Wed Jul 5 10:51:02 PDT 2006 T0358 Make started Wed Jul 5 10:52:47 PDT 2006 Running on shaw.cse.ucsc.edu Wed Jul 5 11:30:27 PDT 2006 Kevin Karplus BLAST gets no good hits in PDB. Best is 2eql 48% over 23 residues E-value 0.75 Only 87 residues long, so we may be able to use new-fold techniques, if the HMMs don't fnd anything. Wed Jul 5 19:06:15 PDT 2006 Kevin Karplus Very weak hit with HMMs to 2f4qA (E-value 0.3). Wed Jul 5 19:54:05 PDT 2006 Kevin Karplus try1 failed, probably because of the buggy version of undertaker from tis morning. (Nope---the bug is an older one, trying to force alignment with zero fragments.) Wed Jul 5 22:21:43 PDT 2006 Kevin Karplus try1 finished, but the results are not impressive---this one will need a lot of thought. Fri Jul 14 12:40:03 PDT 2006 Crissan Harris, Cynthia Hsu I just wanted to throw this out to whomever gets assigned to this protein eventually - it looks A LOT like T0348. Mon Jul 17 15:43:54 PDT 2006 Crissan Harris, Cynthia Hsu Professor Karplus agreed with our opinion on this, so I've transferred our top models of T0348 into the superimpose-best.under file for the sake of comparison. However, RasMol does not seem to say much in best-models.pdb.gz. The first alignment seems to match reasonably well. Examining try1 under the conserved_t06 script, I found that the exposed residues of the sheets did seem to be part of an active site. Due to the absence of Cys residues, "maybe_ssbond" was set to 0. There appears to be a Histidine tag, but it is difficult to identify whether it is supposed to form a metal bonding site with H36. Under Prof. Karplus's advice, we looked up the structure of 2f4q and found that it actually was not a dimer, which does not explain the flat hydrophobic face. Also, the near scripts and the conserved_t06 scripts did not indicate any similarities from which the arrangement of the helices could be drawn. Looking at the rr.constraints, we made the following additions: Constraint F37.CB V40.CB -10. 7.0 14.0 0.5 //hopefully to bring the exposed regions closer together Constraint L63.CB V41.CB -10. 7.0 14.0 1.5 //Will bring the exposed region of the helix closer to the beta sheets. Constraint I67.CB I27.CB -10. 7.0 14.0 1.5 try2 started on pyro on Mon Jul 17 17:14:32 PDT 2006 Tue Jul 18 10:21:15 PDT 2006 Crissan Harris, Cynthia Hsu In general, we liked the shape that came out of this model. It is still a bit too loose and exposed. The beta sheets also looked a little bit warped, in view of this, we changed the constraint on this area to apply to R42 and F37, instead of V40 and F37. We also raised the weight on the other two RR constraints (L63-V41 and I67-I27) to 4, hoping that this would bring them even closer together. "phobic_fit" was raised to 3 as well for this purpose. try3 is currently running on orcas. Tue Jul 18 11:37:46 PDT 2006 Crissan Harris, Cynthia Hsu try3 actually had a very interesting shape, with the N-terminus helix (1-24) actually ended up on the other side of the beta sheets from the other helices. The residues identified by the RR script as belonging close together, L63, V41, I67, and I27 were packed fairly closely together. Also, the target now does fairly well under both the burial and the near scripts. In try4.costfcn, we added the following constraint so that the helix from M1 to F12 might shift so that it is exposed, and a region colored brown will be brought more towards the interior of the protein: Constraint L7.CB Q16.CB -10. 7.0 14.0 2 To bury another region labeled brown by the burial script, we decided to bring together two residues labeled by the RR script, as follows: Constraint W53.CB E81.CB -10. 7.0 14.0 3 Since everything is also starting to pack itself really tightly together, we've decided to raise "soft_clashes" to 30. try4 is currently running on orcas. We were also curious as to whether the novel placement of the helices was due to the phobic fit or the high weight on the RR constraints. Because of this, we decided to copy try3.costfcn, with "phobic_fit" set to 2, the default value. try5 is currently running on furi. Tue Jul 18 13:10:26 PDT 2006 Crissan Harris, Cynthia Hsu We ran make2 as Prof Karplus suggested in his email. Tue Jul 18 13:31:49 PDT 2006 Crissan Harris, Cynthia Hsu We did not like the appearance of try5 is RasMol, and try4 was even worse. However, try5 was our second highest scoring model (after try2) and try4 still scored higher than the try3 model that it was based on. Rosetta actually likes try4 the best, followed by try3, then try5. We copied try4.costfcn to try6.costfcn, then modified the rr constraint functions as follows: Constraint F37.CB R42.CB -5. 7.0 10.0 0.5 //hopefully to bring the exposed regions on beta sheet closer together Constraint L63.CB V41.CB -5. 7.0 10.0 4 //Will bring the exposed region of the helix closer to the beta sheets. Constraint I67.CB I27.CB -5. 7.0 10.0 4 Constraint L7.CB Q16.CB -5. 7.0 10.0 2 Constraint W53.CB E81.CB -5. 7.0 10.0 3 We also commented out the dssp-ehl2, undertaker alignments, and rr constraints. try6 is currently running on orcas. We also decided that given that we liked the shape of try3 the most, and it was the second most favored by Rosetta, we should attempt to polish it some more.\ For try7.costfcn, we copied the try3.costfcn, raised "dry5" to 20, "dry6.5" to 25, "dry8" to 20, "n_ca_c" to 7, "bad_peptide" to 12, "soft_clashes" to 50, and "break" to 120. We also commented out the dssp-ehl2, undertaker alignment, and rr constraints. try7.under took T0358.try3-opt2.gromacs0.repack-nonPC.pdb.gz as input. try7 is currently running on camano. Tue Jul 18 15:26:45 PDT 2006 Crissan Harris, Cynthia Hsu try6 looked very twisted, and had several extremely large breaks. try7 was okay. It scored better than try3, but only slightly, and Rosetta still favored try3 over try7 (also with a minor difference). Due to the failure of try6, I decided that it would probably be in our best interest to just polish the try4 model as is. "dry5" was raised to 20, "dry6.5" was raised to 25, and "dry8" was raised to 20. "n_ca_c" was raised to 7 and "bad_peptide" to 12. "soft_clashes" was set to 60 and "break" was set to 120. try8.under used T0358.try4-opt2.gromacs0.repack-nonPC.pdb.gz as the PDB input. try8 is currently running on camano. Thu Jul 20 11:24:02 PDT 2006 Crissan Harris, Cynthia Hsu try8 does not score as well as the others with Undertaker, and Rosetta still favors try4. Our top five scoring models did not change. Examining our best models, we found that we did not like the way try5 distorted a helix indciated to be such by the ehl2 script. Because of this, we copied the try5.costfcn to try9.costfcn, commented out the dss-ehl2 constraints, the undertaker alignments, and the rr constraints. We then added the following helix constraint: HelixConstraint (T0358)F55 (T0358)S70 2 We also increased the overall constraint to 12. try9 is currently running on camano. Thu Jul 20 13:12:20 PDT 2006 Crissan Harris, Cynthia Hsu try9 produced the interesting result of turning the C-terminal helix into a sheet, and moving the second strand from the beta sheet outside, to form a helix. While it forms a diverse and interesting model, residues 77-87 (in this model, a sheet), were indicated to be exposed by the burial and near scripts. Therefore, we decided that try9 is probably not worth considering. However, we still wanted to try to modify try5, so we copied try9.costfcn to try10.costfcn and did the following: We left the dssp-ehl2 constraints in, then raised the "phobic_fit" to 3. try10 is currently running on squawk. Thu Jul 20 15:18:08 PDT 2006 Cynthia Hsu try10 had the same appearance as try9. It scores better, but only slightly. No changes were made to the superimpose-best.under file. Fri Jul 21 13:10:06 PDT 2006 Cynthia Hsu I made the score-all+servers.unconstrained file and found that try7 does perform worse than the top server alignment (TS1). To accomodate for this, I decided to raise "bad_peptide" to 15, "dry5" to 25, and "dry6.5" to 30. Hopefully this will adjust for it. try11.under was set up as a polishing run that used try7-opt2 as the input PDB. try11 is currently running on abyss. Fri Jul 21 15:23:19 PDT 2006 Crissan Harris, Cynthia Hsu try11 scored better than the try7 model that had similar constraints. Fri Jul 21 17:47:26 PDT 2006 Crissan Harris, Cynthia Hsu The make on the subdomain is currently running on abyss. Make started Fri Jul 21 18:17:24 PDT 2006 Running on whidbey.cse.ucsc.edu Sat Jul 22 13:31:45 PDT 2006 Cynthia Hsu Because try11 scored relatively well, I decided, for try12, to incorporate some suggestions inspired by Prof. Karplus during the meeting. try12.costfcn was copied from try11.costfcn. The following helix constraint was added: HelixConstraint (T0358)T2 (T0358)Y24 2 Hopefully this will fix the kink in the helix that is causing the N-terminus, which should be exposed, to tuck itself into the sheets. It will also align the regions that should be buried, which hopefully will make packing the helix easier. I also decreased the "sidechain" weight to 3, hoping that this might possibly give the helices a little bit more freedom of movement. try12 is currently running on camano. For try13, I copied try5.costfcn. The sheets looked a little bit twisted, so I decided to comment out the undertaker-alignments and hope for the best. I added the following helix constraint: HelixConstraint (T0358)T2 (T0358)T23 2// hopefully will fix the kink in this helix, and free the N terminus I also added the following distance constraints, to bring the exposed helices closer to the beta sheet: constraint((T0358)L39.CA, (T0358)L63.CA) [> 3 = 7 < 10] w=2 //lowering the distance constraint between these two residues will hopefully bring the alpha sheet in closer contact with the beta sheets. constraint((T0358)A64.CA, (T0358)L80.CA) [> 7 = 9 < 12] w=1 // Will keep the C-terminal helices closer together. try13 is currently running on shaw. Sat Jul 22 14:32:14 PDT 2006 Cynthia Hsu try12 turned the C-terminal helix into a beta sheet. In view of this, for try14, I returned "sidechain" to its original weight of 5. I lowered the HelixConstraint on T2-Y24 to 1.5, then copied the other Helix constraints from try11-opt2.helices and gave each a weight of 1. Hopefully this will achieve the desired effect, without affecting it too much. I also added the following distance constraint, to try to draw the helix closer to the sheet slightly: constraint((T0358)F12.CA, (T0358)A52.CA) [> 5 = 10 < 17] w=1 try14 is currently running on camano. Sat Jul 22 14:52:01 PDT 2006 Cynthia Hsu try13 failed to have any of the desired effects, only making the helix more splayed. I lowered the "phobic_fit" to 1, hoping that this would decrease the radius of gyration. I raised the weight on T2-T23 to 3 and added a new constraint: constraint((T0358)V20A.CA, (T0358)W50A.CA) [> 3 = 9 < 13] w=5 I also corrected one of the residues that were inappropriately labeled in the previous cost function. try15 is currently running on shaw. Sat Jul 22 15:31:00 PDT 2006 Cynthia Hsu try14 still failed to correct the sheet back into a helix. I've decided to refrain from this line of reasoning, as it doesn't seem to be getting me anywhere. Sat Jul 22 16:03:26 PDT 2006 Cynthia Hsu The subdomain (M1-R79) finally finished. The appearance is completely different from previously generated models, but it is much more foamy and has many exposed regions. In an effort to draw the C terminal helix closer to the exposed beta sheets, I added the following constraint: Constraint V41.CB I67.CB -10. 7.0 14.0 3 Due to the presenece of only one His residue in this subdomain and the complete absence of Cys residues, "maybe_metal" and "maybe_ssbond" are both set to 0. I also copied the helix constraints from try1-opt2.helices, giving each a weight of 1, except for the C terminal helix, which I gave a weight of 2 and altered so that it would start from G61 instead of residue 63. I also commented out the undertaker alignments. try2 is currently running on lopez. Sat Jul 22 16:43:40 PDT 2006 Cynthia Hsu Like many recent models, try15 turned the N-terminus into a sheet instead of a helix. One of my solutions to this was to raise the rr constraint between I67 and I27 to 6, since the molecules are still very far away in this model. I lowered the "sidechain" weight to 3, hoping that this would allow the helix form and rearrange itself as was necessary. I decided to set "maybe_metal" to 0.1, given that we are assuming the excessive His residues are due to a His tag. I also copied try11-opt2.sheets and gave all of the sheet constraints a weight of 8, hoping that this would force the sheets to remain as they are. As for theN-terminal helix, I also raised this to 8. try16 is currently running on orcas. Sat Jul 22 17:22:19 PDT 2006 Cynthia Hsu try2 for the subdomain turned out quite nicely, but it still looks very foamy. However, given the time, I don't think I will have the opportunity to optimize it anymore than it already is. Looking at all of our best models, I decided to superimpose the subdomain onto the one with the protruding His tag, since the His tag should be exposed. The best choice for this was try5-opt2. For try17, I copied try1.costfcn, but turned off "maybe_metal" and "ss_bond", and raised "soft_clashes" to 50 and "break" to 120. "dry5" was set to 20 and "dry6.4 was set to 25, so that the polishing could be improved as well during optimization. I also commented out the dssp-ehl2 constraints, the undertaker alignments, and rr constraints. try17 is currently running on orcas. Sat Jul 22 17:55:27 PDT 2006 Cynthia Hsu try16 produced an interesting fold. It could still be more tightly packed, however, and some residues are exposed that shouldn't be, so I've decided to try one more model using its constraints before preliminary submission. In view of this, I added the following residue constraint: Constraint A52.CB A59.CB -10. 7.0 14.0 8 Hopefully this will use the exposed Alanines to pack against each other. Other than that, I raised "dry8" and "dry6.5" by 5, each. try18 is currently running on shaw. Sat Jul 22 18:13:51 PDT 2006 Cynthia Hsu While that's being processed, I've decided to go through and chronicle the best models, as follows: ReadConformPDB T0358.try11-opt2.pdb //scores very well on phobic fit and much better than try13 on "n_ca_c", "soft_clashes", and "break". ReadConformPDB T0358.try13-opt2.pdb //scores better than try11-opt2, but the helix looks rather splayed, and in general is rather loosely packed and foamy. ReadConformPDB T0358.try2-opt2.pdb //scores the best. ReadConformPDB T0358.try5-opt2.pdb //scores the second best, but doesn't look very good ReadConformPDB T0358.try16-opt2.gromacs0.repack-nonPC.pdb //Rosetta's favorite, but scores really low with Undertaker. Still appears nice, but is rather foamy and has exposed residues. ReadConformPDB T0358.try1-opt2.pdb //scores relatively high, but helices on either side of T172 stick out a little too much, when this particular residue ought to be relatively buried. ReadConformPDB T0358.try4-opt2.gromacs0.repack-nonPC.pdb //Rosetta's second favorite, and scores relatively high with Undertaker, but the helices are not packed nicely against the beta barrel as one would prefer. With this in mind, I think our favorite models are as follows: try2-opt2 This was the highest scoring of our models. It was a simple Undertaker optimization of the alignments with an emphasis on particular rr constraints in order to bring exposed residues of the target together. T0358.try16-opt2.gromacs0.repack-nonPC.pdb This was the favorite of Rosetta's. The overall shape of the protein looks nice, but it suffers from exposed residues and extreme foaminess. It was meant to correct the formation of a sheet by the N-terminal helix from try13, which had been created to deformities in try5 that lead to some residues being exposed that shouldn't have been. It also utilized distance constraints to bring the helices closer to the sheet, instead of rr constraints. try5-opt2 This was another model greatly favored according to the unconstrained scores, in which different weights on different cost functions were experimented with. In this particular model, high weights were added to the rr constraints to pack the molecule closer together. try11-opt2 try11 is a relatively low scoring model, but it performs well in areas of "phobic_fit", "soft_clashes", "n_ca_c", and "break". We also liked the overall, relatively compact appearance of this molecule, even though it makes an effort to bury the His tag, and it has a slightly different alignment of its three strand beta sheet than the other models. This was a polishing model of try7, which had been generated for purposes of polishing the try3-opt2 gromacs repacked version model. However, that particular model is still more favored by Rosetta than this one. try4-opt2 Since I am reasonably certain that our chimera will do better, I am refraining from commenting on this one at the moment. Sat Jul 22 18:52:58 PDT 2006 Cynthia Hsu try17 just finished running. It does not seem as well packed as I had hoped, and many of its helices developed kinks. It also scored worse than try11 did. It seems that the Undertaker optimization made it worse, instead of better. Regardless, chimera5-2.pdb still does very well, as the second best scoring model. Therefore, I think it should be added to the list with the following description: This model was generated by manipulating the costfunction from an Undertaker alignment of the first seventy nine residues, excluding the LEHHHHHH tag at the end. In particular, RR constraints were added to bring the helices closer together and bury exposed residues. After this was done, a chimera was formed between this subdomain and the His tag of another model, try5, which was chosen because its particular His tag was more exposed than those of other top scoring models. try18 actually scored worse than try11. Although it did slightly better than try16, try16 is still the top-favored Rosetta model. In view of this, our selection for the top five models is as follows: ReadConformPDB T0358.try2-opt2.pdb ReadConformPDB chimera5-2.pdb ReadConformPDB T0358.try16-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0358.try5-opt2.pdb ReadConformPDB T0358.try11-opt2.pdb best-models has been updated accordingly. Sat Jul 22 19:26:06 PDT 2006 Kevin Karplus I modified superimpose-best.under to focus on residues H36, R42, W50, and R51 when doing the superposition, so that the sheets superimpose more clearly and I can see the differences between the models. The sheets are almost identical, but the helices pack quite differently in the different models. All the models are rather foamy (too much emphasis on avoiding clashes, and not enough on packing things tightly). I actually think that models 2 (chimera5-2) and 5 (try11-opt2) may be the most convincing (using "near" to color the atoms and spacefill to view them). Since this is just a preliminary submission, and a more tightly packed submission can be made later, if one is found, I'll stick with Cynthia's listing. I really do not like chimera5-2, as makes the HIS tag clash terribly with the sheet. It would have been better to make a chimera with the HIS tag somewhere else entirely, and let undertaker try to hook it up. I think there is still an advantage to using M1-R79/try2-opt2, but it will need to have a HIS tag put on it *legally* (without gross clashes). Unfortuantely, all 5 of the HIS tags in best-models.pdb.gz conflict with model 2 somewhere, so a simple cut-and-paste among these 5 models won't do. I'll sumbit what Cynthia left, but I think that getting a HIS tag onto the M1-R79/try2-opt2 model may be the highest priority after the preliminary submission. Sat Jul 22 20:00:46 PDT 2006 Kevin Karplus Preliminary submission done, with comment At least 18 undertaker runs were done for T0358, which was treated as a new fold target. All the submitted models share a 3-strand meander, but they pack helices around the sheet in different ways. None of our models seem particularly densely packed, so they are probably not right yet. For the preliminary submission, we are submitting Model 1 is T0358.try2-opt2, our top-scoring model with an unconstrained costfcn. It was a simple Undertaker optimization of the alignments with an emphasis on particular rr constraints in order to bring exposed residues of the target together. Model 2 is chimera5-2, a model optimized on just M1-R79, then with the HIS tag tacked on. Unfortunately, the HIS tag does not fit well here, and causes a serious clash. We hope to replace this model with one that has a similar structure but attaches the HIS tag more reasonably. Model 3 is T0358.try16-opt2.gromacs0.repack-nonPC, the model that rosetta likes best of the ones for which it repacked sidechains. The overall shape of the protein looks nice, but it suffers from exposed residues and extreme foaminess. It was meant to correct the formation of a sheet by the N-terminal helix from try13, which had been created to deformities in try5 that lead to some residues being exposed that shouldn't have been. It also utilized distance constraints to bring the helices closer to the sheet, instead of rr constraints. Model 4 is T0358.try5-opt2, another model greatly favored according to the unconstrained scores, in which different weights on different cost functions were experimented with. In this particular model, high weights were added to the rr constraints to pack the molecule closer together. Unfortunately, this is not a very good way to get tighter packing, and probably contributed to the foaminess of the model. Model 5 is T0358.try11-opt2, a relatively low scoring model, but it performs well in areas of "phobic_fit", "soft_clashes", "n_ca_c", and "break". We also liked the overall, relatively compact appearance of this molecule, even though it makes an effort to bury the His tag, and it has a slightly different alignment of its three-strand beta sheet than the other models. This was a polishing model of try7, which had been generated for purposes of polishing the try3-opt2 gromacs repacked version model. However, that particular model is still more favored by Rosetta than this one. ------------------------------------------------------------ Mon Jul 24 10:57:05 PDT 2006 Cynthia Hsu I copied the try17.costfcn, and lowered "sidehain" to 2 so that the helices would have a bit more freedom of rotation. I also returned "dry6.5" and "dry5" to their original values of 20 and 15, respectively, then, to fix the break in the helix, added a helix constraint from T74 to H87 with a weight of 5. Like try17, try19 used chimera5-2.pdb in the ReadConform. try19 is currently running on orcas. Mon Jul 24 11:30:38 PDT 2006 Cynthia Hsu Except for a very large break around residue 52, I found the overall apperance of try18 rather pleasing. Because of this, I raised "soft_clashes" to 40 and "break" to 100. "phobic_fit" was also lowered to 1. I also commented out the dssp-ehl2 constraints, and added the following helices: HelixConstraint (T0358)G62 (T0358)S70 5 HelixConstraint (T0358)R73 (T0358)H87 8 try20 is currently running on whidbey. Mon Jul 24 13:33:22 PDT 2006 Cynthia Hsu try19, despite avoiding severe clashes, moved the helices into an undesirable position. I think it may be necessary to consider Prof. Karplus's advice on remaking the chimera with a His tag completely out of the way. With this in mind, I decided to attempt another chimera, this time using the His tag from try2-opt2 for residues 80-87. try21.costfcn was identical to try17.costfcn, and try21.under was also the same except for using chimera2-2.pdb as the input file. try21 is currently running on shaw. For try20, setting "phobic_fit" to 1 may have been a bad idea. try22 was identical to try20, excpet for the fact that "phobic_fit" was raised to a value of 3. try22 is currently running on shaw. Mon Jul 24 15:13:31 PDT 2006 Crissan Harris, Cynthia Hsu try22 was extremely foamy. The helices were too far removed from the beta sheet. Mon Jul 24 16:04:57 PDT 2006 Crissan Harris, Cynthia Hsu In try21, the clash with the His tag was removed, but at the expense of two large breaks around residues 34 and 7. In view of this, we've decided to attempt to close the breaks by modifying the weights of the cost function. We rasied "break" to 150 and "soft_clashes" to 70. We also added all of the sheet constraints from try21-opt2.sheets, giving them each a weight of 2. Then we copied over the two C-terminal helices from try21-opt2.helices. To close the break at around L7, we made a HelixConstraint spanning M1 to T21, with a weight of 4. try23 is currently running on lopez. Tue Jul 25 10:12:37 PDT 2006 Crissan Harris, Cynthia Hsu try23 did not perform as we anticipated. Although it closed the breaks, adding the helix constraint may not have been a good idea, as it forced the helices away from the beta sheets. The first thing that we wanted to try to fix this was to raise the RR constraints to bring the helix from A59-A77 closer to the beta barrel. Constraint F55.CB Y66.CB 0. 5 8.5 4 Constraint W50.CB L63.CB 0. 5 9 4 We also changed the C-terminal helix constraint to span from D75 to R79. Hopefully, this will allow it to form a kink if necessary to move the His tag away from the beta sheet. As before, we did not do a polishing run of this model. try24 is currently running on lopez. We also decided that we needed to refine the subdomain so that the C terminus was pointed more in the opposite direction, as it is currently oriented in the direction of the sheets, despite the fact that the script indicates it should be exposed. Exmaining this under the RR script, we decided to bring the following residues closer to F37: I72, Y66, and I67. We copied try2.costfcn, and found that an rr constraint was inappropriately labeled between V41 and I67. This was removed. We also removed "include rr.constraints", given that there were indications of rr constraints on either side of the beta sheets which probably would confuse Undertaker. We also lowered "sidechain" to 4 to give the target a little bit more flexibility of movement. try3 is currently running on shaw. Tue Jul 25 12:03:32 PDT 2006 Crissan Harris, Cynthia Hsu The addition of the rr constraints had the desired effect on try3 of the subdomain, but the C-terminal helix where the His tag would have attached still turned itself inward towards the helix. In view of this, we decided to add distance constraints to turn the helix in the opposite direction. M1-R79/try4.costfcn was the same as try3.costfcn, except for the following added distance constraint: constraint((T0358)F37.CA, (T0358)R79.CA) [> 9 = 10.5 < 16] w=2.5 try4 is currently running on orcas. try24 had the completely reverse effect as was intended; the helices drifted out into the environment exposing the beta sheets. However, we are considering using it for the addition of the His tag, as this may give Undertaker the freedom to work with it as it will. Tue Jul 25 13:16:13 PDT 2006 Kevin Karplus The constraint in M1-R79/try4.costfcn is illegally formatted. What was *meant* was probably constraint F37.CA R79.CA 9 10.5 16 2.5 In M1-R79/try4-opt2 these two atoms are 11.77 Angstroms apart, and bringing them closer is *not* what I believe was intended. Instead, I believe you wanted it 10-20 Angstroms farther than it is now: constraint F37.CA R79.CA 20 25 35 2.5 Tue Jul 25 13:30:16 PDT 2006 Crissan Harris, Cynthia Hsu In accordance with the suggestion of Prof. Karplus, we added the following constraint to try5.costfcn to replace the inappropriately formatted one in try4: constraint F37.CA R79.CA 20 25 35 2.5 M1-R79/try5 is currently running on orcas. Tue Jul 25 14:39:54 PDT 2006 Kevin Karplus M1-R79/try5 did not satisfy the constraint: constraint((T0358)F37.CA, (T0358)R79.CA) [> 20 = 25 < 35] w=2.5 # (weighted) cost= 3.692 dist= 8.23 despite starting over from alignments. Let's try again with this constraint cranked up high. Also, let's fix the typo in the helix constraint (E61, not G61), and turn down the overly-strong rr constraints. Tue Jul 25 14:50:19 PDT 2006 Kevin Karplus M1-R79/try6 started on lopez. Tue Jul 25 15:42:40 PDT 2006 Crissan Harris, Cynthia Hsu In try6, the helices were very far removed from the beta sheets; when examining the subdomain in spacefill, there was a large hole that became apparent in the middle of the protien. In view of this, we decided that try5 might be preferred over try6. The unconstrained scores for the subdomain agree with this judgement, and list try5 as having the lowest cost by 2.02, followed by try3 and try4. Given the time, we have decided that we sould attempt to attach the His tag to try5, and optimize it with Undertaker. As previously noted, we decided to use the His tag from try24-opt2, as this was the most distantly removed from the rest of the target. Hopefully this will enable Undertaker enough flexibility to manipulate the His tag as it best sees fit, thus avoiding any clashes with the rest of the target. This script was run by make-chimera3.under. try25.costfcn was a copy of try1.costfcn, but with "maybe_metal" set to 0, "maybe_ssbond" set to 0, and "break" set to 100, and "soft_clashes" set to 40. try25.under used chimera5-24.pdb as its input file. try 25 is currently running on shaw. Tue Jul 25 19:07:53 PDT 2006 Cynthia Hsu try25 placed the helix in the appropriate place, and there were no clashes present, but I'm not satisfied with the overall appearance in general. It is very foamy, and there are a lot of exposed residues that shouldn't be. However, it scores very well, particularly in "sidechain" and "phobic_fit". I raised "dry5" to 25, "dry6.5" to 30, "n_ca_c" to 7, and "bad_peptide" to 12. I also raised "soft_clashes" to 50 and "break" to 100. try26.under was a polishing run of try25-opt2. try26 is currently running on lopez. Examining our top-scoring models, we have these options: ReadConformPDB T0358.try5-opt2.pdb #ReadConformPDB T0358.try2-opt2.pdb //Most poorly packed. ReadConformPDB T0358.try25-opt2.pdb #ReadConformPDB T0358.try19-opt2.pdb //Too similar to try20 ReadConformPDB T0358.try23-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0358.try20-opt2.pdb ReadConformPDB T0358.try11-opt2.pdb //Could be improved by polishing. #ReadConformPDB T0358.try1-opt2.pdb //Is the initial model, and not terribly different fron the others. After looking through all of our top models, I've decided that our top choices are try25, try23, try5, try11, and try20 I observed from going through these models that try11 looked rather foamy. Surprisingly, tehre were no polishing runs of try11, so in view of this, I copied try11.costfcn to try27.costfcn. I raised "dry5" to 30, "dry6.5" to 36, "bad_peptide" to 20, and "constraints" to 15. try27 is currently running on orcas. Tue Jul 25 20:13:40 PDT 2006 Cynthia Hsu I decided to try one last chimera, out of the fourth subdomain, before giving up. try28 took chimera4-24 as the input PDB. I copied try1.costfcn to try28.costfcn, and commented out the dssp-ehl2 constraints, the alignments, the rr constraints, and turned off maybe_metal and maybe_ssbonds. "break" was set to 100 and "soft_clashes" were set to 40. I raised "dry5" to 20 and "dry6.5" to 30, as well. try28 is currently running on camano. Tue Jul 25 20:30:32 PDT 2006 Cynthia Hsu As a summary of our models: Model 1 is T0358.try26-opt2, a polished Undertaker optimization of a chimera. The chimera was made by using Undertaker alignments to identify an appropriate structure for the subdomain composed of the first seventy-nine residues, after which the remaining residues, forming the LEHHHHHH tag, were added on. The subdomain was optimized by using distance constraints to bring the helices closer to the sheet, as well as changing the weights of various properties in the cost function. Model 2 is T0358.try5-opt2, another model greatly favored according to the unconstrained scores, in which different weights on different cost functions were experimented with. In this particular model, high weights were added to the rr constraints to pack the molecule closer together. Unfortunately, this is not a very good way to get tighter packing, and probably contributed to the foaminess of the model Model 3 is T0358.try27-opt2, a polishing run of an earlier model. It has a relatively compact appearance, although it tried inappropriately to bury the His tag. This was a fairly polished model of several attempts to optimize the original alignment by modifying the weights on the cost function. Residue-residue contact constraints were used to manipulate the helices to pack more tightly against the beta sheets. Model 4 is T0358.try23-opt2.gromacs0.repack-nonPC.pdb, chosen because it is greatly favored by the Rosetta server. This was an Undertaker optimization of the first chimera we produced, which was not a very good optimization given that the His tag had clashed severely with the subdomain from residues 1-79 in the original model. As a consequence of fixing the backbone clashes, the helices were pushed further away from the beta sheets Model 5 is T0358.try20-opt, chosen to add some diversity to the represented folds,as the N-terminal helix had a unique kink in it. However, this model scored poorly and the helices are very far removed from the beta sheets, leaving many residues exposed that should have been buried. This was another Undertaker optimization performed primarily by modifying the weight of various properties in the cost function, in particular "break", "sidechain", and "phobic_fit". The helices were manipulated by the use of residue-residue contact and distance constraints. Wed Jul 26 01:09:29 PDT 2006 Grant Thiltgen I opted to move the best scoring Rosetta model up to the third model because it is probably more likely than a buried HIS tag. For the final submission, we are submitting Model 1 is T0358.try26-opt2. This model is a polished optimization of a chimera. The chimera was made by using alignments to identify the first seventy-nine residues, after which the remaining residues forming the HIS tag were added on. The subdomain was optimized with distance constraints to bring the helices closer to the sheet. Model 2 is T0358.try5-opt2, another model greatly favored according to the unconstrained scores, in which different weights on different cost functions were experimented with. In this particular model, high weights were added to the rr constraints to pack the molecule closer together. Unfortunately, this is not a very good way to get tighter packing, and probably contributed to the foaminess of the model. Model 3 is T0358.try23-opt2.gromacs0.repack-nonPC, the model that rosetta likes best of the ones for which it repacked sidechains. The overall shape of the protein looks nice, but it suffers from exposed residues and extreme foaminess. It is an optimization of the first chimera produced which had a huge clash with the HIS tag. This caused some of the helices to move away from the sheet. Model 4 is T0358.try27-opt2. It is fairly compact, but it tried to bury the HIS tag. This is a fairly polished model of the original alignment. Residue-residue contact constraints were used to manipulate the helices to pack more tightly against the beta sheet. Model 5 T0358.try20-opt2. It was chosed to add some diversity to the represented folds as the N-terminal helix has a unique kink in it. However, this model scored poorly and the helices are far away from the beta sheet, leaving many residues exposed that were predicted to be buried. -------------------------------------------------------------------------- I submitted these models based on what cynthia recommended. If it gets checked in the morning, these submissions may change. I did not change the manual top hits since there was already a submission made on this model. Wed Jul 26 05:33:35 PDT 2006 Kevin Karplus I looked at the best-models and the comments, but do not have the time to look through dozens of other choices. None of the models have the helices well packed against the sheet, so I don't think any of them are "right". The submission appears to have been done correctly. The first model is probably the best of those submitted, though I woul dhave been tempted to run the A59-I72 helix diagonally across the sheet to get better burial. Fri Sep 1 14:22:41 PDT 2006 Kevin Karplus Our best model was align3, with a GDT of 48.1%, but we did not use this in any of our models. Our best full model was try2-opt2.gromacs0.repack-nonPC (GDT 38.3%). Our model1 adds an extra strand to the sheet, pushing the helix around G60-T69 out of the way. The SAM_T06_server_TS4 model is based on a similar alignment to align3, and is the best thing we submitted (better than any of our hand models).