running on vashon. To polish try7 (gromacs repacked), I raised "dry6.5" to 30, "dry8" to 23, "n_ca_c" to 7, "soft_clashes" to 60 and "break" to 120. try9 is currently running on peep. I also thought that try6 (gromacs repacked) should be polished, due to its relatively low score but unique conformation. "dry5" was set to 25, "dry6.5" to 30, "dry8" to 23, "n_ca_c" to 7, "soft_clashes" to 80, and "break" to 120. try10 is currently running on abyss. Tue Jul 18 10:32:09 PDT 2006 Crissan Harris, Cynthia Hsu All of the previous attempts to polish (try8, try9, try10) did worse than the models they were supposed to polish in the unconstrained scores. For try11, we copied the costfcn for try9. We raised "dry5" to 25 and "dry6.5" to 35. "n_ca_c" was raised to 10 and "bad_peptide" was raised to 12. try11.under used try9-opt2. try11 is currently running on camano. Looking at the scores for the individual costs, we found that try10 actually did better than try6 in most, with the exception of "side_chain" and "bad_peptide. In try12, we polished try10 by riasing the weight of "bad_peptide" to 15. try12.under used try10-opt2 as its input file. try12 is currently running on vashon. We also found that try8 did worse in try1 in the area of "bad_peptide" but little else; in view of this, we decided to polish try8 but raise the cost of "bad_peptide to 13. try13.under used try8-opt2 as its input file. try13 is currently running on squawk. Tue Jul 18 12:06:08 PDT 2006 Crissan Harris, Cynthia Hsu try11, try12, and try13 all scored very well with Undertaker, as hoped. We decided to now concentrate our efforts on polishing try5, which could still benefit from better "n_ca_c", "soft_clashes", and "breaks" scores. "dry5" was raised to 25, and "dry6.5" to 30. "n_ca_c" was raised to 7. "soft_clashes" were raised to 60, and "break" weight was raised to 100. try14.under used try5-opt2 as its input PDB file. try14 is currently running on squawk. Make started Tue Jul 18 13:10:22 PDT 2006 Running on whidbey.cse.ucsc.edu Tue Jul 18 13:12:35 PDT 2006 Crissan Harris, Cythia Hsu We ran make2 as Prof. Karplus suggested in his email. Tue Jul 18 14:02:33 PDT 2006 Crissan Harris, Cynthia Hsu try14 successfully polished try5, so it was added to the superimpose-best.under file. We updated superimpose-best.under accordingly: ReadConformPDB T0360.try14-opt2.pdb //polishing run of try5 #ReadConformPDB T0360.try4-opt2.pdb //similar to try5 ReadConformPDB T0360.try11-opt2.gromacs0.repack-nonPC.pdb.gz //Rosetta's favorite; polishing run of tries 9 and 7 - Residues 43-47 form a helix. ReadConformPDB T0360.try12-opt.pdb //polishing run of tries 6 and 10 ReadConformPDB T0360.try13-opt2.gz // polishing run of tries 1 and 8. ReadConformPDB T0360.try3-opt2.pdb //low scoring, but gives diverse model. We also decided to polish try3. try15.costfcn dry weights were updated as follows:dry5 20 dry6.5 25 dry8 18 We also changed "soft-clashes" to 60, "break" to 120, and "n_c_ac" was raised to 7. Thu Jul 20 13:21:44 PDT 2006 Crissan Harris, Cynthia Hsu superimpose-best.under was updated to include the new results. Thu Jul 20 14:11:59 PDT 2006 Kevin Karplus I looked at the new score-all+servers.unconstrained.pretty and noticed that SAM_T06_server_TS1 is scoring as well as some of the top models. Has anyone looked at it? The next highest-scoring server model is ROBETTA_TS5. Thu Jul 20 15:52:14 PDT 2006 Cynthia Hsu In contrast to our models, the server models generally had a kink in the helix at around T127. Usually a kink also exists in K16. With this in mind, I examined the socre-all+servers.unconstrained.pretty, and noted that the SAM_T06_server_TS1 performed worse than models 14, 5, 11, and 12, with a particularly good "sidechain" and "bad_peptide" score. Because it was low-scoring and matched the SAM_T06_server_TS1 fairly well, I decided to modify model 15 so that it more closely resembled the server model. For try16.costfcn, which was copied from try15.costfcn, I added an rr constraint between I27 and I64, as I felt that this would bury certain regions that were exposed in this model. I changed the helix constraint so that there would be a kink between M18 and Q23, so that it would more closely resemble the server model. Other than this, I was not sure what to change. try16 is currently running on vashon. Fri Jul 21 10:28:07 PDT 2006 Crissan Harris, Cynthia Hsu try16 failed to form a kink in the appropriate location. In fact, try16 scored significantly worse than all other models, try15 included. We decided to try again, using only the top scoring SAM_T06_server_TS1. We added the helix constraints, as follows: HelixConstraint (T0360)M1 (T0360)V15 1 HelixConstraint (T0360)V112 (T0360)L128 1 We also modified try17.under to take only the SAM_T06_server_TS1 model for its alignments. try17 is currently running on lopez. Fri Jul 21 14:57:41 PDT 2006 Crissan Harris, Cynthia Hsu There were problems with the try17 file. Try18 is now set up as a polishing run of the SAM_T06_server_TS1 model. For try18.costfcn, we realized that it was probably better to modify the constraints to fit SAM_T06_server_TS1's weaknesses. We removed the constraints at the end, then set "dry5" to 25, "dry6.5" to 30, "n_ca_c" to 10, "soft_clashes" to 80, and "break" to 150. try18 is currently running on vashon. Fri Jul 21 16:36:19 PDT 2006 Crissan Harris, Cynthia Hsu Our top five models are as follows: ReadConformPDB T0360.try14-opt2.pdb Top scoring model. This was a polishing run of try5, which was based on a very similar costfunction to try4. ReadConformPDB T0360.try18-opt2.pdb //polished SAM_T06_server_TS1.pdb model. We took the server model and polished it with the try15.costfcn, but removed the constraints at the end. It ended up scoring the second highest. ReadConformPDB T0360.try11-opt2.gromacs0.repack-nonPC.pdb Rosetta's third favorite, and the third favorite of Undertaker's (excluding try5, which was identical to try14). This was a polishing run of tries 9 and 7, in which residues 43-47 form a small helix. ReadConformPDB T0360.try12-opt2.pdb This was a polishing run of tries 6 and 10. It had a higher sheet constraint than previous models on residues 43, to 45. This was not as high scoring as some other models, but gave a diversity of folds. ReadConformPDB T0360.try16-opt2.gromacs0.repack-nonPC.pdb This model was low scoring with Undertaker, but Rosetta's favorite and it gives diverse model. Fri Jul 21 17:54:52 PDT 2006 Kevin Karplus submitted with comment: For T0360, we had a good fold-recognition hit to 1dvoA, and to no other PDB templates. We did have short alignments to several other templates, which did get used in creating the models. Model 1 is try14-opt2, our top-scoring model. This was a polishing run of try5-opt2, which was created from alignments. Model 2 is try18-opt2.pdb, polished from our SAM_T06_server_TS1 model. We took the server model and polished it with the try15.costfcn, but removed the constraints at the end. It ended up scoring the second highest. Model 3 is try11-opt2.gromacs0.repack-nonPC, which was try11-opt2 reoptimized by gromacs and with sidechains repacked by rosetta. try11-opt2 is polished from try9-opt2, polished form try7-opt2.gromacs0.repack-nonPC, from alignments. This is Rosetta's third favorite, and the third favorite of Undertaker's (excluding try5, which was identical to try14). It is distinguised from other models in that residues 43-47 form a small helix (which appeared in try7). Model 4 is try12-opt2, polished from try10-opt2, from try6-opt2.gromacs0.repack-nonPC, from alignments. The try12 run had a higher sheet constraint than previous models making an antiparallel connection between R87-D89 and A45-P43. This was not as high scoring as some other models, but gave a diversity of folds. Model 5 is try16-opt2.gromacs0.repack-nonPC. This model was low scoring with Undertaker, but it was Rosetta's favorite of the models it repacked sidechains for and it gives a somewhat different model. ---------------------------------------------------------------------- Thu Jul 27 12:26:46 PDT 2006 Crissan Harris, Cynthia Hsu We re-examined the previous models, and decided to try one more polishing run for try16. We copied the costfcn to try19, set "dry5" to 25, "dry6.5" to 30, "dry8" to 20, "phobic_fit" to 3, "n_ca_c" to 10, "bad_peptide" to 15, and turned off "maybe_ssbond". We set try18.under to ReadConformPDB T0360.try16-opt2.gromacs0.repack-nonPC.pdb try19 is currently running on camano. We also decided to make one last attempt to polish the fourth-choice model, try12-opt2. We copied try12.costfcn to try20.costfcn. "dry5" was set to 30, "dry6.5" to 37, "dry8" to 28, and "dry12" to 6. We also raised "phobic_fit" to 4, and "bad_peptide" to 17. try20 used try12-opt2 as the input file. try20 is currently running on lopez. Thu Jul 27 14:30:33 PDT 2006 Crissan Harris, Cynthia Hsu try20 is now currently our third-highest scoring model. However, try19 did substantially worse than the model it was supposed to polish. Rosetta also still favors try16 the most. In view of this, try16 will still be submitted. We also considered that it might be possible to generate an alignment just from the top fold recognition hit. We copied try1.costfcn to try21.costfcn, then turned off "maybe_ssbonds". try21.under took in 1dvoA as its alignment. try21 is currently running on shaw. Thu Jul 27 16:09:50 PDT 2006 Crissan Harris, Cynthia Hsu We examined the model generated from 1dvoA, and were extremely displeased with it. It was very foamy, many of the residues that should have been buried were not, and in particular, the C terminal helix was kinked so that its very end, which should have been exposed, was tucked against the beta sheets. In view of this, we have decided to make the following models our submissions: Model 1 is try14-opt2, our top-scoring model. This was a polishing run of try5-opt2, which was created from alignments. Model 2 is try18-opt2.pdb, polished from our SAM_T06_server_TS1 model. We took the server model and polished it with the try15.costfcn, but removed the constraints at the end. It ended up scoring the second highest. Model 3 is try11-opt2, polished from try9-opt2, polished form try7-opt2.gromacs0.repack-nonPC, from alignments. It is distinguised from other models in that residues 43-47 form a small helix (which appeared in try7). Model 4 is try20-opt2, polished from try12-opt2, from try10-opt2, from try6-opt2.gromacs0.repack-nonPC, from alignments. The try12 run had a higher sheet constraint than previous models making an antiparallel connection between R87-D89 and A45-P43. The high score of this model is accounted for mainly by the numerous polishing runs, but it gives a slight diversity of folds, as its C terminal helix is kinked slightly to give it less of an elongated shape. Model 5 is try16-opt2.gromacs0.repack-nonPC. This model was low scoring with Undertaker, but it was Rosetta's favorite of the models it repacked sidechains for and it gives a somewhat different model. Thu Jul 27 16:18:43 PDT 2006 Crissan Harris, Cynthia Hsu Reviewing our best models, I found that we never really utilized the rr constraints to manipulate the helices. Because of this, I decided to take our worst scoring model (try19) and try to improve on it by adding rr constraints to improve its packing. The rr constraints present are as follows: Constraint I27.CB I64.CB -10. 7.0 14.0 2 Constraint I27.CB F41.CB -10. 7.0 14.0 2 Constraint Y77.CB L81.CB -10. 7.0 14.0 2 Constraint Y77.CB V98.CB -10. 7.0 14.0 2 try22.under is not a polishing run. try22 is currently running on shaw. Thu Jul 27 17:17:27 PDT 2006 Crissan Harris, Cynthia Hsu try22 does better than try19 in terms of Undertaker scores, but unfortunately ROsetta is still favoring try16. We've decided to try polishing try22.costfcn in hopes that it will improve the scores to Rosetta's liking. For try23.costfcn, we copied that of try22, raised "dry5" to 30, "dry6.5" to 37, "dry8" to 25, "and n_ca_c" to 12. try23.under was a polishing run of try22-opt2.pdb. try23 is currently running on lopez. Our other decision was to polish try16 with the try22 cost function. We raised "dry5" to 30, "dry6.5" to 35, and "dry8" to 25, however, to improve the packing slightly. Like try19.under, try24.under was a polishing run of T0360.try16-opt2.gromacs0.repack-nonPC.pdb. try24 is currently running on shaw. Thu Jul 27 18:28:34 PDT 2006 Cynthia Hsu Our top four models are still the same as they were listed in the 16:09 entry from today. Model 5 is as follows: Model 5 is try24-opt2.gromacs0.repack-nonPC. This model was low-scoring with Undertaker, but the top favored by Rosetta, and it shows some diversity among our models. It was a polishing run of try16-opt2, but had added RR residues which helped to bury some regions that were exposed. It was generated primarily by manipulating the weights of various properties in the cost function, in particular the addition of RR constraints. superimpose-best.under and best-models.pdb.gz have been updated accordingly. Fri Jul 28 00:20:05 PDT 2006 Grant Thiltgen Since most of the models were the same as before, with only a few polishing runs that changed models 4 and 5, I kept these models in the order that Cynthia and Crissan suggested. Since there was only one parent in the MANUAL_TOP_HITS target in the makefile, and there was already a soft submission for this target, I went ahead and left it as the only parent. I submitted the five models at around 12:20 am on Friday, July 28.