Tue Jul 11 09:13:06 PDT 2006 T0370 Make started Tue Jul 11 09:14:49 PDT 2006 Running on shaw.cse.ucsc.edu Tue Jul 11 10:00:53 PDT 2006 Kevin Karplus BLAST gets no good hits, but the HMMs see b.45.1.1 as a good possibility. Sun Jul 16 16:53:33 PDT 2006 Cynthia Hsu My initial impression of this model is that it matches up fairly well with the alignment models. However, the match causes numerous very very large breaks, so I suppose a goal of future tries will be to attempt to minimize these breaks, possibly by reducing the length of and constraints on the sheets and the helices. These breaks were located before A37, L46, T117, E149, and W150. In view of this, for the try2.costfcn, I commented out the undertaker-align.sheets (as I felt that they were the primary reasons for the breaks) and the dssp-ehl2 constraints (as many of the alignments did not actually agree with the dssp-ehl2 constraints). The rr constraints were left in the cost function, as I didn't want any major changes to the structure at this point. I then copied the sheet and helical constraints from try1, and set all of the helical constraints with a weight of 1. The helical constraints did not involve any residues that were in the vicinity of breaks, but going through the sheet constraints, I made the following changes: SheetConstraint (T0370)V24 (T0370)A31 (T0370)S44 (T0370)A37 hbond (T0370)A26 1 //Used to span from V24-G32, but forcing G32 into a sheet seemed to cause a break in the hairpin loop from G32-G35, and also at L46. I removed the following sheet constraint, due to the fact that it formed an incomplete beta strand that also forced the break around the L46 area: SheetConstraint (T0370)Y45 (T0370)D48 (T0370)V54 (T0370)T51 hbond (T0370)D48 1 In the following constraint, I shortened the length of the sheet by one residue on either side from the original, given that T51 did not actually form a beta sheet during alignment, nor did A58 or S119. SheetConstraint (T0370)F52 (T0370)P57 (T0370)I125 (T0370)T120 hbond (T0370)V53 1 I removed the following SheetConstraint from the C-terminus, as the dssp-ehl2 constraint insisted that it was a helix, not a SheetConstraint, and forcing it to form a sheet as the alignment models did was causing the break around 149 and 150: SheetConstraint (T0370)E143 (T0370)D147 (T0370)D154 (T0370)W150 hbond (T0370)L144 1 In view of the absence of cysteine residues, I set "maybe_ssbond" to 0. There were also no His residues, so "maybe_metal" was also set to 0. I raised "break" up to 80. I'm hoping most of the problems with breaks will be taken care of by the modifications to the constraints as described above, but we'll have to wait and see. try2 is currently running on shaw. Mon Jul 17 09:25:20 PDT 2006 Cynthia Hsu try2 has formed a rather interesting flattened barrel; I'm not sure what to make of it. There is, however, a break between W47 and T51 that needs to be addressed (all other breaks were closed, thankfully). With this in mind, I've decided to just concentrate on closing the break and then seeing what strange fold I might get out of it. One of the sheet constraints was modified to the following: SheetConstraint (T0370)L53 (T0370)P57 (T0370)R124 (T0370)T120 hbond (T0370)V54 1 This prevented F52 from forming a hydrogen bond with the adjacent beta sheet, which will hopefully give the residues in that region a little bit more freedom of movement. The weight on "break" was raised slightly, up to 100. try3 is currently running on orcas. Tue Jul 18 11:20:31 PDT 2006 Cynthia Hsu The break before F52 is still present. Looking at the sheets, they appear undulating, which makes me wonder if I've set the H-bonding inappropriately. For this reason, I've decided that for try4, I am going to modify try2.costfn incstead, and completely remove the sheet spanning residues T51 and P57. "soft_clashes" was raised to 40, "break" to 120. The overall weight of "constraints" was raised from 10 to 12, and the following sheet constraint was removed: SheetConstraint (T0370)F52 (T0370)P57 (T0370)I125 (T0370)T120 hbond (T0370)L53 1 try4 is currently running on peep. Tue Jul 18 13:13:55 PDT 2006 Cynthia Hsu There is also a break before K81. The break before F52 still exists. To try to close the break in F52: SheetConstraint (T0370)A26 (T0370)T30 (T0370)I76 (T0370)V72 hbond (T0370)V28 1 SheetConstraint (T0370)A26 (T0370)A31 (T0370)P42 (T0370)A37 hbond (T0370)A26 1 I hoped that shifting the Hbond over by one residue would prevent it from contorting. I also hoped by shortening the length of the sheet (from 37-44 to 37-42), it would give the strand a little bit more freedom of movement. try5 is running in lopez. Tue Jul 18 14:52:13 PDT 2006 Cynthia Hsu The break before T51 still exists. I'm not sure what else to do at this point. I've decided that for the try6.costfcn, I'd raise the "break" weight to 200 and lower the overall weight of the "constraint" to 10. try6 is currently running on squawk. Out of desperation, I've also copied over try1.costfcn to try7.costfcn, raised "break" to 200 and "constraint" to 80. I have a feeling that I'm making this protein worse, not better, so if neither of these two tries work, I'll have to ask for help. try7 is currently running on lopez. Tue Jul 18 16:39:03 PDT 2006 Cynthia Hsu On try7, raising the breaks up to 200 did have the desired effect of closing the breaks, but commenting out dssp-ehl2 contraints, undertaker alignments, and rr constraints may have been a bad idea. It scored the worst on unconstrained out of all models; I didn't feel like it was worth modifying this despite the fact that the breaks had been closed up. For try8, I lowered the "break" weight to 160 and the "soft_clashes" weight to 60. I uncommented the dssp-ehl2 constraints, undertaker alignments, and rr constraints. It also occurred to me to lower "sidechain" weight from 5 to 2. I also realized that I hadn't removed the "maybe_ssbond" and "maybe_metal" constraints, both of which I set to 0. try8 is currently running on lopez. Tue Jul 18 17:48:24 PDT 2006 Cynthia Hsu In general, try3 scored the best unconstrained, followed by try1 and try2. try3-opt1 actually had a relatively low "break" weight of 0.4. try6 had the same weight on "break" as try3-opt2 did, but did worse than the other three models in terms of scoring. All models, with the exception of try1, have a very high "n_ca_c" score, which supports my hypothesis that my attempts to close the breaks only resulted in contorting the beta sheets. In try3, the most badly twisted beta sheet appears to have been the one spanning N25 to A33. In view of this, I copied the try3.costfcn to try9.costfcn, and removed all sheet constraints pertaining to the relevant residues: SheetConstraint (T0370)V24 (T0370)A31 (T0370)S44 (T0370)A37 hbond (T0370)A26 1 SheetConstraint (T0370)A26 (T0370)T30 (T0370)I76 (T0370)V72 hbond (T0370)W27 1 I also raised "n_ca_c" to 10. I also lowered "constraint" to 8 and "sidechain" to 3. try9 is currently running on shaw. I decided to do a similar modification of the try2.costfcn. Given that the beta sheet from A136 to N147 was looking a little bit twisted, the sheet constraint that referred to it was deleted: SheetConstraint (T0370)R128 (T0370)E135 (T0370)D147 (T0370)S140 hbond (T0370)V130 1 As before, "n_ca_c" was raised to 10 and "constraint" was lowered to 8. "break" was raised to 150 and "soft_clashes" to 60. I also decreased "sidechain" to 4. try10 is currently running on camano. Thu Jul 20 12:14:57 PDT 2006 Cynthia Hsu Examining try9 in the near script seems to indicate that S44-N48 exhibits incorrect H bonding to V54-F52. This was the same for A37-Y45. Because of this, I added the following sheet constraints to try11 (which was a copy of try9): SheetConstraint (T0370)A26 (T0370)G32 (T0370)G36 (T0370)P42 hbond (T0370)P38 1 SheetConstraint (T0370)S44 (T0370)W47 (T0370)V54 (T0370)T51 hbond (T0370)Y45 1 Hopefully, these constraints will encourage the formation of a hairpin turn. I also raised "soft_clashes" to 40. I raised the weight of "constraints" back up to 12. try11 is currently running on abyss. For try10, I noticed the same pattern with the H-bonding, resulting in a break after S44. For this reason, I made the following sheet constraint: SheetConstraint (T0370)W27 (T0370)A31 (T0370)L40 (T0370)G36 hbond (T0370)V28 2 try12 is currently running on camano. Thu Jul 20 14:24:09 PDT 2006 Cynthia Hsu try12 had about two more breaks than try10 did. When examining the unconstrained scores, I realized, and verified with the all.breaks.gz file, that try5 actually had a very acceptable level of breaks, and try3 was not that bad, either. I think I realize now that my endless efforts to close the breaks were futile, and probably made the protein worse. With this in mind, I decided to focus on refining models try3 and try5. For try13.costfcn, I copied try5.costfcn. I decided to increase "phobic_fit" to 4, hoping this will bring the underside of the beta sheet closer together, to conceal the exposed residues. "soft_clashes" was raised to 60. "constraints" was lowered to 8, and "n_ca_c" raised to 8, in hopes tha this would fix any problems with the hydrogen bonding. For this region, I also changed this sheet constraint: SheetConstraint (T0370)A26 (T0370)A31 (T0370)P42 (T0370)A37 hbond (T0370)W27 1 //originally, Hbond at A26. try13 currently running on squawk. For try14.costfcn, which was copied from try3.costfcn, I removed the following sheet constraint, hoping that this will prevent the H bonds from becoming distorted: SheetConstraint (T0370)A26 (T0370)T30 (T0370)I76 (T0370)V72 hbond (T0370)W27 1 I raised "soft_clashes" to 60 and "break" to 120 as well, to improve these scores. I also lowered "constraints" to 8 and raised "n_ca_c" to 10. try14 is currently running on shaw. Thu Jul 20 16:34:09 PDT 2006 Cynthia Hsu try14 scored worse than try3, (and only slightly better than try2) mainly in the areas of "break", "phobic_fit", and dry weights. In view of this, I lowered "n_ca_c" to 8, and "sidechain" to 2. I also raised "dry6.5" to 25 and "dry8" to 20. try15 is currently running on orcas. try13 appeared very badly, with the break between 44 and 46 increasing, and another one appearing between 117 and 119 that exposed a region of the barrel. In general, however, it seems to have scored better than try5, although it performed poorly in "phobic_fit" and "break". I lowered "phobic_fit" to 1, and "sidechain" to 4. try16 is currently running on lopez. Fri Jul 21 11:21:29 PDT 2006 Cynthia Hsu Both try15 and try16 did significantly worse on unconstrained scores than try14 did. try3 scored the best overall, especially in "phobic_fit" but scored very poorly on "n_ca_c". try14 was an improvement to this, but still scored poorly on breaks. The best scoring model, in terms of "break", was try5, but it generally scored badly in "n_ca_c", which I am concerned about given the appearance of the beta sheets. In view of this, I decided to focus my efforts primarily on making the try14 model work, refraining from worrying about breaks until later. I copied try14.costfcn into the try17.costfcn. Many of the rr constraints seemed to be hindering the burial of certain residues, so I commented these out. I added the following, however, in an effor to close the region: Constraint W27.CB P38.CB -10. 7.0 14.0 1 Constraint W133.CB R129.CB -10. 7.0 14.0 1 I also added the following distance constraints, in an effort to close the breaks: constraint((T0370)Y45.CA, (T0370)L46.N) [> -1 = 0.01 < 1] w=3 constraint((T0370)R115.CA, (T0370)L116.N) [> -1 = 0.01 < 1] w=2 constraint((T0370)A37.CA, (T0370)P38.N) [> -1 = 0.01 < 1] w=2 I also raised "break" to 150" and "soft_clashes" to 70, and then lowered "sidechain down to 4. try17 is currently running on peep. Mon Jul 24 13:16:48 PDT 2006 Cynthia Hsu try17 failed to perform as hoped on breaks. Mon Jul 24 15:01:36 PDT 2006 Sylvia Do, Cynthia Hsu So, we will be attempting to align the H-bonds. According to my notes, we should try to get 46-48, 51-53. (Hbonds between L46-L53, D48-T51) Taking the costfcn of try3, we added SheetConstraint (T0370)L46 (T0370)D48 (T0370)L53 (T0370)T51 hbond (T0370)L53 1 try18 is running on vashon Mon Jul 24 21:17:19 PDT 2006 Sylvia Do Gah! Apparently, the .under was try15's, and not try3's, and would explain why diff worked when I diffed try18's and try3. Bleh. I redid the .under and will rerun try18. Try18 is running on Lopez Tue Jul 25 12:34:55 PDT 2006 Cynthia Hsu The breaks in try18 are still there. It appears that the residues still refused to line up as specified by the SheetConstraint. I raised the weight on the L46-D48, L53-T51 sheet to 10, hoping that this would force the appropriate formation of the sheet. I also raised "soft_clashes" to 40 and "break" to 120. try19 is currently running on lopez. Tue Jul 25 14:55:49 PDT 2006 Sylvia Do, Cynthia Hsu try19, despite the high weight on the added sheet constraint, still failed to produce the desired result. We examined the model and found that the sheet preceding L46-D48 also seemed off by one, so we copied try18.costfcn, replace the try1-opt2.sheets with try18-opt2.sheets, then added the following constraints replacing the faulty ones from try18-opt2.sheets: SheetConstraint (T0370)L46 (T0370)D48 (T0370)L53 (T0370)T51 hbond (T0370)L53 5 SheetConstraint (T0370)D25 (T0370)A33 (T0370)S44 (T0370)G36 hbond (T0370)S44 3 try20 is currently running on camano. Tue Jul 25 15:45:14 PDT 2006 Sylvia Do, Cynthia Hsu try20 successfully closed all breaks. In view of this, we decided to try using the try20.costfcn for all future models, as this was the only one without significant breaks. We noted that the N-terminal helix (up until D23) had swung out into the environment, away from the beta barrel as it had originally been placed. This was contrary to the burial script, which indicated that part of the helix could be used to cover hydrophobic regions of the barrel. In view of this, we decided to try to bring the helix closer to the hydrophobic regions of the barrel. we're using the following contraints: Constraint I13.CA R73.CA 0.0 4.0 8.0 1 Constraint T56.CA G100.CA 0.0 6.0 8.0 1 We also raised "n_ca_c" to 7 and "bad_peptide" to 15, hoping that this will prevent the residues from being twisted out of shape as they are brought closer together. try21 is currently running on peep. Wed Jul 26 10:17:22 PDT 2006 Cynthia Hsu Under the constraints we put in to try21, try20 still performs the best. When viewed in RasMol, it seemed that our RR constraints had little to no effect on moving the helices. I've decided to resort to copying the try20.costfcn to try22.costfcn. This meant abandoning previous attempts to raise the weights on n_ca_c and bad_peptide. I added these constraints: Constraint I13A.CA R73A.CA 0.0 4.0 8.0 5 Constraint T56A.CA V101A.CA 0.0 6.0 8.0 5 The first one is identical to the one used in try21, for the purpose of moving the N-terminal helix closer to the barrel. The second, for moving P94-G110, changed slightly, as in try21, it had labeled an exposed residue. Hopefully, by raising the weight to five, it will be more likely to perform as desired. try22 is currently running on orcas. Wed Jul 26 13:41:56 PDT 2006 Sylvia Do The problem with that helix that we were trying to cover the hydrophobic area is that it is not swinging the way that we want it to. Right now, I'll just try to optimize try20 a little. It is scoring well, but its score for phobic fit is high compared to those around it. In try23, I'll will use try20's costfcn and increase phobic_fit to 6 and dry12 to 6. the dry12 score is in about the same range, so I am not increasing it by much. try23 is running on shaw. I also chose some different residues to get closer, to try to get the helix going the right way. From try22, I changed to: Constraint I13A.CB R71A.CB 0.0 4.0 8.0 5 Constraint A55A.CB V101A.CB 0.0 6.0 8.0 5 try24 is running on peep. Wed Jul 26 16:02:34 PDT 2006 Sylvia Do, Cynthia Hsu Neither try23 nor try24 performed satisfactorily. Both had very large breaks and failed to cover up the hydrophobic regions as desired. In view of this, we took try20-opt2, used ProteinShop to bring the helices closer to the exposed regions of the barrel, which was saved as try20-shifted2.pdb. For try25, we copied try1.costfcn, then commented out the dssp-ehl2, undertaker-align, and rr constraints. try25.under was set up as a polishing run that used ProteinShop/try20-shifted2 as its input model. try25 is currently running on peep. Wed Jul 26 17:05:53 PDT 2006 Sylvia Do, Cynthia Hsu While examining the try25-opt1 model, we noticed that there were multiple patches of exposed residues, more than the covered by the two helices we had attempted to ProteinShop. In view of this, we decided to attempt to generate a new model, in which the C terminus residues starting from E138-D154 could be forced to form a helix, which might give us another sidechain to work with. Because of this, we copied try20.costfcn to try26.costfcn, then removed the following sheet constraint: SheetConstraint (T0370)L144 (T0370)R146 (T0370)L151 (T0370)E149 hbond (T0370)R146 1 We then added an additional HelixConstraint: HelixConstraint (T0370)E138 (T0370)D154 2 We also lowered the weight on "sidechain" to 4. try26 is currently running on abyss. Thu Jul 27 11:22:00 PDT 2006 Cynthia Hsu In try25, the helices were still a bit far from the desired regions. In view of this, I lowered the "sidechain" weight to 3, removed "maybe_metal" and "maybe_ssbond", and added the following distance constraints: Constraint V54A.CA F105A.CA 5. 7 11.0 4 Constraint G102A.CA F123A.CA 4. 6 11.5 4 I also raised "dry6.5" to 25, "dry8" to 20, and "dry12" to 7. try27 was set up as a polishing run of try25-opt2. try27 is currently running on camano. try26 finished, with a very significant break, kinked helices, all without having the desired helical constraint. Thu Jul 27 13:33:17 PDT 2006 Sylvia Do, Cynthia Hsu We examined try26, and we decided that it would benefit from having a simpler cost function. For try28.under, we decided to not make it a polishing run. try28.costfcn was a copy of try25.costfcn. However, "sidechain" was lowered to 4, and we copied the helix and sheet constraints from try26. We also remembered to remove "maybe_metal" and "maybe_ssbond" this time. try28 is currently running on lopez. Thu Jul 27 15:09:58 PDT 2006 Cynthia Hsu try27 has a decent shape, but it is still extremely foamy. try28 formed the helices as desired. They are also placed in the appropriate location to cover the exposed residues, although they are still extremely foamy, and there were several significant breaks. I've decided, however, that in try28, it can still be improved by bringing some of the residues closer together. To do this, I added the following distance constraints: Constraint V54A.CA V101A.CA 2. 6.0 11.0 1 Constraint Y45A.CA F105A.CA 2. 6.0 11.0 1 I also modified the helical constraints, as follows: HelixConstraint (T0370)G96 (T0370)T109 1 HelixConstraint (T0370)N137 (T0370)D154 4 //This is the helix we added "soft_clashes" was raised to 40 and "break" was raised to 100 as well. try29 is currently running on orcas. Thu Jul 27 15:56:02 PDT 2006 Sylvia Do, Cynthia Hsu try27, despite its foaminess is currently our top scoring model. Rosetta, however, prefers try17, try10, and try14 (all gromacs repacked) as its top three. We're not sure which one is better, so we've decided to, in the meanwhile, polish try27. try30.costfcn was a copy of try27.costfcn. We raised "dry6" to 35, "dry8" to 25, and "dry12" to 8. We also raised "n_ca_c" to 10, "bad_peptide" to 13, "soft_clashes" to 60, and "break" to 150. "constraints" was also raised to 12. For try30.under, we decided that instead of specifying try27-opt2, we would ask it to polish all existing models. try30 is running on lopez. Thu Jul 27 17:06:30 PDT 2006 Sylvia Do, Cynthia Hsu try29 did very badly. Fri Jul 28 10:45:33 PDT 2006 Cynthia Hsu try30 is currently our top scoring model. There are still several regions of high cost, however, so in view of this, I've decided to one more polishing run using "include pdb.under" in the try31.under file. "dry5" was raised to 20 and "dry6.5" 37, although "dry12" was lowered to 7. "n_ca_c" was raised to 12 and "bad_peptide" to 15. try31 is currently running on shaw. I've looked through all of our top scoring models, and updated the superimpose-best.under file with the following comments: ReadConformPDB T0370.try30-opt2.pdb #ReadConformPDB T0370.try23-opt2.pdb //Very similar to try30, but some of the sheets are badly skewed. #ReadConformPDB T0370.try25-opt2.pdb //first optimization of ProteinShopped model; similar to try30 ReadConformPDB T0370.try26-opt2.pdb ReadConformPDB T0370.try3-opt2.pdb // except for helix, identical to try28. ReadConformPDB T0370.try28-opt2.pdb ReadConformPDB T0370.try1-opt2.pdb I would still like to make an effort to improve on try28. I copied try28.costfcn to try32.costfcn, then raised the constraint on the desired helix to 4. I also lowered "sidechain" to 2 and raised "dry12" to 6. I also replaced the sheet constraints from try18 with those from try20. try32 started on peep, Fri Jul 28 11:39:48 PDT 2006 Fri Jul 28 13:10:03 PDT 2006 Cynthia Hsu try32 has helices placed in a good orientation, but the desired one failed to form. There were also two very large breaks that formed. try33.costfcn was a copy of try32.costfcn. I found that the last two sheet constraints had been applicable to the E138-D154 helix, and were probably a contributing factor to its inability to form a helix. I commented those two out, and raised the weight on the helix constraint to 8. I then raised "break" to 100, "soft_clashes" to 60, and lowered the overall "constraint" weight to 8. try33 is currently running on shaw. Fri Jul 28 13:34:38 PDT 2006 Sylvia Do I'm going to go back to try25 and try to add the helices. I've taken try25's .under file and changed the ReadConformPDB to try25-opt2. I then took try28's costfcn and changed some of the constraints. I will keep the helices from try25, rather than from try1's in hopes that this run will not change try25's structure except for the helix that was wanted. Changes: HelixConstraint (T0370)P4 (T0370)L22 1 //changed this HelixConstraint (T0370)S60 (T0370)T69 1 HelixConstraint (T0370)G100 (T0370)G110 1 //changed this HelixConstraint (T0370)A136 (T0370)D154 3 //This is the helix we added Increased the weight on the helix we wanted. try34 is running on abyss. Fri Jul 28 14:36:23 PDT 2006 Sylvia Do, Cynthia Hsu try33 achieved the desired result, except for one very large break before L46. try34 is not done yet, but according to the opt1 model, it does not seem to be forming the desired helix. We copied the try33.costfcn to try35.costfcn, then raised "soft_clashes" to 80 and "break" to 200. This was not a polishing run. try35 is currently running on shaw. We have also decided to rerun the try34.costfcn, but this time refraining from making a polishing run. We copied try34.costfcn to try36.costfcn, then used a non-polishing run for try36.under. try36 is currently running on shaw. Fri Jul 28 14:52:58 PDT 2006 Sylvia Do, Cynthia Hsu We decided that it may also be useful to make a chimera out of 30 and 33, as this would take our helix and put it on a barrel that does not have any breaks. We used the make-chimera.under file to generate try30-try33helix.pdb. We then removed the residues in between E135 of the two models (chimera30-33). Fri Jul 28 15:13:15 PDT 2006 Sylvia Do, Cynthia Hsu The chimera30-33.pdb generated a model in which both the helices were on the same side of the barrel, instead of on opposites sides, as we had intended. I decided to use the try30-try33helix file to use the N-termainl helix in try33 as well (spanning from M1-A26). This was inappropriate named chimera30-33-30 (it should have been named chimera33-30-33). For try37.costfcn, I copied try1.costfcn, but commented out the dssp-ehl2 constraints, the undertaker alignments, and rr constraints. try37.under used ReadConfomrPDB chimera30-33-30.pdb. try37 is currently running on lopez. We also considered using chimera30-33 and ProteinShopping the helices into the appropriate locations, but ProteinShop does not seem to want to work at the moment. try36 did badly. Fri Jul 28 16:25:18 PDT 2006 Sylvia Do, Cynthia Hsu We used ProteinShop to move the helices in chimera30-33 to their desired places (ProteinShop/chimera30-33-shifted.pdb). For try38.costfcn, we copied try1.costfcn, commented out the dssp-ehl2 constraints, the undertaker alignments, and the rr constraints. We also removed "maybe_metal" and "maybe_ssbond". try38 is currently running on peep. Sat Jul 29 13:52:07 PDT 2006 Sylvia Do try38 did not keep the helix we wanted (there is, however, a small helix in the area). I'm going to take try38's .under and costfcn and add some helix constraints to keep that helix. The helix constraints will be the one I had for try34's costfcn. try39 is running on peep. I wonder if the helix we have is a little too long and is clashing with the barrel. I'm going to set the helix constraints so that at worst, there will be two smaller helices in the area rather than strands. try40 will be the same as try39, but with these helixconstraints: HelixConstraint (T0370)P4 (T0370)L22 1 HelixConstraint (T0370)S60 (T0370)T69 1 HelixConstraint (T0370)G100 (T0370)G110 1 HelixConstraint (T0370)A137 (T0370)R146 3//changed HelixConstraint (T0370)D147 (T0370)D154 3//added try40 is running on orcas. Sat Jul 29 14:46:38 PDT 2006 Sylvia Do To save us a little time tomorrow, just in case. try30-opt2 < try27-opt2 < try25-opt2 < try20-shifted2 (try20-opt2 < 1dn1A, 2asfA,2fhqA, 1g79A, 1xhnA(last one) ) try26-opt2 < 1dn1A, 2asfA, 1xhnA (last one) try3-opt2 < 1dn1A, 2asfA, 2fhqA (last one) try28-opt2 < 1dn1A, 2asfA, 2fhqA (last one) try1-opt2 < 2aq6A, 1dn1A, 2asfA, 1xhnA (last one) In other news, try31 and 34(opt2) are doing well. chimera30-33-30 and chimera30-33 are high in undertaker's scoring as well. try31-opt2 < try30-opt2 < see above try34-opt2 < try25-opt2 < see above try39 just finished. Looks good, except the N-terminus helix strayed from where we set it. try39-terminusshift.pdb is 39, with the N-terminus moved back where it was wanted - to cover the exposed regions. Try41 will be on try39-terminusshift, with try39's costfcn. Let's hope the helix stays put. try41 is running on peep. Sun Jul 30 12:37:07 PDT 2006 Sylvia Do, Cynthia Hsu try40 had the helices in the appropriate location of where we desired them, but they had strayed slighlty from the intended location. As a consequence, the protein is foamy and with many exposed residues. The same problems exist in try41, but appear slightly less dire. Rosetta really likes try17, and I wonder if it is because the N-terminus is convering the other part of the barrel. For try42, I took try39-shifted.pdb (ProteinShopped), which is like try39-terminusshift, but with the two end helices swapping places (that is, covering the exposed area that the other was in try39-terminusshift). I've just taken try41's .under and costfcn and applied it to try42. try42 is running on peep. Sun Jul 30 13:30:12 PDT 2006 Sylvia Do, Cynthia Hsu After examining our top scoring models, we concluded that these were our five preferred models: ReadConformPDB T0370.try31-opt2.pdb//top-scoring from try20-shifted; still foamy and one region of the barrel exposed ReadConformPDB T0370.try34-opt2.pdb //preferred over try31-opt2; still needs to be polished. ReadConformPDB T0370.try41-opt2.pdb//from try39-shifted; foamy, exposed; but has all four helices ReadConformPDB T0370.try39-opt2.pdb //helices are not in desired location #ReadConformPDB T0370.try40-opt2.pdb.//from try39, but with added helix constraints. - similar to try39, but C-terminal helix does not cover exposed barrel #ReadConformPDB T0370.try26-opt2.pdb #ReadConformPDB T0370.try3-opt2.pdb // except for helix, identical to try28. #ReadConformPDB T0370.try28-opt2.pdb The order is as listed, except for try34 and try31 being reversed, as try34 has not yet been polished as is appropriate. try43.costfcn was a copy of try34.costfcn. "dry5" was raised to 20, "dry6.5" was raised to 25, "dry8" was raised to 20, "phobic_fit" to 3, "n_ca_c" to 10, "bad_peptide" to 12, "soft_clashes" to 40, and "break" to 100. The .under file used ReadConformPDB T0730.try34-opt2.pdb try43 is currently running on camano. try44.costfcn was a copy of try41.costfcn. "dry5" was raised to 20, "dry6.5" to 25, "dry8" to 20, "dry12" to 7, "n_ca_c" to 7, "bad_peptide" to 12, "soft_clashes" to 40, "backbone_clashes" to 3, "break" to 100, and "constraints" to 12. try44.under to ReadConformPDB try41-opt2.pdb. try44 is currently running on lopez. try45.costfcn was a copy of try39.costfcn. "dry6.5" was raised to 25, '"dry8" was raised to 17, "soft_clashes" was raised to 40, "break" was raised to 100, and "constraints" was raised to 12. try45 is currently running on shaw. Sun Jul 30 15:08:14 PDT 2006 Sylvia Do, Cynthia Hsu try42 did very badly, creating many breaks and placing the helices in a completely different direction from what was desired. It also scored very poorly with Rosetta. In view of this, we decided to take try17-opt2 and attempt to polish it to our standards. We replaced the sheet constraints in the try17.costfcn with those from try31-opt2, then changed the distance constraints to the appropriate format and lowered the weights on each of them to 1. We also raised "dry5" to 20, "dry6.5" to 25, and raised "constraint" to 10. "soft_clashes" was also raised to 80 and "break" to 160. try46.under used ReadConformPDB T0370.try17-opt2.gromacs0.repack-nonPC.pdb try46 is running on peep. Sun Jul 30 15:57:42 PDT 2006 Sylvia Do, Cynthia Hsu Our selections for the top five models are as follows: T0370.try43-opt2.pdb Although this was only our second-highest scoring unique model, we preferred it because its N-terminal helix was not kinked, whch seemed to improve the accuracy of its burial. It was a polishing run of the following: try43-opt2 < try34-opt2 < try25-opt2 < try20-shifted2 (try20-opt2 < 1dn1A, 2asfA,2fhqA, 1g79A, 1xhnA(last one) ) try25-opt2 was a polishing run of try20-shifted2, a ProteinShopped model that moved the two helices so that they would be closer to the exposed residues of the beta barrel. The starting model for this was try20-opt2, our first model produced in which the hydrogen sheets in the barrel had the correct phase, sealing the breaks that had been produced during the initial alignment by Undertaker. T0370.try31-opt2.pdb This was our best scoring model according to Undertaker. However, it is less preferred than try43-opt2, because a kink in the N-terminus helix at A9 causes the N-terminus residues to be inappropriate buried. try31-opt2 was a polishing run of the following: try31-opt2 < try30-opt2 < try27-opt2 < try25-opt2 < try20-shifted2 (try20-opt < 1dn1A, 2asfA,2fhqA, 1g79A, 1xhnA(last one) ) try25-opt2 was a polishing run of try20-shifted2, a ProteinShopped model that moved the two helices so that they would be closer to the exposed residues of the beta barrel. The starting model for this was try20-opt2, our first model produced in which the hydrogen sheets in the barrel had the correct phase, sealing the breaks that had been produced during the initial alignment by Undertaker. T0370.try44-opt2.pdb try44 was selected because it had four helices, as predicted by the ehl2 script, and these were placed to that they, for the most part, buried the regions that they should have. However, this still did not solve the problem of exposed residues, and the protein still remains rather foamy. try44 was a polishing run of try41, which polished the ProteinShopped model try39-terminusshift.pdb. try39-terminusshift.pdb was based on a polishing run of a chimera, chimera30-33shifted.pdb. This chimera was inappropriate named, but it took the C terminal helix and its orientations from try33-opt2, and the barrel from try30. Unlike previous models, try33, and therefore try44, had a C-terminal helix, as predicted by the ehl2 scripts. ProteinShop was used to manipulate the helices so that they would cover the exposed regions of the barrel. T0370.try45-opt2.pdb.gz try45 had four helices, as predicted by the ehl2 script, although they failed to sufficiently cover the exposed regions of the beta barrels, resulting in a foamy protein that does not score as well as the other models. It was a polishing run of try39-opt2, which was an Undertaker optimization of the chimera30-33-shifted. Unlike previous models, try33, and therefore try44, had a C-terminal helix, as predicted by the ehl2 scripts. Weights were adjusted in the cost function to produce this model. decoys/T0370.try46-opt2.gromacs0.repack-nonPC.pdb.gz This gromacs repacked version of try46-opt2 is the most-favored model according to Rosetta. It is significantly different from the other models, as the N-terminal helix is oriented parallel to the beta barrel, not perpendicular as it is in the other models. It was a polishing run of try17-opt2.gromacs0.repack-nonPC.pdb, which was the most favored by Rosetta prior to this model. try17 was originally produced by adding helical, rr, sheet, and distance constraints, and adjusting the weights on the costfunction. However, this produced many breaks, as the beta sheets in the hydrogen barrel had the wrong phase. To adjust for this, in try46, we polished try17, but replaced the original sheet constraints with those of try31-opt2, our best-scoring model according to Undertaker. In reviewing our best-models, we realized that one of our top scoring models, chimera30-33-30 was never used. Its Undertaker optimization, try37-opt2, was fairly low-scoring. Because of this, we decided to do one last polishing run on try37 and see if it scores comparably to try44 and try45. try47.costfcn was a copy of try37.costfcn. "dry6.5" was raised to 25, "n_ca_c" was raised to 7, "soft_clashes" to 40, and "break" to 80. try47.under used try37-opt2 as its input file. try47 is currently running on lopez. Sun Jul 30 16:40:38 PDT 2006 Cynthis Hsu, Sylvia Do All in all, there are the top models: try43-opt2 < try34-opt2 < try25-opt2 < try20-shifted2 (try20-opt2 < 1dn1A, 2asfA,2fhqA, 1g79A, 1xhnA(last one) ) try31-opt2 < try30-opt2 < try27-opt2 < try25-opt2 < try20-shifted2 (try20-opt2 <1dn1A, 2asfA,2fhqA, 1g79A, 1xhnA(last one) ) try44-opt2 < try41-opt2 < try39-terminusshift (try39-opt2 < chimera30-33-shifted) try45-opt2 < try39-opt2.pdb < chimera30-33-shifted try46-opt2.gromacs0.repack-nonPC < try17-opt2.gromacs0.repack-nonPC (try17-opt2 < 1dn1A, 1xhnA (last one)) (Alternate choice for 4th model) chimera30-33-30 (Note for the chimeras: try30-opt2 < try27-opt2 < try25-opt2 < try20-shifted2 (try20-opt2 < 1dn1A, 2asfA,2fhqA, 1g79A, 1xhnA(last one) ) and try33-opt2 < 1dn1A, 1xxoA, 1xhnA (last) ) Sun Jul 30 17:17:14 PDT 2006 Sylvia Do, Cynthia Hsu After examining try47-opt1, we decided that Undertaker could not properly optimize the chimera as desired. As in try37, the two helices were both moved to the exterior of the barrel, which did not seem to be an accurate representation of what was supposed to happen. In view of this, we decided that we would refrain from Undertaker optimizations of the chimera. However, we would still like to suggest chimera30-33-30 as an alternate fourth choice that is different from try45-opt2, as it is one of our highest scoring models. In addition, the orientation of the helices is as desired, covering the exposed regions both on the interior and the exterior of the barrel. The chimera was produced using the N and C terminal helices (residues 1-23 and from 135 on) from try33, and using the barrel and other helices from try30. Sun Jul 30 18:02:57 PDT 2006 Kevin Karplus I'm not that fond of try46-opt2, as the N terminal helix has its hydrophobic face pointing out away from the sheet. chimera3-33-30 has some of the same problem, and similar problems with the its C-terminal helix. Based on burial patterns, it would make more sense for the N-terminal helix to cover the outside hydrophobic patch and the C-terminal helix conver the inside hydrophobic patch. Perhaps we could put constraints on the following pairs to tuck the helices this way: L17 V28 L17 P42 I13 L40 A9 P38 L139 A132 l144 V85 M145 V130 L151 P127 The constraints C-terminal helix may not be completely consistent. I'll put them into a costfcn and see which existing model scores best, take a look at it, and see if I can come up with better constraints. Sun Jul 30 18:24:29 PDT 2006 Kevin Karplus It looks like try31-opt2 gets the N-terminal helix in a reasonable position, but has a strand in place of the C-terminal helix. Why did that score so well. Ah---that's because I took sheet constraints from the top-choice try43-opt2, which *also* has the C-terminus as part of the sheet. I should take them from try44 or try45 if I want a C-terminal helix. Trying again for the try48 costfcn. With the try48 costfcn, try45-opt2 and try44-opt2 do indeed score best, but we could improve them by taking all but E135-D154 from the best model (try43-opt2). I could change the C-terminal constraints to Constraint L144.CB A132.CB -10 5.8 8 0.5 Constraint L144.CB V83.CB -10 5 8 0.5 Constraint V152.CB P127.CB -10 5 8 0.5 Constraint L151.CB P127.CB -10 5 8 0.5 (requesting a tighter packing of the C-terminal helix) and change the N-terminal constraints to Constraint I13.CB L40.CB -10 6.4 8 0.5 Constraint I13.CB P38.CB -10 4.5 8 0.5 Constraint A9.CB P38.CB -10 4.5 8 0.5 Constraint L17.CB V28.CB -10 4.3 8 0.5 Sun Jul 30 18:57:48 PDT 2006 Kevin Karplus I started try48 on abyss as a polishing run for chimera-43-44, with the constraints listed above to get the packing. After that I should do a polishing run with breaks and clashes turned up, because the models are still a little bit gappy. Sun Jul 30 20:00:00 PDT 2006 Kevin Karplus I killed try48, because I had a typo, doing 500 conformations instead of 50 conformations per generation in the second OptConform. I think that try48-opt1 is pretty good, so I'll try polishing it without waiting for the opt2 model. Sun Jul 30 20:07:21 PDT 2006 Kevin Karplus try49 polishing try48-opt1 on abyss. I should probably also do a polishing run for try43. I made a try50 costfcn, but it likes try31 much better than try43, which is way down the list. Perhaps try43 is not a good starting point, except for the N-terminus. Maybe I should make a chimera-43-31 with M1-D23 from try43-opt2 and the rest from try31-opt2 and polish it instead. Sun Jul 30 20:20:56 PDT 2006 Kevin Karplus try50 started on whidbey to polish chimera-43-31. Sun Jul 30 20:35:16 PDT 2006 Kevin Karplus I should probably include try47 (both helices outside) as one of the models to submit. I like the N-terminal helix of try47 better than that of try43. Perhaps I should have pasted it into try31 to make a chimera to polish. I should have at least one each of Nhelix-outside-Chelix-inside, Nhelix-outside-Csheet, Nhelix-outside-Chelix-outside. try47-opt2 will probably be the representative for Nhelix-outside-Chelix-outside. try50-opt2 will probably be the representative for Nhelix-outside-Csheet try49-opt2 will probably be the representative for Nhelix-outside-Chelix-inside Nope: try44 scores much better with the try49 costfcn. It looks like try48 and try49 were a waste of time. I made a new "clash.costfcn" similar to unconstrained.costfcn, but with clashes and breaks turned up. It currently favors try31-opt2, but it may favor try50 when that run is finished. I may want to do yet another chimera to put the C-terminal helix on a good initial model (try43 was not all that great an initial model). Sun Jul 30 21:09:48 PDT 2006 Kevin Karplus try50 does not quite catch up to try31 on the try50 costfcn nor the clash.costfcn. Let's try making a chimera-47-31 with M1-A26 from try47-opt2 and the rest from try31-opt2. Sun Jul 30 21:20:14 PDT 2006 Kevin Karplus try51 started on abyss to try to polish chimera-47-31 Currently, I am favoring ReadConformPDB T0370.try31-opt2.pdb # best-scoring helix-outside, sheet ReadConformPDB T0370.try44-opt2.pdb # best of helix-out, helix-in ReadConformPDB T0370.try45-opt2.pdb # second best helix-out, helix-in ReadConformPDB T0370.try46-opt2.gromacs0.repack-nonPC.pdb.gz # best rosetta, helix-inside, sheet ReadConformPDB T0370.try47-opt2.pdb # helix-out, helix-out But I might throw out try45-opt2 in favor of some other model (maybe the try43-opt2 that Cynthia and Sylvia liked best). With any luck, try51 will replace try31. Sun Jul 30 21:58:38 PDT 2006 Kevin Karplus The best-scoring server model is ROBETTA_TS1, which does not close the barrel but has N-helix-inside, C-sheet conformation (like try46). Sun Jul 30 22:16:07 PDT 2006 Kevin Karplus try51 *does* beat try31 (barely) on try51.costfcn, so I will promote try51 to the top slot. Note: with the clash.costfcn, try51-opt2 comes after try31 and try50, because of slightly higher clashes and breaks. Sun Jul 30 22:47:46 PDT 2006 Kevin Karplus submitted with comment We had several good hits, all to PNP-oxidase-like family (Pyridoxine 5'-phoshate oxidase). The automatic method did not result in a compact fold---we had to fuss around for quite a while to get the barrel to close. Even after we got a nice barrel, we still had the termini to place. We ended up with two choices for the N-terminus and three for the C-terminus: N-terminal helix outside the barrel N-terminal helix inside (covering the end of) the barrel C-terminal sheet C-terminal helix outside the barrel C-terminal helix inside (covering the end of) the barrel There are 5 combinations that make sense (there doesn't seem to be room for two helices inside). Model 1 is try51-opt2, our best-scoring model with N-terminal outside, C-terminal sheet. This model is low in clashes, but seems a bit foamy. Model 2 is try44-opt2, the best we managed to get with N-terminal outside, C-terminal helix inside. Model 3 is try46-opt2.gromacs0.repack-nonPC, rosetta's favorite of the backbones we had it repack sidechains for. It is an N-terminal inside, C-terminal sheet model. Model 4 is try47-opt2, an N-terminal outside, C-terminal helix outside model. (Note: try51-opt2 has the same N-terminal. In fact, we made a chimera of try31-opt2 with the N-terminus of try47-opt2 and optimized it to get try51-opt2.) Model 5 is chimera30-33-30, an N-terminal inside, C-terminal outside model. We did not have time to get undertaker to optimize this model (attempts to do so resulted in models like try47, with both helices outside). As a result, this model scores poorly and is unconvincing even to those who created it. We include it here for completeness. ------------------------------------------------------------ Fri Aug 11 03:37:05 PDT 2006 Kevin Karplus T0370 was selected as a refinment model. tr370 scores terribly, but the score can be improved somewhat by running through gromacs and repacking sidechains. This is an N-terminal inside, C-terminal strand model, though the C-terminal strand is not well formed. The packing of the helix is a bit different form our model 3 (try46-opt2), and it seems to be trying to add M1-T3 as yet another strand to the sheet. It might be worthwhile to try optimizing tr370 without constraints first, then try adding sheet constraints to form the C-terminal sheet, if it isn't found by itself. Fri Aug 11 06:48:51 PDT 2006 Kevin Karplus We didn't do the optimization for the original models in a dimer context, but we probably should have, as the top templates are all dimeric. Now that we have the topology determined, dimerization is more straightforward. Mon Aug 14 16:01:37 PDT 2006 Sylvia Do I'm making dimers from try51-opt2, but one of the helices keep smashing into the dimer helix. It has done this from three different templates: 2arzA and 2fhqA and 2aq6A I'm going to ditch try51 and try try46-opt2.gromacs0.repack-nonPC instead, since that has the N-terminus on the inside. The dimer looks much better. I haven't run any of them through Undertaker yet, so I don't know for sure which will turn out better. dimer/decoys try1 is based off the dimer using try46-opt2.gromacs.repack-nonPC. It is running on orcas. Tue Aug 15 10:14:58 PDT 2006 Cynthia Hsu dimer/try1-opt2 looks pretty nice. There are, however, a number of fairly large breaks: T0370.try1-opt2.pdb.gz breaks before (T0370)P248 with cost 7.77076 T0370.try1-opt2.pdb.gz breaks before (T0370)P94 with cost 5.35304 T0370.try1-opt2.pdb.gz breaks before (T0370)A37 with cost 4.64799 T0370.try1-opt2.pdb.gz breaks before (T0370)A191 with cost 4.53743 T0370.try1-opt2.pdb.gz breaks before (T0370)G254 with cost 3.00814 T0370.try1-opt2.pdb.gz breaks before (T0370)T205 with cost 2.94048 T0370.try1-opt2.pdb.gz breaks before (T0370)G100 with cost 2.50411 T0370.try1-opt2.pdb.gz breaks before (T0370)T156 with cost 2.0098 T0370.try1-opt2.pdb.gz breaks before (T0370)M155 with cost 1.74913 T0370.try1-opt2.pdb.gz breaks before (T0370)R80 with cost 1.44989 None of these breaks seemed to have existed in the original monomer on which it was based. I found one problem, at least. I added a KnownBreak before M155, where the two chains would have separated. This will hopefully prevent the formation of the random helix that seems to be joining the two monomers. Because I'm hoping that the breaks in the dimer were mostly because off-phase beta sheets, i added the SheetConstraints from try46-opt2.sheets with a weight of 2. For thoroughness, I also included try46-opt2.helices, although these were only given a weight of 1. I removed "maybe_ssbond" because of the absence of cys residues. I also removed "maybe_metal", as there were no His residues either. I raised "n_ca_c" to 7, and "bad_peptide" to 20. I also raised "break" to 100. Because dimer-from-try46-opt2.gromacs0.repack-nonPC.pdb.gz-template-2aq6A.pdb scores the best out of all models, I decided to leave it as the ReadConformPDB input. Examining the dimerization interface in the original model, I thought it might also be worth putting in distance constraints between the residues in the area. I also turned off TweakMultimer in the try2.under file, in order to pack a tighter interface. I considered adding the "Multimer 2" command to make it into a cyclic dimer, but I have my doubts as to whether this really what is considered a "cylic dimer" and hence, whether that would actually be a good idea. On second thought, it was already in the original file, so perhaps it is better left as is. When I look at the model, the two dimers do seem to interface in a cyclic fashion (the two identical regions always being on opposite sides of the circle), so I decided that perhaps this was indeed a cylic dimer. I also commented out all references to including undertaker alignments. dimer/mult2 is currently running on shaw. Tue Aug 15 12:57:03 PDT 2006 Sylvia Do I'm going to do another dimer based off try46 gromacs repack but with the template 2arzA. I'm also going to make one from template 1dnlA. the best top hit for t06. I'll make one for all of them too (1rfeA). I won't add or change any of the constraints until the second try for each dimer. mult3 wiill be try46-based-off-2arzA dimer. mult3 is running on lopez. Tue Aug 15 14:05:59 PDT 2006 Cynthia Hsu I've decided, looking at the model generated for 1rfeA, to add distance constraints to the try4.costfcn, despite Sylvia's intentions. For try4.under, I turned TweakMultimer up to 0.5, hoping to build a tighter interface. I left the Multimer 2 option in the .under file, and set up ReadConformPDB dimer-from-try46-opt2.gromacs0.repack-nonPC-template-1rfeA.pdb dimer/mult4 is currently running on orcas. I also used make-dimer3.under to generate the new dimer dimer-from-try46-opt2.gromacs0.repack-nonPC-template-1dnl.pdb. Tue Aug 15 15:19:12 PDT 2006 Sylvia Do Since the dimers created from the different templates are mroe or less the same, I'm just going to use the dimer used in try1/2 (dimer-from-try46-opt2.gromacs0.repack-nonPC.pdb.gz-template-2aq6A.pdb) and the new dimer dimer-from-try46-opt2.gromacs0.repack-nonPC-template-1rfeA.pdb, which was the top hit for t2k, t04, t06 combined. I'm currently optimizing try2 by increasing the weight for breaks to 150. The KnownBreak didn't work though. I'm just going to tweak the .under file slightly to see if it was something in that. mult5 is running on vashon. mult6 will be like mult5, but with the 1rfeA tempplate instead. mult6 is running on whidbey Wed Aug 16 14:00:28 PDT 2006 Cynthia Hsu Just reviewed the README, and it seems that we have been approaching this the wrong way. We should be try optimizing then dimerizing the original model, not our own alignments. Following Professor Karplus's advice, I examined the unconstrained scores, and found that the gromacs repacked version of the refinement model was the top scoring of those released. Because of the hideous "n_ca_c" and "bad_peptide" scores, I doubled both of these to 10 and 20, respectively. "soft_clashes" was raised to 40 and "break" to 100. It seems that given the current score of 8.2, it will take much more than raising the weight on "break" to fix it, but as this is a refinement model, I'm not entirely sure what should be done. Other than that, I decided to see if this would improve the refinement model, so the remainder of the weights were left identical to those in try1. try52.under used ReadConformPDB tr370.gromacs0.repack-nonPC.pdb.gz try52 is currently running on orcas. Wed Aug 16 17:56:24 PDT 2006 Cynthia Hsu try11 looks a bit foamy, but as I don't want to stay for the night, I've decided to just let it optimize as a dimer. On second thought, looking at the unconstrained scores, it scores very very badly on all accounts. I've decided instead to copy the try52 costfcn to try53.costfcn. "dry5" was raised to 25, "dry6.5" to 30, and "dry8" to 20. "n_ca_c" was raised to 20, and "bad_peptide" to 30, as these still have high costs. I also raised "soft_clashes" to 60 and "break" to 150. I also remembered to comment out dssp-ehl2.constraints, undertaker-align.sheets, and rr.constraints this time. try53.under used ReadConformPDB T0370.try52-opt2.pdb. try53 is currently running on orcas. Thu Aug 17 11:47:54 PDT 2006 Cynthia Hsu try53 still scores pretty badly, but not as badly as try52 does. I wonder if its low scores are due to the large number of exposed residues at the dimerization interface. However, in light of the fact that the C terminal sheet, despite having appeared, looks as if it's about to disintegrate at any second, it will probably be better to run one more optimization, this time with a sheet constraint. I raised "dry5" to 30 and "dry6.5" to 35, but refrained fro modifying "dry8" as I didn't want the dimerization interface to close. I also was concerned about "n_ca_c", which I raised to 30. I also removed the "maybe_ssbond" weight, due to the absence of Cys residues. I also added the following sheet constraint, although I am not quite sure I got the phase of it right: SheetConstraint (T0370)R129 (T0370)E135 (T0370)M145 (T0370)L139 hbond (T0370)R134 1 try54.under used try53-opt2.pdb for ReadConformPDB. try54 is currently running on orcas. Thu Aug 17 13:28:52 PDT 2006 Sylvia Do try54-opt1 just finished, and it doen't look like the sheetconstraints worked properly. For try55, I'm just going to take try46's sheet constraints for that area and use it instead of the sheetconstraints in try54. try55 is running on vashon. Thu Aug 17 14:35:18 PDT 2006 Cynthia Hsu I'm concerned that the SheetConstraints in try46 do not match those desired by the refinement model, so I've instead modified try54.costfcn, raising the weight on the SheetConstraint to 10. I also raised "constraints" to 12, and removed "maybe_metal". In addition, I raised "dry5" to 35, "dry6.5" to 30, "n_ca_c" to 30, "bad_peptide" to 35, "soft_clashes" to 100, and "break" to 240. try56.under used try54-opt2.pdb for its ReadConformPDB. try56 is currently running on camano. Thu Aug 17 15:48:50 PDT 2006 Sylvia Do try56 does not seem to be working.I'im going to try again with try54, but with these constraints: SheetConstraint (T0370)R129 (T0370)W133 (T0370)M145 (T0370)G141 hbond (T0370)R134 5 SheetConstraint (T0370)G141 (T0370)M145 (T0370)T153 (T0370)Q149 hbond (T0370)R146 5 Maybe together each can keep the other as a sheet. try57 is running on vashon. Fri Aug 18 12:09:54 PDT 2006 Cynthia Hsu try57 still failed. I copied try56.costfcn to try58.costfcn, then raised "constraints" to 15. I lowered "sidechain" to 2. I also added the second SheetConstraint from try56: SheetConstraint (T0370)G141 (T0370)M145 (T0370)T153 (T0370)Q149 hbond (T0370)R146 5 I also decided, as an afterthought, to raise "hbond_geom_backbone" to 15, "hbond_beta" to 70, "hbond_beta_pair" to 120. This seems like a rather last ditch attempt. I think if this fails, it may be in our interest to give up on the sheet and just form the dimer, regardless. try56-opt2 is still the best scoring when unconstrained, so I've decided to use that in the try58.under ReadConformPDB. try58 is currently running on orcas. Fri Aug 18 14:41:23 PDT 2006 Cynthia Hsu try58-opt1 actually did better than its opt-2, but neither try did better than try56. I've decided to just form the dimer as is. I used the top alignment from Undertaker, 1rfeA, as the template. I used make-dimer4.under to generate dimer-from-try56-opt2-template-1rfeA.pdb I then copied the try2.costfcn to try7.costfcn. I decided to not include any constraints, as those in try56 still were not satisfactory. I then replaced the distance constraints with those from RasMol set pick distance from dimer/decoys/dimer-from-try56-opt2-template-1rfeA.pdb. I also raised "n_ca_c" to 10, "bad_peptide" to 20, and restored "soft_clashes" and "break" to their original weights of 40 and 100, respectively. Like the previous dimerization runs, try7 had TweakMultimer turned off and a Multimer 2 command. It used ReadConformPDB dimer-from-try56-opt2-template-1rfeA.pdb dimer/mult7 is currently running on lopez. Sat Aug 19 12:51:20 PDT 2006 Cynthia Hsu I apparently overwrote dimer/mult6. That's currently running on camano; dimer/mult7 is currently running on orcas. Sat Aug 19 16:58:36 PDT 2006 Cynthia Hsu Our refinement model, in dimer form, is doing worse than all previous models except for try6. I raised "dry5" to 20, "dry6.5" to 25, and "dry8" to 20. I raised "n_ca_c" to 15 and "bad_peptide" to 30. I also raised "soft_clashes" to 60, "break" to 150, "hbond_geom_beta" to 60, and "hbond_geom_beta_pair" to 110. dimer/mult8 is currently running on orcas. Sun Aug 20 13:26:36 PDT 2006 Cynthia Hsu The try8 of the dimer is an improvement over try7, but just barely. I raised "n_ca_c" to 30, "bad_peptide" to 45, "soft_clashes" to 80, and "break" to 200. try9.under was a polishing run of try8-opt2. dimer/mult9 is currently running on orcas. Sun Aug 20 17:40:03 PDT 2006 Cynthia Hsu try9 only showed marginal improvement over try8. I decided to experiment with removing the distance constraints, and raising "soft_clashes" to 100 and "break" to 240. try10.under used ReadConformPDB to polish try9-opt2. try10 is currently running on orcas. Sun Aug 20 23:50:02 PDT 2006 Cynthia Hsu try10 improved in breaks, but did marginally worse in other areas, so that it had the same overall score as try9. I copied try9.costfcn to try11.costfcn, then raised "break" to 300 and "soft_clashes" to 120. try11.under used try9-opt2.pdb for polishing. try11 is currently running on orcas. Mon Aug 21 10:39:57 PDT 2006 Cynthia Hsu try11 did worse than try9 and 10. Interestingly enough, the raised "break" weight made the break cost higher and lowered that of "n_ca_c". "phobic_fit", however, is significantly worse. I've decided to apply the try11.costfcn to try10-opt2, and see if there is any improvement in results. I also raised "n_ca_c" to 35 and "bad_peptide" to 50. On second thought, I lowered the weight on "break" to 280. Perhaps the incrementation was too rapid. In addition, I lowered the weight on the distance constraints to 0.5 each. dimer/mult12 is currently running on orcas. Mon Aug 21 11:33:44 PDT 2006 Cynthia Hsu After discussing with Professor Karplus, I realize that we shouldn't have abandoned the formation of the sheet constraint so soon. I copied try58.costfcn to try59.costfcn, but lowered all values related to polishing: "soft_clashes" to 30, "break" to 80, "n_ca_c" to 7, "bad_peptide" to 12, "dry5" to 15, "dry6.5" to 25, and "dry8" to 20. I also lowered "hbond_geom_backbone" to 15, "hbond_geom_beta" to 50, and "hbond_geom_beta_pair" to 100. I then raised the weight of "constraints" to 20. I also decided to add the following SheetConstraints from try51-opt2.sheets: SheetConstraint (T0370)E86 (T0370)L92 (T0370)R128 (T0370)Y122 hbond (T0370)L90 5 SheetConstraint (T0370)V130 (T0370)E135 (T0370)M145 (T0370)S140 hbond (T0370)A132 10 SheetConstraint (T0370)R142 (T0370)R146 (T0370)T153 (T0370)E149 hbond (T0370)L144 10 Running make score-all.try59.pretty on this showed that try31 actually had the best score in this regard. try55 seems to be the best of those models generated from the refinement. Because of this, I decided that it would be best to polish try55-opt2. Following Professor Karplus's advice, I've raised "ReduceConstraint" to 20 and lowered "InsertSpecificFragment" to 5 in the try59.under file. try59 is currently running on camano. Mon Aug 21 14:30:38 PDT 2006 Cynthia Hsu The added SheetConstraint caused a very large break. I've decided that it may be slightly more effective to use some of our original sheet constraints. I also found that the sheet constraint for the third sheet, the one already in the barrel, was not actually applicable to the desired sheet. The SheetConstraints were modified to the following: SheetConstraint (T0370)R129 (T0370)E135 (T0370)M145 (T0370)L139 hbond (T0370)V130 10 SheetConstraint (T0370)R142 (T0370)R146 (T0370)T153 (T0370)E149 hbond (T0370)L144 10 SheetConstraint (T0370)R80 (T0370)G87 (T0370)E135 (T0370)R128 hbond (T0370)L82 5 try60 is currently running on orcas. Mon Aug 21 16:40:54 PDT 2006 Cynthia Hsu try60 still didn't form the sheet. I examined the model and found, interestingly enough, that L144 and M145 were pointed the same direction, despite the fact that they were adjacent to each other. I decided that this may have been because of the higher "bad_peptide" weighthich would manipulate it in a certain direction, and also the low "sidechain" weight, which would allow the coil too much freedom of movement. I returned "bad_peptide" and "sidechain" to their original values, of 10 and 5, respectively. "dry6.5" was also returned to 20 and "dry8" to 15. try61.under was identical to try60.under (polished try55-opt2). try61 is currently running on orcas. Tue Aug 22 09:06:53 PDT 2006 Cynthia Hsu try61 still doesn't have the desired sheet. Examining hydrophobic CBs in spacefill, I found that 136 and 139 seemed out of phase. Because the sheets currently have the three residue hairpin turn between them, I didn't want to extend the sheets. However, I realized that I had inappropriately labeled residue 135 and Alanine when it was actually Glutamine. I made the appropriate change, and then raised the weight on "constraints" to 25. try62 is currently running on orcas. Tue Aug 22 13:35:36 PDT 2006 Sylvia Do, Cynthia Hsu try62 failed to form a complete sheet, but at the very least, more of a sheet is formed (from residues E138-G141). I doubled the weight on the SheetConstraints, as follows: SheetConstraint (T0370)R129 (T0370)E135 (T0370)M145 (T0370)L139 hbond (T0370)V130 20 SheetConstraint (T0370)R142 (T0370)R146 (T0370)T153 (T0370)E149 hbond (T0370)L144 20 SheetConstraint (T0370)R80 (T0370)G87 (T0370)E135 (T0370)R128 hbond (T0370)L82 10 We ran the make on score-all.try62.pretty, hoping to find the best model to polish for improved SheetConstraints. Because of this, in try63.under, we used try62-opt2 for ReadConformPDB. try63 is currently running on orcas. Tue Aug 22 16:13:31 PDT 2006 Sylvia Do, Cynthia Hsu try63 still has no improvement in the sheet. A132 and R142 are adjacent to each other, but the residues are pointing in opposite directions. We found that the phase in question was not between these two sheets, but within the next group of sheets. As a result, we changed the SheetConstraint to the following: SheetConstraint (T0370)R142 (T0370)R146 (T0370)D154 (T0370)W150 hbond (T0370)L144 20 Hopefully this will work. Because it did the best out of the refinement models on score-all.try64, we decided to use try63-opt2 for polishing. try64 is currently running on orcas. Wed Aug 23 10:24:21 PDT 2006 Cynthia Hsu Still no luck. For the try65.costfcn, I changed the hbond from L144 to E143. Hopefully this will slide the sheet down by one residue, forming a three helix turn and the appropriate sheet. I also lowered "hbond_geom" to 3 and "hbond_geom_backbone" to 5, to prevent hydrogen bonding between irrelevant members of the backbone from pulling apart the sheet. As before, I used the score-all.try65.pretty functionality to determine the best-scoring model under constraints, which in this case was T0370.try64-opt2.pdb. I was quite relieved that it was scoring better than the models we had generated before the refinement. try65.under was a polishing run of try64-opt2. try65 is currently running on orcas. Wed Aug 23 12:22:57 PDT 2006 Kevin Karplus The tr370 model is a N-terminal helix inside, C-terminal sheet model. Our best such model was try46. For refinement models, we only want to look at try52 and later. None of the dimer runs were properly done, so they aren't much use. Wed Aug 23 12:55:34 PDT 2006 Sylvia Do So, mult6 was done incorrectly. Prof. Karplus pointed out that in the .under file, the ReadConformPDB was not reading anything (the dimer model was written into the wrong area). Anyway, Fixed this and reran mult6. mult6 is running on orcas. Wed Aug 23 14:00:49 PDT 2006 Cynthia Hsu Under Grant's suggestion, I modified the try65-opt2.pdb to have a break, reduced the cost of "break" and "soft_clashes", then repeated the run, hoping that this would induce sheet formation. try66.under uses ReadConformPDB T0370.try65-opt2_broken.pdb.gz try66 is currently running on orcas. Wed Aug 23 15:59:36 PDT 2006 Cynthia Hsu Under Professor Karplus's suggestion, I decided to make a monomer out of try46-opt2 and the top scoring polishing of the refinement model, try64-opt2. This produced the model try64-opt2+try46-opt2.pdb, which I then converted into chimera64-46.pdb. For polishing chimera64-46.pdb in try67.under, I copied the try1.costfcn, removing "maybe_metal" and "maybe_ssbond". I then raised the weight on "soft_clashes" to 40 and the weight on "break" to 100. "n_ca_c" was raised to 7, "bad_peptide to 12, and "dry6.5" to 25 and "dry8" to 20. The dssp-ehl2 constraints, undertaker-align.sheets, and the rr constraints were commented out. try67 is currently running on lopez. Wed Aug 23 16:41:22 PDT 2006 Sylvia Do Did a dimer of our best scoring model so far (try64-opt2). Now, I'm going to polish it. Added some distance constraints to make sure that the dimer won't get too far from each other, and changed some costs to make the dimer interface closer. Namely, dry6.5 is upped to 25, dry8 is upped to 20 soft_clashes upped to 50 break upped to 75 mult13 is running on orcas. Thu Aug 24 03:34:29 PDT 2006 Sylvia Do Looks like try66 never finished? I'd look at the log files, but it's 3 in the morning. try67 looks promising. It has a better sheet, but it also has a helix there now. mult13 is doing okay, but has a few breaks. I've upped the breaks to 150 in mult 14. mult14 is running on orcas. Thu Aug 24 10:48:59 PDT 2006 Cynthia Hsu Polishing the chimera produces a short helix where a sheet should be. I copied over the following SheetConstraint from try46-opt2.sheets: SheetConstraint (T0370)R128 (T0370)E135 (T0370)D147 (T0370)S140 hbond (T0370)V130 15 I also raised "constraints" up to 15. try68.under polished try67-opt2.pdb. try68 is currently running on orcas. Thu Aug 24 12:37:33 PDT 2006 Sylvia Do mult14 still has bad breaks. Gah! I'm going to up the break costs. If this doesn't close up some of the gaps, I'll have to try something else. break is now 250. Try13 is actually doing better than try14. One of the breaks did get worse in try14, so no big surprise. Though as a whole, 14 scored better in breaks. Anyway, the dimer interface is still foamy, so I also upped dry6.5 to 30, dry8 to 25, and dry12 to 7. I also upped soft_clashes to 75 and backbone_clashes to 5.Also, it is not doing well in n_ca_c, so I now have it at 7. mult15 is running on vashon. Thu Aug 24 16:10:56 PDT 2006 Sylvia Do Yay! Cynthia's sheet constrant is working well; however, now one of the other sheets has disappeared. More sheetconstraints, coming up. If all works well, this should fix things. I think it will only restore the former sheet, but it won't connect to the new sheets. We'll try to fix it when we get there. //Added from try67-opt2.sheets SheetConstraint (T0370)D81 (T0370)G87 (T0370)R134 (T0370)R128 hbond (T0370)L82 15 try69 is running on peep. Thu Aug 24 18:14:08 PDT 2006 Cynthia Hsu The sheet still insists on forming a helix. I viewed the protein model in RasMol, with the carbon betas in spacefill. I then used set pick distance to add distance constraints to hopefully hold the two strands of the sheet together. try69-opt2 was polished by try70.under try70 is currently running on orcas. Fri Aug 25 10:41:20 PDT 2006 Cynthia Hsu The CB atoms of the two adjacent beta strands seem to fall within the appropriate ranges, but they are off-centered. I decided to lower the distance constraint ideal by 0.3, and to lower the upper bound of these to 0.1 less than the ideal was, originally. try71 is currently running on orcas. According to Professor Karplus, I seem to have done the chimerization improperly. This time, instead of using the make-chimera.under script, I used the refined-models.pdb.gz to superimpose the models at the appropriate location, between A132 and W133. This made the pdb file chimera64-46-no2. For the try72.costfcn, I copied try67.costfcn. try72 is currently running on orcas. Fri Aug 25 12:53:19 PDT 2006 Sylvia Do Mult15 is scoring better than mult13 by rosetta, but worse by undertaker. Try72 is a bloody success! Awesome! No major breaks either. Let's see how well it does on the other constraints. Well, it beats the other tr models we have so far. I'm going to optimise try72. In th .under file, I've change the readconformpdb to try72-opt2. The break cost is now 125, n_ca_c is now 9, phobic_fit is now 4, and soft_clashes is upped to 50. try73 is running on peep. Fri Aug 25 15:19:18 PDT 2006 Sylvia Do I made a dimer out of try72 (I will make a dimer out of 73 if it comes out better). I am now running it through undertaker. I've added the following distance constraints to keep the chains together: Constraint A31.CB A185.CB -10 5.5 6.0 0.5 Constraint W27.CB W181.CB -10 4.3 4.8 0.5 Constraint P38.CB P192.CB -10 6.3 6.8 0.5 Constraint A37.CB A191.CB -10 12.1 12.6 0.5 Constraint L82.CB L236.CB -10 18.4 6.2 0.5 mult16 is running on vashon. Fri Aug 25 16:10:12 PDT 2006 Sylvia Do, Cynthia Hsu Yay! try73 is great, and we're sneaking up the unconstrained.pretty scores. We are going to optimise it a little more by upping the costs for n_ca_c, bad_peptide, soft_clashes, and breaks. n_ca_c 12 bad_peptide 15 \ soft_clashes 70 backbone_clashes 2 \ break 175 \ try74 is running on peep. Fri Aug 25 16:56:01 PDT 2006 Sylvia Do, Cynthia Hsu Current top models (in Cliff's notes summary): try73 - chimera, optimized, from try46-opt2 (C-terminal sheet) and try64-opt2 (polished refinement) try64 - best based only on refinement model - lacks C-terminal sheet. dimer/try13 - dimerization from try64. dimer/try16 - dimerization from try72. try67 - chimera, inappropriate superimposition - best favored by Rosetta. Sat Aug 26 07:38:14 PDT 2006 Sylvia Do mult16 is doing very well in the dimers' score-all.unconstrained.pretty. With that note, since try73 did better than try74, I made a dimer of that one. I used mult16's constraints, since the distance constraints shouldn't be that far off. current top models: same as above, but move mult16 up one. mult17 is running on peep. Sun Aug 27 20:13:42 PDT 2006 Kevin Karplus I've put all the models mentioned above into superimpose-refine.under, to take a look at them. I'm a little disturbed that they all come from chimeras, as I think we need at least one model that has been refined from the tr370 model with any cutting and pasting. InFilePrefix decoys/ ReadConformPDB T0370.try73-opt2.pdb ReadConformPDB T0370.try64-opt2.pdb InFilePrefix dimer/decoys/ ReadConformPDB T0370.try17-opt2.unpack.pdb chain A ReadConformPDB T0370.try13-opt2.unpack.pdb chain A InFilePrefix decoys/ ReadConformPDB T0370.try67-opt2.gromacs0.repack-nonPC.pdb Sun Aug 27 20:19:01 PDT 2006 Kevin Karplus OK, I'm not willing to submit try67, no matter how much rosetta likes it. try73-opt2 and dimer/try17-opt2 both seem very similar, with similar movements of the N-terminal helix and with the C-terminal sheet forming fairly well. There are only subtle differences in sidechains between the two. The dimeric optimization does not seem to be changing much. It is interesting that the beginning and end of the chain are attempting to add another strand to the sheet in try73 (and hence dimer/try17): SheetConstraint M1 T2 T153 V152 hbond M1 1 There are still some bad breaks: T0370.try73-opt2.pdb.gz has 23 breaks T0370.try73-opt2.pdb.gz breaks before (T0370)Y120 with cost 0.642922 T0370.try73-opt2.pdb.gz breaks before (T0370)P113 with cost 0.631333 T0370.try73-opt2.pdb.gz breaks before (T0370)D112 with cost 0.525201 T0370.try73-opt2.gromacs0.repack-nonPC.pdb.gz has 9 breaks T0370.try73-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0370)G96 with cost 1.06683 T0370.try73-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0370)P5 with cost 0.607997 T0370.try73-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0370)A95 with cost 0.591913 T0370.try73-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0370)N18 with cost 0.52062 T0370.try73-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0370)T88 with cost 0.504242 dimer T0370.try17-opt2.pdb.gz has 41 breaks T0370.try17-opt2.pdb.gz breaks before (T0370)M155 with cost 81.1966 T0370.try17-opt2.pdb.gz breaks before (T0370)P192 with cost 0.618025 T0370.try17-opt2.pdb.gz breaks before (T0370)G189 with cost 0.603534 T0370.try17-opt2.pdb.gz breaks before (T0370)S198 with cost 0.509293 T0370.try17-opt2.pdb.gz breaks before (T0370)T263 with cost 0.507081 (All these breaks are in the second monomer!) It looks like try17 did not have "multimer 2" on for the OptConform commands, so the two monomers did not remain identical. I'll do another optimization (starting from try16 and try17) but with multimer optimization turned on to maintain symmetry. Actually, I'll start not with all the try16 and try17 models, but only those run through gromacs, since they have smaller clashes and fewer breaks. Also, try18 will have the sheet constraints from try17-opt2, but no constraints to hold the monomers together. Breaks are turned up to a more reasonable level, and the dry terms are turned up a bit also. I'll reduce the number of generations and conformations per generation to speed things up a little. Sun Aug 27 20:59:28 PDT 2006 Kevin Karplus dimer/mult18 started on cheep. OK, while I'm working on a new dimer optimization, perhaps I should try doing one more optimization from a non-chimeric model. I believe that the non-chmeric models based on tr370 are try52 through try66, and dimer/try7 through dimer/try15 dimer/try7 < dimer-from-try56-opt2-template-1rfeA.pdb dimer/try12 < dimer/try10 < dimer/try9 < dimer/try8 < dimer/try7 dimer/try11 < dimer/try9 dimer/try13 < dimer-from-try64-opt2-1rfeA.pdb dimer/try14 < dimer-from-try64-opt2-1rfeA.pdb dimer/try15 < dimer-from-try64-opt2-1rfeA.pdb From chimeras: dimer/try16 < dimer-from-try72-opt2-1rfeA.pdb dimer/try17 < dimer-from-try73-opt2-1rfeA.pdb Of the non-chimeric dimers from tr370, it looks like rosetta likes try12-opt2.unpack.gromacs0.repack-nonPC best (though the chimeric try17 beats it). With the try18 costfcn, try12-opt2 is the best of the non-chimeric dimers also. The best-scoring of the non-chimeric monomers (with the unconstrained costfcn) is try64-opt2. Of the dimers made from try64, try14-opt2 scores the best with the try18 costfcn. It may be worth reoptimizing with the try18 costfcn, to get a non-chimeric dimer optimization. Sun Aug 27 21:29:05 PDT 2006 Kevin Karplus dimer/mult19 started on peep to optimize try13-opt2,try14-opt2,try15-opt2, starting from gromacs optimized models (gromacs0 and gromacs0.repack-nonPC). Sun Aug 27 21:37:27 PDT 2006 Kevin Karplus Looking at dimer/try18-opt1, I see some burial problems. In particular, I think that V41 and L43 should be buried, perhaps by W287 (W133 of chain B) perhaps by L293 (L139 of chain B). Perhaps for try20, I'll try adding some distance constraints for these pairings (and the corresponding ones with chain A and chain B swapped). Sun Aug 27 22:06:50 PDT 2006 Kevin Karplus dimer/try18 looks slightly better than try17 in the try18 costfcn, but Rosetta thinks try18 is a *big* improvement: try18-opt2.unpack.gromacs0.repack-nonPC -202.6 try17-opt2.unpack.gromacs0.repack-nonPC -104.6 Mon Aug 28 07:22:16 PDT 2006 Kevin Karplus Neither rosetta nor try19.costfcn care much for dimer/try19-opt2. The runs optimized from monomer try64 are not doing as well as those from try56. The try20 dimer scores well with the try20 costfcn, beginning to get the constraints for burying the specified residues, but rosetta does not like try20 (though better than try19). The try20 run was interesting in that each superiteration started out optimizing try17-opt2, but most switched to one of the other models as a base. The final optimization was based on try17-opt2, however. The L139 residue does come over to bury V41 and L43 of the other chain, but at the cost of a really horrendous break between R142 and E143. I should try polishing try20, with the new constraints turned down and the break weight turned up. I'll also remove the constraints on the tryptophan. Mon Aug 28 07:51:23 PDT 2006 Kevin Karplus dimer/mult21 started to try to reduce the breaks in try20 (starting from the two gromacs optimized models, which have fewer clashes and slightly smaller breaks, though worse packing scores and sidechains). Mon Aug 28 09:57:37 PDT 2006 Kevin Karplus rosetta doesn't care much for dimer/try21, liking it only slightly better than dimer/try20, but it scores best with the try21 costfcn, and breaks and clashes are way down from try20 (though not as low as try18). There are still 3 bad breaks in each monomer: T0370.try21-opt2.pdb.gz has 21 breaks T0370.try21-opt2.pdb.gz breaks before (T0370)L144 with cost 0.907765 T0370.try21-opt2.pdb.gz breaks before (T0370)L298 with cost 0.905978 T0370.try21-opt2.pdb.gz breaks before (T0370)A249 with cost 0.668222 T0370.try21-opt2.pdb.gz breaks before (T0370)A95 with cost 0.666619 T0370.try21-opt2.pdb.gz breaks before (T0370)E143 with cost 0.621564 T0370.try21-opt2.pdb.gz breaks before (T0370)E297 with cost 0.620521 gromacs makes the breaks worse: T0370.try21-opt2.unpack.gromacs0.pdb.gz has 20 breaks T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)G96 with cost 1.23898 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)L97 with cost 1.0086 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)L144 with cost 0.998467 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)G186 with cost 0.977235 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)A95 with cost 0.737711 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)P159 with cost 0.735011 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)A249 with cost 0.723392 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)P5 with cost 0.645708 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)E143 with cost 0.533693 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)P196 with cost 0.520088 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)A185 with cost 0.518874 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)E297 with cost 0.509593 T0370.try21-opt2.unpack.gromacs0.pdb.gz breaks before (T0370)P42 with cost 0.502816 I should do another polishing run with *no* constraints, with breaks and dry terms turned up. Mon Aug 28 10:21:38 PDT 2006 Kevin Karplus dimer/mult22 started on cheep to polish the gromacs optimized versions of try20 and try21, with no constraints. Mon Aug 28 12:57:12 PDT 2006 Kevin Karplus dimer/try22 doesn't score quite as well as try18 and try21 with the try22 costfcn, but it has reduced gaps and clashes from try21 (its sidechain score is not quite as good as try21's). try22-opt2.unpack.gromacs0.repack-nonPC does score almost as well as the try12-opt2 version with the rosetta cost function, but it still way behind try17 and try18. I should probably polish try22 with something like the unconstrained costfcn, with the hope of getting it to be a bit more acceptable to the cost functions. I like it the best of the models we've generated, even though it isn't scoring that well yet. Mon Aug 28 15:59:59 PDT 2006 Kevin Karplus I think we may have reached diminishing returns now, and the deadline is only 20 hours away, so I've submitted refinement models ReadConformPDB T0370.try18-opt2.unpack.gromacs0.repack-nonPC.pdb chain A # best with rosetta ReadConformPDB T0370.try23-opt2.unpack.gromacs0.repack-nonPC.pdb chain A # dense packing, good rosetta score ReadConformPDB T0370.try12-opt2.unpack.gromacs0.repack-nonPC.pdb chain A # best non-chimeric with rosetta ReadConformPDB T0370.try17-opt2.unpack.pdb chain A # best with unconstrained, try23 ReadConformPDB T0370.try21-opt2.unpack.pdb chain A # best with try22.costfcn (other than try18) with comment For refining tr370, we tried both monomeric refinement and dimeric refinement. In the end, all the models that we are submitting have been refined as dimers. Our basic refinement protocol consisted either of repeatedly optimizing with undertaker, or of optimizing the model using undertaker, then reoptimizing with gromacs, and repacking the sidechains (except CYS and PRO) using rosetta. For each iteration of this protocol, we changed the undertaker cost function, to emphasize those features of the model that we felt needed the most work. This protocol alone did not seem adequate for getting the C-terminal sheet cleaned up, so we also tried making a chimera, using the refinement model for M1-A132, but taking W133-D154 from a model we had generated independently, which had better H-bonds for the final strands. Refinement model 1 is try18-opt2.unpack.gromacs0.repack-nonPC chain A, which had the lowest Rosetta energy of any of the dimer refinement models that Rosetta repacked sidechains for. It was optimized by undertaker from dimer/try17-opt2.unpack.gromacs0, then reoptimized with gromacs and with sidechains repacked by rosetta. dimer/try17-opt2 was optimized by undertaker from dimer-from-try73-opt2-1rfeA, which was a superposition of the monomer try73-opt2 on the 1rfe dimer. The monomer try73-opt2 was optimized by undertaker from try72-opt2 (no gromacs or rosetta), which in turn was optimized by undertaker from chimera64-46-no2, the chimera of try64-opt2 and try46-opt2. Try46-opt2 was a model we had generated ourselves (not derived for tr370). try64-opt2 < try63-opt2 < try62-opt2 < try55-opt2 < try53-opt2 < try52-opt2 < tr370.gromacs0.repack-nonPC. That is, the tr370 model was optimized with gromacs and had sidechains repacked by rosetta, then six runs of undertaker were done before the chimera was made. Refinment model 2 is chain A of dimer/try23-opt2.unpack.gromacs0.repack-nonPC, which has dense packing and a good rosetta energy (though not as good as model 1). This model was built to try to get better burial for V41, L43, W133, and L139, which seemed to exposed in try18. dimer/try23-opt2 < dimer/try22-opt2.unpack.gromacs0.repack_nonPC dimer/try22-opt2 < dimer/try21-opt2.unpack.gromacs0 dimer/try21-opt2 < dimer/try20-opt2.unpack.gromacs0 dimer/try20-opt2 < dimer/try17-opt2 Refinment model 3 is chain A of dimer/try12-opt2.unpack.gromacs0.repack-nonPC, which had the lowest Rosetta energy of any of the non-chimeric models optimized directly from tr370. dimer/try12-opt2 < dimer/try10-opt2 < dimer/try9-opt2 < dimer/try8-opt2 < dimer/try7-opt2 < dimer-from-try56-opt2-template-1rfeA The monomeric try56-opt2 was optimized by undertaker from try54-opt2 < try53-opt2 < try52-opt2 < tr370.gromacs0.repack-nonPC For dimer/try12-opt2.unpack.gromacs0.repack-nonPC, optimization by gromacs and repacking by rosetta ocurred only at the beginning (on tr370) and at the end (on dimer/try12-opt2). We probably could have gotten faster convergence had the cycling protocol (undertaker, gromacs, rosetta) been used, since gromacs is good at removing small clashes. Refinment model 4 is chain A of dimer/try17-opt2, which scored the best of the dimeric models with some of undertaker's cost functions, including the unconstrained costfcn and try23.costfcn. Its provenance was described above, as this model was on the path to both try18 and try 23. Refinement model 5 is chain A of dimer/try21-opt2, which scored the best with the try22 costfcn. It is on the path to try23-opt2. ---------------------------------------------------------------------- Sun Sep 24 12:57:14 PDT 2006 Kevin Karplus Several of our refinement models were better than tr370. The best monomeric model was try56-opt1-scwrl (I've not looked at the dimeric models). Sun Sep 24 13:42:42 PDT 2006 Kevin Karplus Our best model was dimer/try21-opt2.repack-nonPC and 2nd best was dimer/try21-opt2 (submitted as refinement model 5). All of our refinements improved on tr370, and all but one of the submitted ones improved on tr370.gromacs0: refine1 dimer/try18-opt2.unpack.gromacs0.repack-nonPC -78.2432 refine2 dimer/try23-opt2.unpack.gromacs0.repack-nonPC -81.6140 refine3 dimer/try12-opt2.unpack.gromacs0.repack-nonPC -80.3779 refine4 dimer/try17-opt2 -75.0710 refine5 dimer/try21-opt2 -87.6441 tr370 -67.4063 tr370.gromacs0 -76.7666 We did not do a great job of selecting the best of the refinements to submit. All the dimer models were created after getting tr370. Monomer models up to try51 are from before tr370. Of our pre-tr370 models, the best was align3 -4.89 try46-opt1-scwrl -4.12 try10-opt1-scwrl -2.94 model3 -1.95 model1 was fairly poor 42.11 bet servers CIRCLE_TS5 -64.30 FAM_TS2 -64.30 best model 1 karypis.srv_TS1 -49.04 It looks like we messed this one up, but so did ROBETTA, whose best was ROBETTA_TS1 30.31 Our server was like our hand alignment SAM_TO6_server_TS3 -1.27 SAM_TO6_server_TS1 10.75 Wed Nov 29 20:38:16 PST 2006 Kevin Karplus Our GDT scores and CASP's for tr370 don't seem to agree. This needs examining---are my GDT scores wrong? Theirs? Mon Jan 1 15:33:24 PST 2007 Kevin Karplus One problem may be that they were using just residues 8-115, not the whole of either chain A or chain B of 2iab. I'll try making T0370_1.real.pdb.gz using just 8-115 of 2iabA, and see if those results match better. For model1, I am reporting GDT=55.9, but they report 64.41. This is possible if my GDT is not sampling sufficiently. The GDT curve I output indicates that I have the GDT computed correctly from the curve. Also from the GDT curve, I get GDT_HA (average of 0.5,1,2,4 Ang)=36.11%, but they report 45.83%. For Zhang-Server_TS1, I am reporting GDT=65.8, but they report 71.53. Tue Jan 2 09:56:18 PST 2007 Kevin Karplus I tried using chain 2iabB instead of chain 2iabA, but this did not imprve my GDT values any. I'll have to look at doing more sampling and see if it makes a difference. First, I'll make a short undertaker script that just evaluates model1, so I can test changes. Tue Jan 2 10:52:11 PST 2007 Kevin Karplus So far, I have determined that the GDT numbes I'm getting correspond to what I get if I do no attempt to re-optimize transformations, but just take the n that I get from each sequential superposition of 3 residues. This is *not* what I want---there are definitely better superpositions out there. It *is* better than just taking a single transform based on all points. Sat Jan 20 07:07:10 PST 2007 Kevin Karplus I did eventually get GDT computation fixed---more directed sampling was the key. The CASP7 GDT scores are slight underestimates of true GDT. This model is about the only one in which our refinement improved a refinement target, but the refinement results on the CASP7 website seem to be junk---the RMSD and GDT values don't match my computations at all.