Tue Jul 18 09:49:06 PDT 2006 T0385 Make started Tue Jul 18 09:52:11 PDT 2006 Running on lopez.cse.ucsc.edu Tue Jul 18 14:55:27 PDT 2006 Kevin Karplus BLAST does not find strong hits. (best is 2almA or 1ox0A, 23% over 76 residues, E-value 0.085). We are getting moderate hits with HMMs (it will be a little while before all the HMMs are done). Tue Jul 18 16:37:22 PDT 2006 Kevin Karplus Top HMM hit 1vjxA at 2e-03 All the top hits seem to be to a.25.1.1, and the alignments are fairly long. Tue Jul 18 18:33:44 PDT 2006 Kevin Karplus The models from undertaker alignments all look good for a 4-helix bundle. I think that this is fairly simple fold recognition. Tue Jul 18 19:23:22 PDT 2006 Kevin Karplus We're going to have to add some constraints from the alignment models to keep the bundle together, because try1-opt2 let it fall apart when forming the loop from R76-S95. Making that loop will probably be the hardest part of the model. We may want to put constraints on V69, F129, A29, and A113, for example. Tue Jul 18 21:58:59 PDT 2006 Kevin Karplus I measured a bunch of distances from align1 (to 1vjxA). V32.CB W114.CB 8.1 A35.CB M106.CB 8.5 V43.CB A99.CB 5.8 V51.CB A96.CB 7.0 L54.CB V147.CB 7.1 L54.CB W144.CB 9.1 A58.CB M140.CB 5.8 V43.CB V55.CB 4.7 M72.CB F129.CB 6.9 M72.CB A130.CB 8.0 V69.CB A130.CB 8.5 A25.CB A116.CB 6.1 A25.CB V117.CB 4.7 L26.CB V117.CB 6.6 A29.CB A113.CB 6.2 A29.CB W114.CB 6.1 Tue Jul 18 22:20:30 PDT 2006 Kevin Karplus I put those distances in as constraints for try2.costfcn, together with the helices and sheets of try1-opt2. Started try2 on cheep. Wed Jul 19 09:36:03 PDT 2006 Kevin Karplus try2 got the helical bundle ok, but left a big gap on the linker between the first two helices and the last two. The helix A84-P90 is probably wrong, as it shortens the linker too much. Let's make a try3, putting a weak strand constraint on that region to keep it from forming a helix. I'll alos replace any of the helix packing constraints from align1 that were badly violated with new constraints from try2-opt2. Wed Jul 19 10:09:13 PDT 2006 Kevin Karplus try3 started on cheep. Wed Jul 19 16:47:56 PDT 2006 Kevin Karplus The biggest problem is the alignment of the helices with each other---they can slide up or down a turn of the helix, resulting in very similar structures. Which alignment is best? (try2 and try3 pick essentially the same one, because of the contraints in try3 being based on try2). The relationship between H62 and M106 seems to be the same throughout but there are two positions for M140 in the alignments and three for V32. So it isn't the middle turn that is so variable (despite the hinging in try1) but the 1st and 3rd turns. Looking at the n_sep, o_sep, n_notor2, and o_notor2 predictions, we can get the following Hbonds Hbond P48.O V51.N Hbond L73.O V78.N Hbond S74.O G77.N Hbond A83.O G86.N Hbond N145.O A150.N Hbond R146.O G149.N These don't really help with the critical turns, but may be useful anyway. Mon Jul 31 16:05:54 PDT 2006 Crissan Harris, Cynthia Hsu On a note, tries 2 and 3 are very similar, with try3 being the better scoring. Following Grant's suggestions, we went through the Undertaker alignments and picked out those that showed reasonably good packing between the hydrophobic CB atoms of the helices. We settled on model 2, 3, and 5 (2fkzA, 1tjoA, and 1za0A). For try4.under, we set it up so that it would read in alignments from these families. We considered copying the try1.costfcn to try4.costfcn, but decided that its foamy and skewed appearance made it not as preferable a choice as try3.costfcn. We put the following back into the costfunction: include T0385.dssp-ehl2.constraints include T0385.undertaker-align.sheets include rr.constraints We also removed all other constraints, so that the model could be generated from anew. try4 is currently running on abyss. Tue Aug 1 10:45:26 PDT 2006 Cynthia Hsu In try4, the helices became unpacked, completely. I decided it would better to include the constraints from try3 into the try5.costfcn. Hopefully this will achieve a much better result. try5 is currently running on lopez. Tue Aug 1 11:24:27 PDT 2006 Crissan Harris, Cynthia Hsu After examining the summary.html files, we've decided that given that V78 had a high I prediction in the ncep file and G77 had a high G prediction, there was a possibility of a capping motif over the end of the helix that did not exist in try3. Because of this, we copied try3.costfcn to try6.costfcn, then changed the helix constraint from P49 so that it spanned to M72 instead of A75. We also raised the weight to 1. We also added the following Hbond constraints> hbond V78.N L73.O 1 hbond G77.N S74.O 1 Examining the ehl2 script sseemed to indicate a sharp three-residue turn from residues 121-123, which in try3 were a helix. Because of this, we raised the weights on the helix constraints as follows: HelixConstraint A96 H120 1.5 HelixConstraint A124 R146 2 The n_sep also indicated that there was a possibility of a 5 separation from A150, so we added the following hbond constraint to accomodate for it: hbond A150.N N145.O 0.6 try6 is currently running on shaw. Tue Aug 1 14:00:31 PDT 2006 Crissan Harris, Cynthia Hsu try5 seems to have some closely packed residues, but in general is fairly foamy and still has exposed regions. try6 does not appear as well packed as try3, but the alignment of its helices is slightly different. Selecting the beta carbons of hydrophobic residues seem to indicate good packing, however. try6 did not form a hairpin from 121-123, as desired. Nor did it produce the helix cap motif. The primary difference between try3.costfcn and try6.costfn seems to be the coil from A150 to A155. We copied try6.costfcn to try7.costfcn. This time, we decided to remove the helix constraints from try2-opt2. We added the following constraint, to force the link: StrandConstraint A121 T123 0.5 We also changed the weights on the helix constraints, as follows: HelixConstraint A96 H120 2 HelixConstraint A124 R146 3 try7 is currently running on lopez. Tue Aug 1 15:55:38 PDT 2006 Crissan Harris, Cynthia Hsu For reasons that we can't explain, try7 still did not execute the desired effects. We raised the overall weight of "constraints" to 15, then raised the "StrandConstraint" from A121 to T123 to 1. We were concerned about the effects of an H bond constraint between two residues located in the same helix, so therefore, this Hbond constraint was removed: hbond A150.N N145.O 0.6 try8 is currently running on lopez. Wed Aug 2 10:39:52 PDT 2006 Cynthia Hsu The helices in this model do seem nicely packed, but the desired hairpin turn was not produced. Looking at the models in superimpose-best.under, the differences between try8 and try3 are marginal. In fact, try3 seems much less foamy than try8 does. For try9.costfcn, I raised the strand constraint on A121-T123 to 5. I also raised the HelixConstraint on A124-R146 to 8. try9 is currently running on orcas. Wed Aug 2 14:03:14 PDT 2006 Cynthia Hsu At this point, the desired results still have not been achieved. I've decided to remove all the distance constraints from the try10.costfcn and hope that this will have the desired effect. try10 is currently running on lopez. Wed Aug 2 14:06:30 PDT 2006 Grant Thiltgen The models from try3 and try5 are based primarily on 2fkzA which is the second fold recognition hit. I am going to start a new run removing the 2fkzA alignment and only using the other two top alignments 1vjxA and 1bcfA. Since undertaker likes to pull the helices apart, I am going to include the distance constraints to keep the helices together. Okay, try12 is going to use the two top alignments, and try11 is going to use the other alignments that Cynthia and Crissan liked. I'm not including 2fkzA in either of these two runs. I'm sending both of these to the cluster to run. Wed Aug 2 15:42:46 PDT 2006 Crissan Harris, Cynthia Hsu Without the distance constraints, try10 fell apart completely. The two helices are completely splayed at an approximately 140 degree angle from each other. Wed Aug 2 16:46:18 PDT 2006 Crissan Harris, Cynthia Hsu Because of the limited similarity between the two alignments that were chosen for try11 (1tjoA, and 1za0) and the actual sequence itself, try11 had helices that were not very well packed at all and several significant breaks. try12 had a much better result, with the exception of two significant breaks before T123 and Q188. Viewing it under the near script and selecting for the beta carbons of hydrophobic residues also produced decently satisfactory results. When examining our current favorite models in best-models.pdb.gz, we find that try12's helices are more tightly packed than those of try3, but that the alignment of the helices is not significantly different. We've decided to attempt to produce the results we were attempting with previous tries (the helical cap near residues 73-78) and the hairpin turn from 121-123. We removed the following helix constraint, as it was contradictory to another one: HelixConstraint D126 W151 1 Then we shortened the helix constraint from A97-D125 so that it would stop before H120. We raised the helix constraint on A124-R146 to 5, then added a StrandConstraint to A121-T123. Then we raised the overall weight of "constraints" to 15. For the helical cap, we added the following Hbond constraints: # to produce the capping motif from 73-74 hbond V78.N L73.O 1 hbond G77.N S74.O 1 try13 is currently running on abyss. Thu Aug 3 10:19:07 PDT 2006 Crissan Harris, Cynthia Hsu try13 appeared broken, most notably right before T123, the last residue in the hairpin turn. However, these did form a coil, as desired. We decided that this could be fixed, and the foaminess improved, by the following distance constraints: Constraint Y114.CB L134.CB -10 5 5.5 1 Constraint Y114.CB R115.CA -10 4.5 6 1 Constraint M106.CB M140.CB -10 8 10 1 We also raised "break" to 100 and "soft_clashes" to 30. try14 is currently running on lopez. Thu Aug 3 10:55:37 PDT 2006 Crissan Harris, Cynthia Hsu Starting from try3, which we felt to be our best model, we decided to attempt to shift the helices. We removed the helix constraints that were not from T0385.dssp-ehl2.constraints, and we changed the helix from S20 to S44 so that it would span to A45. The weight on this was raised to 2, as was the weight on the helix from P49 to A75. To promote the formation of the hairpin turn as indicated by L46 and P48, we added a strand constraint, with a weight of 5. We also commented out the helix constraints that would interfere with the shifting of this helix. On the helices in the vicinity of 121-123, we raised the weights up to 1.5, and added a strand constraint with a weight of 3. try15 is currently running on shaw. Thu Aug 3 12:12:13 PDT 2006 Grant Thiltgen I think we approached things a bit wrong yesterday when I was half asleep for aligning these models. I think we are going to rebuild from specific alignments but use the residue constraints from those specific alignments to make our model. For try16, I am using 1vjxA as a model. It appears that there were already constraints in place using 1vjxA as a model, but it may have used a different model for alignments. 2fkzA might have been the one. So I am rebuilding from 1vjxA alignments and constraints. I am then going to look at 2fkzA constraints and build a model from that. I'm hoping to get some variety, but a similar four helix bundle from most of these. Try16 started on whidbey. Thu Aug 3 12:23:08 PDT 2006 Cynthia Hsu As with Grant's suggestions, Crissan selected residue constraints for 1za0A and I selected residue constraints for 1tjoA. I copied try2.costfcn to try17.costfcn, then removed the distance constraints that were in existence. Examining 1tjoA in RasMol, I selected distance constraints for residues that were in close proximity, and gave each of them a weight of 0.3. The helix constraints that were not from the dssp-ehl2 constraints were also removed. try17.under included alignments only from 1tjoA. try17 is currently running on orcas. Thu Aug 3 12:27:07 PDT 2006 Crissan Harris As stated above, I added distant constraints from the 1za0A alignment with a weight of 0.3 as well as the helix constraints from the dssp-ehl2 constraints. I also added the hbond constraints that Professor Karplus had noted from the sep and notor predicitons with a weight of 0.2. try18.under was set up with 1za0A as it's only alignment. try18 is currently running on peep. Thu Aug 3 12:48:31 PDT 2006 Cynthia Hsu Following the procedures as before, I repeated the procedure for 2fkzA, using the same cost function as try17.costfcn assigning the distance constraints with a weight of 0.3. as before, try19.under used only alignments from 2fkzA. try19 is running on orcas. Thu Aug 3 15:53:49 PDT 2006 Crissan Harris, Cynthia Hsu try18 is broken, especially at residues 95, 77, and 106. Interestingly enough, 95 and 77 were both labeled as coil by the ehl2 script. We considered trying to seal these breaks, but considering how many kinked helices there were, and the fact that they were arranged so as to produce a hole in the vicinity of residues that should have been buried, we decided to discard this model. try19 also has breaks, but appears to have more densely packed helices than try18. It is slightly foamy, but in general, the alignments look good. In try17, the alignment seems good, and the packing seems tight, although it is still slightly foamy. One of the helices briefly became unwound to a coil from F129-A132, but other than this, there do not appear to be large problems with the model. try16 has two very large breaks, before D98 and V51. It may be possible to fix these by altering the constraints on strands and helices. It has an interesting and unique shape, and although is not as tightly packed as the others, could improve with some modifications and raised dry weights. try15 did not form either of the desired hairpin turns. It was also broken in many places, so we decided this model should be discarded as well. try14 looks quite nice, in terms of producing a turn (although not quite a hairpin) at residues 121-124. The turn at the end was also fairly tight. It also had no large breaks. We now have the following list of models that we like: ReadConformPDB T0385.try19-opt2.pdb ReadConformPDB T0385.try17-opt2.pdb ReadConformPDB T0385.try16-opt2.pdb ReadConformPDB T0385.try14-opt2.pdb ReadConformPDB T0385.try12-opt2.pdb ReadConformPDB T0385.try3-opt2.pdb #ReadConformPDB T0385.try8-opt2.pdb - much foamier than try3, otherwise virtually identical. ReadConformPDB T0385.try5-opt2.pdb #ReadConformPDB T0385.try1-opt2.pdb Thu Aug 3 16:23:42 PDT 2006 Crissan Harris, Cynthia Hsu try19 had huge breaks. The two largest breaks are before T123 and P49. Since both of these are places that are likely sites for hairpin turns, we've decided to modify the cost function to allow for these to form. The following strand constraints were added: # to produce hairpin turns StrandConstraint A121 T123 5 StrandConstraint L46 P48 5 We also raised "soft_clashes" to 30, "break" to 100, and lowered "sidechain" to 3. We also raised "dry8" to 20. Like try19, try20 used alignments from 2fkzA. try20 is currently running on orcas. For try17-opt2, we copied the try17.costfcn to try21.costfcn. Looking at the all.breaks.gz document, we noticed significantly large breaks near P49, A133, A85, and Q88. However, the ones at residues 85 and 88 were in coil regions, so we decided these would probably be closed simply by raising the weight on "break". For P49, we added the following strand constraint: StrandConstraint L46 P48 5 Because there was also a break near A133, which was a region in which the helix had unwound itself into a coil, we decided to change all distance constraints on A130 such that there upper bound was twice as it was, so that they would have more flexibility of movement. We also raised "dry8" to 20, "soft_clashes" to 30, "break" to 80, and lwoered "sidechain" to 3. Like try17, try21 used alignments from 1tjoA. try21 is currently running on camano. For try16, there was a very large break before D98, and another one (although not as bad) before V51. However, examining the location of the breaks, we've decided that it can be fixed by merely raising the weight on "break". "soft_clahses" for try22 was raised to 40, and "break" was raised to 120. We also lowered "hbond_geom_backbone" to 9, "hbond_geom_beta" to 20, and removed "hbond_geom_beta_pair". "sidechain" was lowered to 4. "dry8" was raised to 25. try22.under used the same family of alignments as try16 did (1vjxA). try22 is currently running on camano. Thu Aug 3 17:02:57 PDT 2006 Crissan Harris, Cynthia Hsu As said before, try14 was a good model as it had tight turns at the end of its helices and no large breaks. However, the coils on the end were rather loose and did not do well. After examining our unconstrained scores and discussing things with Grant, we decided instead to work on try15, despite the large break, and forgive the absence of our much desired hairpin turns. One thing that we noticed was that several exposed residues, namely A133 and L134, should have been buried, according to the near script, and we were concerned by the distance constraints that would hinder the freedom of movement. We then added the following distance constraint to try to get the helix to turn: Constraint W114.CB A138.CB -10 8 9.8 0.3 We raised "soft_clashes" to 40 and "break" to 110. We also raised "dry6.5" to 40 and "dry8" to 25. "sidechain" was lowered to 3 to give the helix a little bit more freedom to rotate. Like the previous try15, try23.under included all alignments. try23 is currently running on squawk. Thu Aug 3 18:10:22 PDT 2006 Cynthia Hsu try20 is a little bit foamy, but other than that, looks very very nice. However, it scores significantly worse than try19 does, particularly in areas of "dry5", "dry6, "phobic_fit" (which given the shape of the molecule may not be particularly important), "n_ca_c", "soft_clashes", and "hbond_geom_beta". I copied try20.costfcn to try24.costfcn. "dry5" was raised to 25 and "dry6.5" is raised to 35. I decided to raise "phobic_fit" slightly to 0.5. I also decided to return "sidechain" to its original weight of 5, as the helices don't seem to have much of a need to rotate anymore. "n_ca_c" was raised to 10 and "bad_peptide" to 12. I raised "soft_clashes" to 50 and "break" to 110. Because I liked the shape so much, I decided to set up try24.under as a polishing run of try20-opt2.pdb. try24 is currently running on camano. try21 ended up with several large breaks, most notably before A80, V51, G86, T132, and S47. However, in examining the unconstrained scores, it seems to do better than try17 in "break", although still with a very high cost, and it does worse in all other fields. After debating whether or not it was worth polishing, I opened grep-best-rosetta and noticed that try21 actually scored the best, by a significant margin, with try4 being the only exception (which was really bad). In view of this, I decided it might be a good idea to attempt to polish the gromacs repacked version. For try25.costfcn, I copied try21.costfcn and raised "dry5" to 25, "dry6" to 38, and "dry8.5" to 25. I raised "n_ca_c" to 7 and "bad_peptide" to 17. "breaK' rwas raised to 120 and "constraints" to 12. try25 used ReadConformPDB T0385.try21-opt2.gromacs0.repack-nonPC.pdb Since people seem stressed about meeting deadlines, try25 is currently running on bacchus, as I won't be needing it until tomorrow morning anyways. try22 was one of our worst scoring models, so future polishing runs are probably not worth pursuing. Fri Aug 4 09:25:56 PDT 2006 Cynthia Hsu try24 scored very well, much better than the try20 model that it was based on. Its scores can still be improved in "dry6.5", "dry8", "phobic_fit", "n_ca_c", "soft_clashes", and "break". However, it still has one very large break before P152, and I am not sure as to how that might be changed. try25 had poor "dry6.5", "dry8", "phobic_fit", "n_ca_c", and "bad_peptide" scores. For try26.costfcn, try25 was copied to try26. "dry6.5" was raised to 43, "dry8" to 30, "phobic_fit" to 0.5, "n_ca_c" to 15, "bad_peptide" to 23, and "soft_clashes" to 50. try26.under used ReadConformPDB T0385.try25-opt2. I considered using the gromacs-repacked version, as it was the highest scoring with Rosetta, but decided that probably using one gromacs model was enough of a starting point. try26 is currently running on orcas. Fri Aug 4 10:17:20 PDT 2006 Crissan Harris, Cynthia Hsu Reexamining try24 in RasMol, we decided to refrain from adding additional constraints, and to instead just attempt to have Undertaker try to close the break. "dry6.5" was raised to 40, "dry8" to 25, "sidechain" was lowered to 3, "n_ca_c" raised to 12, "soft_clashes" to 60, and "break" to 140. try27.under used ReadConformPDB T0385.try24-opt2.pdb try27 is currently running on shaw. Fri Aug 4 10:22:22 PDT 2006 Crissan Harris, Cynthia Hsu try23 had three very large breaks, before V104, S47, and T2. We've decided in this case to try to use the gromacs repacked version of the model, since it only had one large break before V104. In addition, it scores much better on "soft_clashes". "dry5" was raised to 30, "dry6.5" was raised to 43, "n_ca_c" to 8, "bad_peptide" to 20, "soft_clashes" to 60, and "break" to 140. "constraints" was lowered to 8 and "sidechain" to 2. try28.under used T0385.try23-opt2.repack-nonPC.pdb as the ReadConformPDB file. try28 is currently running on shaw. Fri Aug 4 11:21:18 PDT 2006 Crissan Harris Grant suggested using some of the constraints that were determined from the other alignments (1vjxA, 1tjoA, 1za0A, 2fkzA) that we used in tries 16-19 and then performing runs using all the alignments. Since try16 used the same constraints that were used in try3, but only aligned to the specific alginment 1vjxA, I've decided not to run this again since it would be virtually the same as try3. try29 uses the same costfcn as try17 (determined from 1tjoA). try29.under was copied from try3.under. try29 is currently running on shaw. Since try18 came out so badly (due to the fact that there were not enough residues aligned to give tight enough constraints) we decided not to use the costfcn for another try. try30 uses the same costfcn as try19 (determined from 2fkzA). try30.under was copied from try3.under. try30 is currently running on camano. Fri Aug 4 12:54:45 PDT 2006 Crissan Harris, Cynthia Hsu try27 scores among the best of our recent models, but it still appears rather foamy. try28 is less foamy than try27, but still has visible holes. try26 looks very foamy, when one considers how many polishing runs it has been through. However, it scores relatively well, and could potentially be improved by dry weights. We examined all of our polished models and found, to our dismay, that we had generated about six virtually identical models (only try5-opt2 looked significantly different). We decided to refrain from polishing and hope that try29 and try30 appeared differently. Fri Aug 4 14:29:38 PDT 2006 Cynthia Hsu Both try29 and try30 appeared very similar to each other and to try27. They also had extremely large breaks. However, try30 had a score comparable to our top models, so I thought it might be worth repeating the procedure, but with slightly different weights. try31.costfcn was copied from try30.costfcn. "dry8" was raised to "20", "soft_clashes" to 40, and "break" to 100. "n_ca_c" was raised to 7, and "bad_peptide" to 15. try31.under was identical to try30.under. try31 is currently running on lopez. Going through the best-models.pdb, I've made the following comments on our best models: ReadConformPDB T0385.try30-opt2.pdb //ReadConformPDB T0385.try29-opt2.pdb //too similar to try30, and low scoring //ReadConformPDB T0385.try28-opt2.pdb // similar to try23, which polished try14 - model IS unique from try27, slightly (long C-terminal helix). - BUT, it is identical to try3 and worse-scoring ReadConformPDB T0385.try27-opt2.pdb //polished try24, polished try20, polished try19 ReadConformPDB T0385.try26-opt2.pdb // polished try25, polished try21, polished try17 ReadConformPDB T0385.try16-opt2.pdb ReadConformPDB T0385.try12-opt2.pdb ReadConformPDB T0385.try3-opt2.pdb #ReadConformPDB T0385.try8-opt2.pdb - much foamier than try3, otherwise virtually identical. ReadConformPDB T0385.try5-opt2.pdb The commented out models were removed from superimpose-best.under. There are seven models in total. Fri Aug 4 16:48:16 PDT 2006 Cynthia Hsu I apparently had set up try31 as a new model from alignments, instead of as a polishing run. I copied try31.costfcn to try32.costfcn, then set up try32.under to ReadConformPDB T0385.try30-opt2.pdb try32 is currently running on camano. Sat Aug 5 02:34:50 PDT 2006 Kevin Karplus In reading through the README today, I was initially worried that people were playing with just the Helix constraints and not the distance constraints that affect the helix packing, but I see that got straightened out already. I've not had a chance to look at the models yet (download speed is only about 100kbs), but will try to today. Sat Aug 5 12:53:33 PDT 2006 Cynthia Hsu try32 scores the second best out of all models, but with a ridiculously high cost of "break" (4.7). I'm not sure what the best decision would be at this point. Sat Aug 5 13:59:17 PDT 2006 Grant Thiltgen So I'm going to run a few runs, attempting to shift the helices by starting from the original try3 model, but with different distance constraints to see if we can shift the helices up and down from where they started. I think try3 is good, because it's different, and the other models are all slightly different, but they helices look like they may be in the same general region with each other, though they are different than try3. I'm starting try33 which is starting from try3, but with shifted helices. I used the constraints from try29, that were the constraints from 1tjoA. try33 started on camano. I'm also starting try34 which is starting from try3, but with shifted helices with distance constraints to 2fkzA. Try34 started on whidbey. Try35 is going to be starting from all the alignments with the distance constraints from 1tjoA. Try36 is going to also start from all the alignments with distance constraints to 2fkzA. Try35 started on whidbey. Try36 started on lopez. I want to see how these work. I don't know which way will work better for shifting the helices, and I want to see how they look before I start new models from the other alignments. Sat Aug 5 16:19:02 PDT 2006 Grant Thiltgen It appears that starting from try3 and shifting the helices scores better in undertaker than starting anew from alignments and attempting to move things around. The best scoring model from the shifting of helices is try34, which uses 2fkzA as a model. I believe that try3 did 2fkzA for alignments, which is why try34 may score best overall. So far I have try32, try34, try36, and try6 are the most diverse so far. I am going to try some new runs starting from try3, but using different alignments to build the models from. Sat Aug 5 16:52:37 PDT 2006 Cynthia Hsu try32.costfcn was copied to try37.costfcn. The dry weights were left as is, but we raised "n_ca_C" to 10, "bad_peptide" to 20, "soft_clashes" to 70, "break" to 150, and "constraints" to 12. The dssp-ehl2 constraints were removed and replaced with those from try32-opt2.helices, which were each given a weight of 1. try37.under was a polishing run of try32-opt2. try37 is currently running on lopez. try38.costfcn was a copy of try34.costfcn. I raised "bad_peptide" to 15, "n_ca_c" was 7, "soft_clashes" to 40, and "break" to 100. try38.under polished try34-opt2. try38 is currently running on camano. try39.costfcn was a copy of try33.costfcn. "dry5" was raised to 23, "dry6.5" to 35, "dry8" to 20, and "dry12" to 7. "n_ca_c" was raised to 7, "bad_peptide" to 16, "soft_clashes" to 40, and "break" to 100. the dssp-ehl2 helices were replaced with those from try33-opt2, and given weights of 1. try39.under polished try33-opt2. try39 is currently running on camano. try40.costfcn was copied from try6.costfcn, with try6-opt2.helices each with a weight of 1 replacing the previous ehl2 constraints. "dry5" is raised to 25, "dry6.5" to 40, "dry8" to 20, "n_ca_c" to 7, "bad_peptide" to 12, "soft_clashes" to 50, and "break" to 160. "constraints" was raised to 12. try40 is currently running on orcas. Sat Aug 5 17:00:50 PDT 2006 Grant Thiltgen I am going to try a few more distance constraints from the alignments and starting from try3. Just to see if we get more variety. Try41 is using the try3-opt2 PDB file and using distance constraints from the alignment for 2gyqA. Try41 started on whidbey. Sat Aug 5 19:00:05 PDT 2006 Kevin Karplus Please make decisions about which models to submit soon. Sat Aug 5 21:31:12 PDT 2006 Kevin Karplus Pick for variety, but it looks like try37, try38, try1, try25-opt21.gromacs0.repack-nonPC may be good choices based on scores. Sat Aug 5 21:49:57 PDT 2006 Grant Thiltgen Okay, so looking at the models which we ran, I would say try37 and try38 are probably are best two choices. Try1 scores well, but the bundle is not held together, so I really don't like that model. Try41 also scores decently, but I'm not sure how it looks in comparison to the other models we tried. I'm going to look at that now. Hmm. I've decided on the first four: try37-opt2 try38-opt2 try41-opt2 try25-opt2.gromacs0.repack-nonPC I still don't know about the fifth model yet. I'm not even sure I like try41, it seems very similar to try37 So now I have these: try37-opt2 try38-opt2 try25-opt2.gromacs0.repack-nonPC Okay my final decision on models is try37-opt2 try38-opt2 try25-opt2.gromacs0.repack-nonPC try39-opt2 try11-opt2 ---------- For this model we got many hits to four helix bundles with hits to SCOP a.25.1.1. All of our alignments showed the four helix bundle in differing positions, so we attempted to align our helices up slightly different for each of our submissions. Our best results appeared to stem from alignments to 2fkzA and 1tjoA. Model 1 is try39-opt2, which was our best scoring model with our unconstrained costfcn for undertaker. This model was a refined model based on alignments to 2fkzA and using residue constraints to 2fkzA to keep the four helix bundle in place. Model 2 is try38-opt2, which was our second best scoring model with our unconstrained costfcn. This model was a polished model with residue constraints to 2fkzA, but started with another model that we had originally aligned to 1vjxA. For this, we attempted to move the helices in an already established model instead of starting from original alignments, which ended up giving us a different bundling of the helices. Model 3 is try25-opt2.gromacs0.repack-nonPC, which is a model we created which scores very well with the rosetta scoring function. It started from alignments to 1tjoA, and it included constraints to keep the four helix bundle together using residues from this alignment. This model is a polished model from the original model run with these alignments. Model 4 is try39-opt2, which also started with the model we had originally aligned to 1vjxA, but then we shifted using the constraints from the alignments to 1tjoA. This model was also created from a full model instead of alignments, and is included because it gives variety from the previous models. Model 5 is try11-opt2, which is a model started from alignments to 1tjoA and 1za0 and using distance constraints to alignments from 1vjxA in order to pack the helices. The model doesn't score that well, but is included because it differs from the other models. -------- Interesting, models that aligned to certain pdb files seemed to score better than others, but aligning them in different ways caused the helices to pack very differently. Models submitted August 5th, 10:28 pm.