Thu Jun 22 10:29:43 PDT 2006 T0342 Make started Thu Jun 22 10:31:05 PDT 2006 Running on shaw.cse.ucsc.edu Thu Jun 22 11:51:43 PDT 2006 Kevin Karplus BLAST gets no hits in PDB, but HMMs get strong matches to 1vkbA, 1g0qA, and 1xhsA. Fri Jun 23 04:13:11 PDT 2006 Kevin Karplus We have a good model from the alignments up to about P135, but the helices at the C-terminus (P135-L188) are a bit arbitrary. The initial helix (before D13) is also a bit arbitrary. It might be good to make another model or two from random starts, to get some sampling of the possibilities. The rr constraints look pretty good for this target, so it might be worth including those that involve residues P135 and later. Fri Jun 23 04:38:40 PDT 2006 Kevin Karplus try2 started on shaw Mon Jul 3 18:12:20 PDT 2006 Crissan Harris, Cynthia Hsu Running the rr script, we noticed that there were two residues that should be brought closer together, namely His65 and Tyr139. We also hoped that this would bring the protruding coil (from Phe53 to Ile68) closer to the exposed region of the beta barrel (which should be hydrophobic according to the burial script). The value of this constraint is as follows: Constraint H65.CB Y139.CB -10. 7.0 14.0 0.5 try3 is currently running on lopez. Tue Jul 4 16:50:58 PDT 2006 Cynthia Hsu Somehow, only changing try3 by the constraint on one residue produced completely different results; the residues from 117-188 are on the completely opposite side of the protein from where they were in the try2 model. My gut feeling is that our reasons for making the changes that we did are still valid, but it's really hard to tell. I lowered the constraints on H65.CB Y139.CB from 0.5 to 0.06, since none of Prof. Karplus's constraints on individual residues were over 0.048, so maybe 0.5 was too high for individual residues. I'm also hoping that by simply raising the weight on the breaks to 150, it'll help prevent the backbone from breaking with loose ends appearing at random locations (ie, Ser102 being a loose end). try4 is currently running on camano. Wed Jul 5 10:52:10 PDT 2006 try4 performed better than try3 according to the unconstrained scores, but still not as well as Prof. Karplus's try2 model. The "soft_clashes" for try4 were significantly improved (from 3.1 to 1.9), but other than that, the scores were generally worse, particularly in "phobic_fit", "sidechain", "break", "hbond_geom_beta", and "h_bond_geom_beta_pair". The "phobic_fit" cost was increased from 2 to 5, and the "sidechain" cost was increased from 5 to 7. try5 started on camano on Wed Jul 5 11:11:10 PDT 2006. Wed Jul 5 18:20:32 PDT 2006 Crissan Harris, Cynthia Hsu After the tutorial today, we've actually discovered that try4, since it scores higher than try2 as a constrained model, is actually not that bad. try5 scores the highest in both constrained and unconstrained, though. try5 is significantly worse from try4 in the following: bad_peptide, soft_clashes, and break. In particular, the cost of the break is set at 12.6. We increased "break" to 180, "soft_clashes" to 80, and "bad_peptide" to 15. We also increased the "constraints" cost to 15, with hope that this would put more emphasis on the added constraints. try6 is currently running on whidbey. Thu Jul 6 10:59:53 PDT 2006 Crissan Harris, Cynthia Hsu try 6 did the third best under its own constraints (after try4 and try5), but in the unconstrained models, try5 was still the best, followed by try2 and try4. We decided to focus mainly on the strong attributes of try5, in comparison to 2 and 4. We eliminated try4 from possible models to consider, since it had significantly exposed regions under the burial script. Between try5 and try2, we were unable to identify which one was better. With this in mind, we decided to run subsequent polishing runs for each of them and hope that the results were satisfactory. try7.under was set up as a polishing run of the try2 model. We copied try7.costfcn from try2.costfcn, making the following changes: "phobic_fit" was raised from 2 to 3, "breaks" were raised from 100 to 180, and "soft_clashes" were raised 40 to 80. To help with packing, "dry6.5" was raised to 25 and "dry8" was raised to 20. We also examined the presence of His and Cys; the Cysteine residues did not appear to be in close enough proximity to form a metal bond, so we set the value of this cost to 0. try7 was started on lopez on Thu Jul 6 11:32:46 PDT 2006 try8.under will be the polishing run of the try5 model. In the costfcn, "breaks" were raised to 200 and "soft_clashes" were raised to 60. "dry6.5" was raised to 25 and "dry8" was raised to 20. "maybe_metal" was also set to 0. try8 was started orcas on Thu Jul 6 11:46:59 PDT 2006. Thu Jul 6 13:29:19 PDT 2006 Crissan Harris, Cynthia Hsu Both try7 and try8 returned virtually identical models in RasMol, so we've decided to focus mainly on polishing try7 since it had better unconstrained scores. The protein still has a lot of exposed regions which RasMol indicates to be buried. We attempted to fix these by making the following changes to the dry weights: dry5 20 dry6.5 30 dry8 25 dry12 20 breaks were also increased to 200. try9 is currently running on lopez. Thu Jul 6 16:12:07 PDT 2006 Crissan Harris, Cynthia Hsu Grant explained to us how to use Read-PDB.under to specify which model we wanted to polish, so we decided to rerun the polishing of the try2 model. We copied the try7.under file, then added "ReadConform PDB try2-opt2.pdb". We also used the same costfcn (try7) for try10. Fri Jul 7 10:56:45 PDT 2006 try9-opt2 and try9-opt1 still score the best out of all the models, but we're concerned because the helices form from residues 133-188 are protruding into the environment, and a region of the barrel which is predicted to be buried appears exposed. Looking through all of our best-scoring tries, we noticed that this is a common trend with all of the previous models, so we may have been premature in starting our polishing runs. The protein with the best overall appearance is still try1, based on the fact that most of the regions predicted to be buried are buried, and it has a good, compacted shape (there are no stray helices protruding into the environment). We decided to run try11 using the original costfcn and .under file. Looking at the exposed regions, and the rr constraints, we decided that P153, W80, and R45 should be moved closer together, so that the three terminal helices between 133-188 would move closer together. This would also bring I176 and I180 closer to I143, L160, and Q157, as suggested by the rr script. Wit this in mind, we added the following constraints: Constraint P153.CB W80.CB -10. 7.0 14.0 3 Constraint P153.CB R45.CB -10. 7.0 14.0 4 try11 is currently running on whidbey. Fri Jul 7 13:52:22 PDT 2006 Crissan Harris, Cynthia Hsu Despite the added constraints, P153, W80, and R45 acually seemed to move further apart. Grant suggested that we tried using Carbon-alpha as constraints, instead of the carbon betas. Other than that, try12.costfcn was copied from try1.costfcn. Also, given the lack of Cys and His residues in close proximity, we set "maybe_metal" and "maybe_ssbond" to 0. "break" was also increased from 50 to 100, hoping this will give us a more accurate image of what the model will look like. try12 is currently on Whidbey. Fri Jul 7 15:56:42 PDT 2006 Crissan harris, Cynthia Hsu try12's exposed regions appear less prominent than those on try11, according to the Rasmol visualization. Moreover, the helices are now surrounding the beta sheets, instead of being distributed over to one side. According to the scores, however, trys 9, 7, and 8 are still the best, and try 11 is still listed as scoring substantially better than try12 (175.76 vs. 181.42 in the unconstrained, 174.6 vs. 181.8 in try12). We increased "phobic_fit" to 5. The dry weights were set to the following weights: dry5 20 dry6.5 25 dry8 20 dry12 10 "soft_clashes" was increased to 80 and "break" was increased to 200. We also copied over the helix and sheet constraints from "decoys/try12-opt2.sheets" and "decoys/try12-opt2.constraints", and commented out dssp-ehl2.constraints and undertaker-align.sheets try13 is currently running on whidbey. Sat Jul 8 14:04:07 PDT 2006 Cynthia Hsu try13 doesn't look anything at all like a protein... our alpha helices seem to have flattened themselves into a beta trough surrounding what used to be the beta barrel (but now also appears very much like a trough. Given that we used mostly the same constraints in try13 as we did in try12, I can't come up with a logical explanation for any of it, other than the fact that we may have been overambitious in increasing some of the other costs. With this in mind, I raised the overall weight of "constraints" to 15, reduced "soft_clashes" to 60, "break" to 150. I also decided to lower "dry8" to 18 and return "dry12" to 5, since given the size of the protein, it may have been overly ambitious of us to increase these values so much. try14 is currently running on orcas. Sat Jul 8 20:12:55 PDT 2006 Cynthia Hsu try14 doesn't look any more like try12 than try13 did. In comparing the costfcns for tries 12 and 13, I've decided that it may be in our best interest to include the dssp-ehl2.constraints and the undertaker-align.sheets. I've also reduced "phobic_fit" to 3, raised "sidechain" weight to 7 (hopefully this will address our original problem of packing the helix closer to the barrel) and decrease "dry8" to 15. I also decreased "soft_clashes" to 40. I'm not sure what the effect will be, but it's certainly worth a try. try15 is currently running on shaw. Sun Jul 9 14:21:32 PDT 2006 Cynthia Hsu The try15 model seems to have an approximate shape of what I think the protein should look like.. however, it's still very foamy and according to the burial script, has way too many regions that are exposed. Looking at our best models superimposed, I noticed that try11 and try7 both had significantly exposed helices, so the conclusion was that neither of these were good models to attempt to polish. Of the unconstrained models, try9 still scores significantly better than try11, which still scores higher than try15, but try15 does do better than try12. However, RasMol seems to indicate that in the try9 model, the protein does not fold as would be expected by the RR scripts. I'm not sure which of the three models remaining (try15, try12, or try1) are the best representation of the protien, but I'd like to continue to try and refine tries 15 and 12. For try16, I copied the costfcn of try15, and then changed the "phobic_fit" to 5. I'm not sure what else to do; I'm afraid the helices might unwind and pack into beta sheets the way they did in try13 and try14. I'm not sure what to do with try12, since its costfcn actually isn't that different from try15. try16 is currently running on camano. Mon Jul 10 12:37:53 PDT 2006 Crissan Harris, Cynthia Hsu We decided after much discussion that our favorite models were try12 and try16, since only these succeeded in placing the helices in the appropriate location so as to bury the barrel. Given the minimal differences between the costfunction of try12 and the costfunction of try16, we decided to continue modifying the weights of try16.costfcn for future runs, at least until we had the protein's helices in a satisfactory position. We decided to decrease "sidechain" to 5 (since try16 had a higher sidechain weight than try12, and its sidechains did not appear as well-placed). We also raised "dry8" from 15 to 20. Mon Jul 10 17:13:31 PDT 2006 Crissan Harris, Cynthia Hsu We added the following to the Makefile: K140-L188: /projects/compbio/experiments/protein-predict/casp7/scripts/split-into-domains \ -first 140 -length 999 \ -advance 99999 \ --nomake < ${TARGET}.a2m The subdomain is currently running on Whidbey. Mon Jul 10 17:24:02 PDT 2006 Crissan Harris, Cynthia Hsu Because try9-opt2 and try11-opt2 both score better than try17-opt2, which looks the best in RasMol, we've decided to compare where try17 scores worse than both of these and change these weights as appropriate. In comparison to try9, try17 scores worse in the following: wet6.5, dry5, dry6.5, dry8, n_ca_c, soft_clashes, and breaks. The only improvement try11 has over try17 that try9 does not have is in the dry12 weight. With this in mind, we did the following to the dry weights: dry5 25 dry6.5 30 dry8 23 dry12 7. We also raised "n_ca_c" to 10 and "soft_clashes" 50, and "break" to 180. try18 is currently running on vashon. Tue Jul 11 10:29:23 PDT 2006 Crissan Harris, Cynthia Hsu We examined the subdomain (K140-L188) models that were returned as best-models from the HMMs, in particular, try1. For try19, we copied try9, since this was our top scoring model and we assumed that his meant that it had the best packing except for the helices. We then cut and paste the helical constraints from the subdomain into the costfunction, weighting them as follows: #from subdomain K140-L188 HelixConstraint (T0342)K140 (T0342)N150 0.9 HelixConstraint (T0342)P153 (T0342)E164 1 HelixConstraint (T0342)A162 (T0342)G170 1 HelixConstraint (T0342)S173 (T0342)L188 1 We commented out the helical constraints from the previous tries, and also commented out all of the individual constraints that had been left there from previous runs. Tue Jul 11 15:59:28 PDT 2006 Crissan Harris, Cynthia Hsu We do like the appearance of try19, but the helices do not appear as compact as they did in the subdomain. We decided that the solution to this would be to cut and past the other rr constraints from the K140-L188 subdomain into the costfcn. We also raised the constraints up to 15. try20 is currently running on shaw. Tue Jul 11 22:18:03 PDT 2006 Kevin Karplus To get a helical subdomain intact from a subdomain model, it isn't enough to use the helix constraints, as those put no constraints between helices. The rr constraints will only help if they are extraordinarily correct. The easiest thing to do is to superimpose the subdomain and the whole chain, focusing on some segment of the model that you believe they should have in the same location (maybe K140-N150??), then cut and paste the models. I'll try this for try19-opt2 and K14-L188/try1-opt2 with script make-chimera.under Tue Jul 11 22:30:31 PDT 2006 Kevin Karplus decoys/chimer-19-1.pdb.gz combines these two models, switching from try19-opt2 to K14-L188/try1-opt2 before G146. Tue Jul 11 22:40:55 PDT 2006 Kevin Karplus I made try21.costfcn with sheet and helix constraints from the appropriate parts of the two pieces Tue Jul 11 22:45:27 PDT 2006 Kevin Karplus try21 to try to clean up chimera-19-1 started on lopez. Wed Jul 12 00:34:17 PDT 2006 Kevin Karplus try21 scores almost as well as try9 on the try21 costfcn (better than try19) try21-opt2.gromacs0.repack-nonPC is now rosetta's favorite. The low clashes look fine, but the C-terminal domain is not tightly packed. Perhaps I should do a polishing run on the subdomain and the redo the chimera. Wed Jul 12 00:48:31 PDT 2006 Kevin Karplus K140-L188/try2 started on cheep Wed Jul 12 01:01:56 PDT 2006 Cynthia Hsu Looking at the all.breaks.gz file, it appears that try21 seems to have a very large break at V106. I don't know if there's much of a point to doing this, given that Prof. Karplus is redoing the chimera, but I thought that I might just do a trial run of doing the stuff that I usually do to increase packing and close clashes, given that not that many people wil be using the computers at this hour. I've copied try21.costfn and increased "dry5" to 20, "dry6.5" to 30, and "dry8" to 20. I've rasied "soft_clashes" to 60 and "break" to 20. I also raised "constraints" to 15. I've also changed try22.under to polish try21opt2. try22 started on lopez. Wed Jul 12 09:20:33 PDT 2006 Kevin Karplus With the try22 costfcn, the best models are try22, try9, try8, try21, try7. With unconstrained the order is try22, try9, try21, try7, try8. Rosetta likes best try22-opt2.gromacs0.repack-nonPC In K140-L188/ try2-opt2 scores slightly better than try1-opt2 (slightly less helix, but fewer clashes and maybe tighter packing). Rosetta definitely likes K140-L188/ try2-opt2.gromacs0.repack-nonPC better than other C-terminal models it has repacked, mainly due to the reduced clashes. The differences between K140-L188/ try1-opt2 and try2-opt2 are tiny, but I'll try making a chimera of try22-opt2.gromacs0.repack-nonPC and K140-L188/ try2-opt2.gromacs0.repack-nonPC and optimizing that. It should replace try22 and try21 as our best example of that C-terminal domain. We should also pick out the 4 other models with distinctly different (but plausible) C-terminal domains to submit. I superimposed them with make-chimera2.under, then crossed over between M145 and G146. Note: the "gromacs" models throw away the residue numbering, and undertaker did not quite recover it in making the superposition. Residues before G146 are misnumbered in try22+K140-L188-try2.pdb It looks like the helices were pried apart a little in try22 (and probalby try21) to avoid clashes between P137 and S173. We may want to move around the 3-helix bundle with ProteinShop to reduce the clashes before optimizing. Since I don't have ProteinShop at home, I won't try it, but it looks like a logical step. The linker region from T128 to Y139 should all be flexible. Wed Jul 12 10:44:26 PDT 2006 Kevin Karplus It might be a good idea to make a dimer based on 1v30.mmol from http://pqs.ebi.ac.uk/pqs-doc/macmol/1v30.mmol This might help place the C-terminal domain. The top two hits are monomeric, so this may not be a big help. I fetched the 1v30.mmol file (in 1v30A/1v30.mmol), but have not made the dimer yet. Wed Jul 12 11:54:48 PDT 2006 Kevin Karplus try23-opt2 is the best scoring (with the try23 costfcn), but try22-opt2.gromacs0.repack-nonPC is still rosetta's favorite. Wed Jul 12 12:00:55 PDT 2006 Kevin Karplus I made a dimer dimer/decoys/dimer-1v30-chimera-22-2.pdb using make-dimer.under, superimposing chimera-22-2 on 1v30. Wed Jul 12 12:45:13 PDT 2006 Kevin Karplus I proteinshopped try23-opt2 to make decoys/ProteinShop/try23-shifted.pdb but ProteinShop seems to have messed up portions that I didn't want moved. I'll take F53-I68 and Q138-end from the proteinshop model, the rest from try22-opt2.gromacs0.repack-nonPC, to make chimera-22-shifted.pdb Wed Jul 12 12:56:46 PDT 2006 Kevin Karplus Problem, try22-opt2 and try23-opt2 had the really bad break near Y105. ProteinShop did not see the problem. I'll try wiggling the helix K88-E96 a bit to move the loop around for undertaker to optimize. (This is ProteinShop/try23-shifted2.pdb) Now I'll superimpose the K140-L188/try2-opt2 on chimera-22-shifted.pdb using make-chimera4.under. Nope---that's no good---I didn't move the C-terminal piece far enough in ProteinShop. Wed Jul 12 13:31:04 PDT 2006 Kevin Karplus I tried another ProteinShop edit (try23-shifted3.pdb, and will try superimposing K140-L188/try2-opt2 on it with make-chimera5.under Wed Jul 12 16:17:58 PDT 2006 Kevin Karplus I tried optimizing chimera-23-shifted4-2 as try24-opt2. It came out reasonably good, so I'll make a dimer from it (using make-dimer2.under to produce dimer-1v30-try24.pdb It looks ok, EXCEPT that N55-I68 conflict between the monomers. Moving the tiny helix in the C-terminal direction by about half its length would probably fix the problem. Wed Jul 12 16:57:50 PDT 2006 Kevin Karplus I tried doing that as try24-shifted.pdb and try24-shifted2.pdb Wed Jul 12 17:03:55 PDT 2006 Kevin Karplus oops it looks like I moved the wrong piece. Let's try again. Wed Jul 12 17:33:28 PDT 2006 Crissan Harris, Cynthia Hsu The superimpose-best.under file has been updated with our favorite four models that show a diverse arrangement of helices. We're a little bit hesitant about including try1-opt2, as it has some regions predicted to be exposed that are not exposed, but it scores better than our other fourth-choice model, try12-opt2, which looks very similar to try19-opt2. Following Prof. Karplus's advice, we decided to see if the model could possibly be optimized as a dimer. Going through our favorite models, we decided not to use try22, since that had been used in Prof. Karplus's dimer (dimer/decoys/dimer-1v30-chimera-22-2.pdb), and did not use try19-opt2, as it was difficult to visualize how this would form a dimer interface given the orientation of the helices. With this in mind, we decided to make a dimer out of try16-opt2.pdb. We made the appropriate changes to make-dimer4.under, so that it would use try16-opt2 as an input and output the dimerized result to dimer-1v30-try16.pdb. Wed Jul 12 17:34:53 PDT 2006 Kevin Karplus I ran the ProteinShop model try24-shifted4 through gromacs and rosetta repacking (to get rid of some of the worst of ProteinShop damage, then dimerized it with make-dimer5.under as dimer-1v30-24-shifted4-repack There is a bit of clashing still, but I'll try optimizing this as a dimer. Wed Jul 12 17:38:09 PDT 2006 Crissan Harris, Cynthia Hsu dimer-1v30-try16.pdb looks very very foamy, and has many regions predicted to be buried that are exposed. We don't think it is actually worth optimizing in the form of a dimer. Wed Jul 12 17:43:13 PDT 2006 Kevin Karplus I started dimer/try1 to optimize dimer-1v30-24-shifted4-repack (which is the best of the current dimers, sad as it is). Wed Jul 12 20:21:25 PDT 2006 Kevin Karplus The dimer optimization in dimer/try1 is fairly useless, as the dimerization has changed. Perhaps I should give up on dimerizing and just make a monomer. I don't have time to do another ProteinShop run tonight, though one could be done by Crissan and Cynthia before the hard deadline. Looking at the dimer with the "near" script, it looks like the 3-helix bundle needs to be flipped over and nestled against the N-terminal domain---each has a hydrophobic patch that could be lined up. Wed Jul 12 20:44:16 PDT 2006 Kevin Karplus Preliminary submission done: InFilePrefix decoys/ ReadConformPDB T0342.try22-opt2.gromacs0.repack-nonPC.pdb InFilePrefix dimer/decoys/ ReadConformPDB T0342.try1-opt2.unpack.pdb chain A InFilePrefix decoys/ ReadConformPDB T0342.try19-opt2.pdb ReadConformPDB T0342.try16-opt2.pdb ReadConformPDB T0342.try1-opt2.pdb Date: Thu, 13 Jul 2006 11:38:34 -0700 (PDT) Subject: T0342 in ProteinShop From: cthsu To: karplus To Professor Karplus: In your last update of the README, you said that you wanted us to proteinshop an additional monomer before the hard deadline, but Crissan and I were unsure as to which model we should attempt to protein shop. We also wanted to know if we should attempt to work with all of the sidechains that were not placed as indicated by the burial and near scripts, or if we should focus on the helices generated by the subdomain. Thanks! Sincerely, Cynthia Hsu Date: Thu, 13 Jul 2006 12:18:37 -0700 From: Kevin Karplus To: cthsu Subject: Re: T0342 in ProteinShop Start with the first model in superimpose-best.under, that is T0342.try22-opt2.gromacs0.repack-nonPC.pdb If you look at it with the "near" script, you should see two hydrophobic patches: one involving V172, I163, I176, I180 the other involving V79, L51, A69 (you may have to hide the C-terminal domain to see the earlier hydrophobic patch). Take the C-terminal domain as a unit (by selecting the Y139-E149 helix) and turn the whole domain over to bring the hydrophobic patches close. This may require first moving it out to straighten the linker region, turning it over, then bringing it back in. ------------------------------------------------------------ Thu Jul 13 13:54:29 PDT 2006 Crissan Harris, Cynthia Hsu Under Prof. Karplus's instructions (as copied above, we Protein-Shopped try22-opt2 gromacs repacked to bring residue V172 and V79 together. The file was saved as try22-shifted.pdb For try25.costfcn, we copied the original cost function for try1, but commented out the original constraints and added the ones from try22-opt2.sheets and try22-opt2.helices, giving each a weight of 1. try25.under was a polishing run, as was try22.under, but it took try22-shifted.pdb as its input. try25 is currently running on lopez. Thu Jul 13 15:59:38 PDT 2006 Crissan Harris, Cynthia Hsu For some reason, Undertaker, when optimizing, attempts to form a hydrogen bond between E119 and N150. Moreover, examining our Protein-Shopped try22-shifted, we found that we actually did not move the subdomain close enough to the hydrophobic regions. We did a second run of ProteinShop, labeled as try22-shifted2.pdb. We also decided to try a third model, in which the orientation of the three helix subdomain bundle was reversed from the way it was in try22-shifted2. We prefer the appearance of try22-shifted3 to try22-shifted2, but will refrain from making any judgements until Undertaker has optimized both models. try22-shifted2 was the input for try26.under. try26 is running on bacchus. try22-shifted3 was the input for try27.under. try27 is running on iris. Fri Jul 14 11:27:03 PDT 2006 Crissan Harris, Cynthia Hsu both try26 and try27 came out very very badly. In try26, Undertaker insisted on moving the subdomain so that it was on the other side of the barrel, which also leaves the residues H65-I68 (that form the loop) exposed, contrary to the burial script. In try27, the subdomain is forced perpendicular to the barrel so that it extends out into the environment, and also ends up on the other side of the hydrophobic regions. After about an hour of working in ProteinShop and examining our try22-shifted models in Rasmol, we realized that part of the reason why it failed is because of severe clashes in which the coils would intersect. We attempted to fix the ones in try22-shifted3 slightly (try22-shifted5) and we've decided to go ahead and put it through Undertaker to see how it would optimize the model. try28 is currently running on iris. Fri Jul 14 16:27:39 PDT 2006 Sylvia Do, Cynthia Hsu In try28 the terminal helix (from K144-L188) was for some reason pushed into the environment, destroying the three-helix bundle that was created by the subdomain. Also, for some reason, the placement of the subdomain was very different from what it was in ProteinShop. For future reference, when we run the undertaker optimization of our ProteinShopped models, we should include the rr constraints from the subdomain to preserve the structure of the three-helix bundle. Fri Jul 14 17:00:32 PDT 2006 Sylvia Do, Cynthia Hsu After about half an hour of ProteinShop, we seem to have the subdomain placed to our liking. However, examining the target under the near and burial scripts, we're a bit concerned because it seems that the orientation of the helices so that they are close to the barrel will bury residues that should be exposed. We've decided to run try29 with the rr constraints for the subdomain on try22-shifted7, but will be doing another protein-shopped run in the event that the subdomain should have been facing the other direction. try29 is running on camano. Sat Jul 15 12:52:18 PDT 2006 Cynthia Hsu The subdomain remained intact, but we appeared to have BROKEN the coils from 100-108 and 125-140 into fragments. Our hydrophobic patches in the near script did manage to stay together, however. There are two things I think we can do at this point. First, I can rerun try29, but this time raise the weight of breaks really high. Second, I can try to ProteinShop the domain again, this time being very very careful to avoid breaks. For try30, I copied the try29.costfcn. I raised the breaks up to 150. Then, for try30.under, I changed it so that it would accept try29-opt2.pdb as the model to polish. try30 is currently running on shaw. After carefully examining try22-shifted7 in RasMol and the gromacs model that it was based on, I've found that the break in the coil from 100 to 108 had actually existed there in the try22 model. With this in mind, I've decided to wait and see how try30 turns out before attempting another ProteinShop run. As a reminder to myself, I should, in the future, start from the original gromacs model when ProteinShopping, instead of from try22-shifted3 as I have been doing in the past. Sat Jul 15 19:16:44 PDT 2006 Cynthia Hsu The breaks still haven't gone away. I've decided to try one more run with "break" raised to 200 and "soft_clashes" raised to 70. For try31.under, I used the try30-opt2 model as the input. At first I considered restarting it from the try22-shifted7 model, but looking at the unconstrained scores, it seemed that try30 actually scored really well. try31 is currently running on shaw. I considered that it might be necessary to ProteinShop the target again, but examining all of our best models showed that the break in the coils from 100 to 108 and 125-140 were common across all models. In view of this, I decided to continue optimizing try22 to attempt to get rid of the breaks, providing us with a slightly different model from which we could ProteinShop. For try32.costfcn, I raised the weight on "break" up until 200. I debated whether to use try22-opt2 or the gromacs version for the input model, but decided to attempt try22 since tries 29-31 were all based on the gromacs versions. try32 is currently running on orcas. Sun Jul 16 16:17:25 PDT 2006 Cynthia Hsu Even with "break" weight turned up to 200, RasMol showed no visible difference in the size of the break; try31 still insisted on forming hydrogen bonds between the fragments rather than closing the gaps in the backbone. In try32, the break also failed to close, which was immensely frustrating. I checked to see if residues G98 to M104 had any helix constraints that were forcing them to form a helix and therefore preventing them from sealing the break, but there was no such luck. I'm wondering if this problem may come from the fact that we are telling Undertaker to do polishing runs, therefore preventing any major change in conformation. For try33.costfcn, I added weights of 2 to all of the present helix constraints. Hopefully this will discourage Undertaker from forming accessory helices. I raised the "break" weight up to 250, although given what little difference raising the break weight has made in the past, I'm not sure if this will help. Because of our previous problems with closing the breaks, I decided to NOT make try33.under a polishing run. Hopefully this will solve the problem. [Tue Jul 18 17:34:38 PDT 2006 Kevin Karplus Increasing the weight of a helix constraint does not discourage other helices from forming. There is no "not a helix" constraint (other than StrandConstraint and SheetConstraint). ] try33 is currently running lopez. Sun Jul 16 17:54:10 PDT 2006 Cynthia Hsu try33 did close both of the breaks (the ones from 100-108 and 124-140) but it produced a new one from V83 to K85 that I'm concerned about. To deal with this, I changed the SheetConstraint: SheetConstraint F41 D48 W84 D77 hbond C42 1 to the following: SheetConstraint V43 D48 V82 D77 hbond A44 1 Hopefully this will prevent residues 83 and 84 from attempting to hydrogen bond with other beta sheets and seal the break. I've also lowered the weight of "constraints" to 12 to deal with this. I was also concerned by the fact that our three helix bundle seemed to be disintegrating, so I added the rr.constraints from the subdomain into the try34.costfcn. try34 started on lopez Sun Jul 16 18:09:41 PDT 2006 Mon Jul 17 09:00:42 PDT 2006 Cynthia Hsu I'm concerned because our three helix bundle seems to have disintegrated completely. I've removed the "bonus" part from the rr.constraints, but I don't know if this will have the desired effect. I finally found the pesky constraint that was causing the break: N87 was in both a sheet and a helix constraint; the helix constraint had a much higher weight. I changed these two constraints as follows: SheetConstraint F17 Y22 N87 V82 hbond F17 2 HelixConstraint N89 E98 0.5 Hopefully that will, if nothing else, close the gap. Just for good measure, the weight on "break" was raised to 270. try35 currently running on shaw. Mon Jul 17 11:02:20 PDT 2006 Crissan Harris, Cynthia Hsu The breaks were successfully closed, but the three helix bundle doesn't look anything at all like a three-helix bundle anymore. I decided to attempt to make a chimera with try35-opt2 and the try2-opt2 model of the dimer. The superimposed output file was try35-opt2+K140-L188-try2.pdb. The file chimera-35-2.pdb contains the desired chimera. We're concerned because chimera-35-2 appears to have a very large break in the vicinity of K141 and Q138. Looking at the PDB file, we're not sure how to fix it, but we think that the superimpose had problems. Mon Jul 17 11:59:59 PDT 2006 Crissan Harris, Cynthia Hsu We readjusted make-chimera8 so that it would start the superimposition starting at K140 intead of K141. We also kept the constraints from residue 141. However, this just moved the break back one residue. We've decided to attempt to optimize it by using Undertaker to close the break. For the try36 optimization of chimera-35-2-2.pdb, we removed the helical and sheet constraints that were relevant to residues 1-140 and instead just left those relevant to the subdomain. "break" was left at the same value of 270. try36.under was a polishing run that would take chimera-35-2-2.pdb as an input (the second chimera was chosen instead of the first because we felt that the gap in the middle of the helix would be easier for Undertaker to close). try36 is currently running on shaw. Mon Jul 17 16:24:11 PDT 2006 Crissan Harris, Cynthia Hsu try36 performed satisfactorily (the subdomain remained in a three-helix bundle and there were no particularly large breaks. We protein-shopped it according to Professor Karplus's instructions (above) and then polished it in Undertaker. Because there were so many changes between the optimal try22 and the very heavily modified try36, for try37 (since this was the model upon which try37-shifted was based), we copied try1.costfcn. We then added in the helical and sheet constraints for try36, with the addition of the helical constraints and the rr constraints from the subdomain. Given Undertaker's history with ProteinShopped models, I removed the "bonus" tags on the RR constraints, and raised the weight of "break" to 150 and the weight of "soft_clashes" to 50. [Tue Jul 18 17:37:01 PDT 2006 Kevin Karplus The RR constraints are probably not the right ones to be using. What you want to do is to measure the distances (using rasmol "pick distance") between atoms (say, CB atoms) on a model that has the region the way you want it and create constraints based on those distances, allowing more or less slop, depending on how much you trust the measured model. The rr constraints (1) may not apply to the right residues and (2) may not have the right distances. ] try37 is currently running on squawk. Mon Jul 17 19:02:15 PDT 2006 Cynthia Hsu Undertaker distorted the C terminal helix so that it is protruding into the environment. The subdomain itself appears distantly removed, no longer covering the barrel as well as it had originally. I turned off "phobic_fit", hoping that this would prevent the sidechain from elongating. I also lowered the weight of "break" slightly to 140. In addition, I decided to remove the try36-opt2 helix and sheet constraints, given that try36 was not very well optimized in the first place. [Tue Jul 18 17:39:34 PDT 2006 Kevin Karplus You need constraints (measured from the proteinshopped model) *between* the domains to hold the subdomain in place until the backbone can close. ] Because try37 failed to perform satisfactorily, try36-shifted.pdb was left as the input file. try38 is currently running on Whidbey. Tue Jul 18 10:07:55 PDT 2006 Crissan Harris, Cynthia Hsu try38, unlike try37, did maintain the structure of the three helix bundle, but the subdomain failed to remain in place. Our guess that this is a consequence of turning the "phobic_fit" off. It may have been a better idea to raise "phobic_fit", so that the subdomain would still remain in the vicinity of the rest of the protein. In view of this, we raised "phobic_fit" to 4. "break" was also slightly lowered to 120. try39 is currently running on shaw. Tue Jul 18 11:55:35 PDT 2006 Crissan Harris, Cynthia Hsu try39 failed to maintain the three helix bundle. However, the target, when viewed in RasMol, generally did not contradict the predictions of the burial script and the near script. We've raised "constraints" to 15, in hopes that this will force the helices to stay together. The dry weights were raised to the following, to try to encourage the target to pack more closely together: dry5 20 dry6.5 30 dry8 20 dry12 7 try40.under took try39-opt2 as the input model. try40 is currently running on pyro. Tue Jul 18 15:50:58 PDT 2006 Crissan Harris, Cynthia Hsu try40 still did not maintain the three helix bundle, but it is the best-scoring model we've had since try32 (in other words, since we fixed the break). We considered making a chimera out of try40, but the superimposed model that was produced had a really bad clash between residue 187 and 88. We decided to repeat the process with try39 using the make-chimera10.under file. Attempts to make a chimera using try1-opt2 of the dimer also failed, always resulting in the terminal helix being tucked too close to the barrel. Tue Jul 18 16:53:04 PDT 2006 Cynthia Hsu Following Prof. Karplus's suggestions, I've decided to close the breaks in our previous models (the ones successfully optimized by Undertaker from Photoshop) by lowering the constraints. In try41.costfcn, which was copied from try32.costfcn, I've changed a helix constraint to span from N87 to Q97, instead of to E98 as it had originally, as the helical structure from 98-104 is part of what's causing the break. I've also lowered "sidechain" to 3. Lastly I took "constraints", which was originally set to 15, and lowered it to 8. I refrained from setting this up as a polishing run, as I wanted the target to do as much as it could to fix the breaks. [Tue Jul 18 17:41:28 PDT 2006 Kevin Karplus I'm not sure what you mean by "I refrained from setting this up as a polishing run". To me, a polishing run is one that starts from existing models and tries to modify them in small ways (reducing clashes, closing small gapgs, packing tighter, ...). It looks like try41 is starting all over from alignments---unless you get very lucky, it is unlikely to do as well as what you have previously polished, as it has to find a fold-recognition alignment and build all the loops from scratch. ] try41 is currently running on peep. Tue Jul 18 17:44:19 PDT 2006 Kevin Karplus superimpose_best.under seems to be specifying 3 different treatments of the C-terminus: try22-opt2.gromacs0.repack-nonPC & try32-opt2 try19-opt2 try40-opt2 & try39-opt2 try22-opt2... is still Rosetta's favorite, though the backbone around Y105 is a mess. (Rosetta assumes that the backbone always has good bond angles and bond lengths, but this is not always true of gromacs-optimized models.) Looking at decoys/all.breaks.gz, we can see that undertaker is aware the gromacs messed up: Conformation[86] T0342.try22-opt2.pdb.gz has 23 breaks T0342.try22-opt2.pdb.gz breaks before (T0342)V106 with cost 5.42092 T0342.try22-opt2.pdb.gz breaks before (T0342)P36 with cost 0.63821 Conformation[84] T0342.try22-opt2.gromacs0.pdb.gz has 12 breaks T0342.try22-opt2.gromacs0.pdb.gz breaks before (T0342)V106 with cost 4.44504 T0342.try22-opt2.gromacs0.pdb.gz breaks before (T0342)Y105 with cost 2.97639 T0342.try22-opt2.gromacs0.pdb.gz breaks before (T0342)V107 with cost 2.52842 T0342.try22-opt2.gromacs0.pdb.gz breaks before (T0342)S37 with cost 1.47246 T0342.try22-opt2.gromacs0.pdb.gz breaks before (T0342)P36 with cost 0.967777 T0342.try22-opt2.gromacs0.pdb.gz breaks before (T0342)S102 with cost 0.621294 I think you want to go back to your favorite proteinshopped model, add some constraints for the hydrophobic regions you want to hold together. I tried looking at try22-shifted7, but I don't like how M127-P135 wraps so tightly around the protein. Also, the 3-helix bundle seems very loose. Do we have a tighter 3-helix bundle anywhere? Did we ever try packing the 3-helix region as a subdomain? try36-shifted has the loose bundle and a very awkward loop for F53-I68. Tue Jul 18 18:31:26 PDT 2006 Cynthia Hsu As Professor Karplus predicted, try41 turned out to be completely different from expected. I decided to rerun the try again, but this time having it specifically polish try32-opt2. I refrained from doing this in try41, as I thought that specifying an input model would prevent it from performing the large structural changes that would be necessary to close the very large gaps in try32-opt2, but given that try41 came out with a completely different structure, it may be worth trying. try42 is currently running on peep. Thu Jul 20 10:17:35 PDT 2006 Crissan Harris, Cynthia Hsu Not only did try42 fail to close the break, it also moved the subdomain. We reviewed Prof. Karplus's advice in using RasMol to pick the distance constraints, and with Pinal's help, added the following distance constraints from the try36-shifted model to the try43.costfcn (copied from try37.costfcn): constraint((T0342)V79.CA, (T0342)I163.CA) [> -10 = 7 < 14] w=1 # (weighted) cost= 1.014 dist= 5.724 constraint((T0342)G81.CA, (T0342)I180.CA) [> -10 = 7 < 14] w=1 # (weighted) cost= 1.014 dist= 6.763 constraint((T0342)I43.CA, (T0342)I180.CA) [> -10 = 7 < 14] w=2 # (weighted) cost= 1.014 dist= 9.842 We chose to leave the distance constraints with a weight of 1 in hopes of allowing Undertaker some flexibility during optimization. We also lowered the weight or the rr constraints by 0.2, each. try43.under used try36-shifted as the input PDB. try43 is currently running on peep. We also decided to attempt to close the breaks in try31-opt2. We copied the try31 costfcn, and then as before, lowered the rr constraint weights by 0.2 each. We added the following distance constraints from RasMol: constraint((T0342)I43.CA, (T0342)I180.CA) [> -10 = 7 < 14] w=1 # (weighted) cost= 1.014 dist= 9.737 constraint((T0342)G81.CA, (T0342)I180.CA) [> -10 = 7 < 14] w=1 # (weighted) cost= 1.014 dist= 12.362 "soft_clashes" was raised to 80, and "break" was raised to 240. I also lowered "sidechain" to 3. try44 is currently running on lopez. Thu Jul 20 12:52:30 PDT 2006 Crissan Harris, Cynthia Hsu Despite our distance constraints, in try43 Undertaker still refused to keep the subdomain intact, or even hold it in the same location as it was previously. I raised all of the rr.constraints by 0.3 (0.1 more than they were originally). Because one of the helices had been split in two, I heightened the helix constraint as follows: HelixConstraint (T0342)K140 (T0342)N150 4 We also modified the constraints as follows: constraint((T0342)V79.CA, (T0342)I163.CA) [> 5.4 = 5.7 < 6] w=5 # (weighted) cost= 1.014 dist= 5.724 constraint((T0342)G81.CA, (T0342)I180.CA) [> 6 = 6.8 < 7.5] w=5 # (weighted) cost= 1.014 dist= 6.763 constraint((T0342)I43.CA, (T0342)I180.CA) [> 9 = 9.8 < 10.5] w=5 # (weighted) cost= 1.014 dist= 9.842 try45 is currently running on peep. Thu Jul 20 13:49:21 PDT 2006 Crissan Harris, Cynthia Hsu Try44 failed to close the breaks, as we had hoped. We decided to attempt to close these by adding distance constraints to bring together the appropriate residues. We added the following distance constraints: //from RasMol pick distance, to close the breaks: constraint((T0342)Y105.CA, (T0342)V106A.N) [> 0 = 0.5 < 1] w=3 constraint((T0342)V106.CA, (T0342)V107A.N) [> 0 = 0.5 < 1] w=3 We also modified the distance constraints on the subdomain as follows: constraint((T0342)I43.CA, (T0342)I180.CA) [> 9 = 9.7 < 10.5] w=1 # (weighted) cost= 1.014 dist= 9.737 constraint((T0342)G81.CA, (T0342)I180.CA) [> 11.5 = 12.4 < 13] w=1 # (weighted) cost= 1.014 dist= 12.362 try46 is currently running on lopez. Thu Jul 20 15:10:16 PDT 2006 Crissan Harris, Cynthia Hsu In try45, the subdomain remaind in its appropriate form, but the subdomain was still displaced to the other side of the barrel. We noticed that the beta sheet was contorted as the subdomain was brought over the other side of the barrel, so we decided to raise "n_ca_c" to 7 and hope tha tthis would have a role in keeping the subdomain in place. We raised the overall "constraint" weight to 15, and set the constraints below to have a weight of 10: constraint((T0342)V79.CA, (T0342)I163.CA) [> 5.4 = 5.7 < 6] w=10 # (weighted) cost= 1.014 dist= 5.724 constraint((T0342)G81.CA, (T0342)I180.CA) [> 6 = 6.8 < 7.5] w=10 # (weighted) cost= 1.014 dist= 6.763 try47 is currently running on peep. Thu Jul 20 16:55:35 PDT 2006 Cynthia Hsu The breaks in try46 did not close. In hopes of giving it a little bit more flexibility, I've modified the constraints as follows: //from RasMol pick distance constraint((T0342)I43A.CA, (T0342)I180A.CA) [> 4 = 9.7 < 11] w=3 # (weighted) cost= 1.014 dist= 9.737 constraint((T0342)G81A.CA, (T0342)I180A.CA) [> 6 = 12.4 < 14] w=3 # (weighted) cost= 1.014 dist= 12.362 constraint((T0342)L27A.CA, (T0342)I180A.CA) [> 1 = 6.8 < 8.5] w=3 //from RasMol pick distance, to close the breaks: constraint((T0342)Y105.CA, (T0342)V106A.N) [> 0 = 0.5 < 1] w=15 constraint((T0342)V106.CA, (T0342)V107A.N) [> 0 = 0.5 < 1] w=15 I've also lowered "sidechain" to 1. I used try44 as the input PDB for try48.under, as try44 performed better on the breaks overall. try48 is currently running on camano. Looking at the unconstrianed scores, I also noticed that try46 performed the best out of all models produced thus far, besides large breaks. try42 did very well on breaks, with only one arge break before V106. try47 perofrmed even better on breaks. However, it failed to keep the subdomainin the appropriate location. Fri Jul 21 10:18:14 PDT 2006 Crissan Harris, Cynthia Hsu The gap befor Y105 still refused to close. We decided that this may have been too large of a gap to close with a polishing run, so for try49 we decided to, keeping the costfcn for try48, but make the following changes: We replaced the helical constraints from try22-opt2 pertaining to the subdomain, and instead replaced them with those from the subdomain directory, eahc of which were given a weight of 2. We also added the following two distance constraints, to preserve the structure of our bundle: constraint((T0342)I176A.CA, (T0342)S24A.CA) [> 4 = 9.6 < 11] w=3 constraint((T0342)E155A.CA, (T0342)F41A.CA) [> 8 = 12.5 < 13.5] w=3 "sidechain" was set to 0. We also used the try42-opt2 helix and sheet constraints, instead of the try22 ones, as try42 was a much better scoring model. try49 is currently running on orcas. Fri Jul 21 12:33:15 PDT 2006 Cynthia Hsu try49 completely obliterated everything. I'm wondering if the constraints were inappropriately weighted, as we had focused more on closing the breaks than keeping the subdomain intact and in place, and that was what the consequential result of try49 was. For try50.costfcn, I raised the weights of all the rr constraints by 0.4, then raised the weights on the distance constraint that kept the helix together to 10. The weights of the other distance constraints were lowered from 15 to 8. I also lowered "break" slightly to 170, and therefore reduced "soft_clashes" to 60. The overall weight of "constraints" was raised to 15. I also commented out the t04 and t06 alignments from the try50.under file, hoping that this would limit the number of alignments that it used and therefore emphasize the constraints we had included in the costfcn. try50 is currently running on squawk. Fri Jul 21 16:01:49 PDT 2006 Crissan Harris, Cynthia Hsu Professor Karplus advised for us to just focus on packing things tighter instead of trying to fix the large breaks or move the subdomain. In view of this, we went through and selected our favorite models, with the following comments: InFilePrefix decoys/ ReadConformPDB T0342.try22-opt2.gromacs0.repack-nonPC.pdb //Rosetta's favorite //Should be polished so that it packs much tighter. InFilePrefix decoys/ ReadConformPDB T0342.try42-opt2.pdb //similar to try22 (breaks were somewhat closed) //refrain from packing until Rosetta has been packed. ReadConformPDB T0342.try44-opt2.pdb //try packing ReadConformPDB T0342.try19-opt2.pdb //favorite pre-ProteinShop ReadConformPDB T0342.try40-opt2.pdb // top scoring model with minimal breaks. #ReadConformPDB T0342.try28-opt2.pdb //destroys three-helix bundle; pushes one into the environment - therefore does not do well under the burial script. ReadConformPDB T0342.try26-opt2.pdb //displaces subdomain. ReadConformPDB ProteinShop/try36-shifted.pdb // favorite protein-shopped model, unoptoimized - looks decent unpacked, but has some severe clashes. #ReadConformPDB ProteinShop/try22-shifted7.pdb //does not look good as try36. SShould not be included in best-models, as exists in optimized form. First, in try51, we polished the top Rosetta model. In the cost function (copied from try22), "dry5" was raised to 25, "dry6.5" was raised to 36, "dry8" was raised to 23, "n_ca_c" was raised to 7, "bad_peptide was raised to 12, "soft_clashes was raised to 80, and "break" was raised to 180. try51 is currently running on shaw. After reexamining the unconstrained scores, we decided that it would be better to polish the repacked try31-opt2 instead of try44, since they were based on similar models but try31 scored much better. For try52, we copied try31.costfcn, then set "dry5" to 20, "dry6.5" to 25, "dry8" to 20, "sidechain" to 3, "n_ca_c" to 8, "bad_peptide" to 12, and "break" to 240. try52 is currently running on orcas. Instead of packing try19, we decided that we visually preferred try12's appearance the most out of all of our pre-proteinshopped models. We increased the dry weights as follows: dry5 25 dry6.5 30 dry8 20 "n_ca_c" was raised to 7, "bad_peptide" was 11, "soft_clashes" was raised to 80 and "break" was raised to 180. We commented out the dssp-ehl2 constraints, the undertaker alignments, the rr constraints, and then added the helix and sheet constraints from the respective try12-opt2 files. try53 is running on lopez. For try54, we copied the costfcn for try40. We raised "dry8" to 25, "n_ca_c" to 8, "bad_peptide" to 11, and "soft_clashes" to 60. try54 is currently running on camano. For try55, we copied the costfcn for try26. We raised "dry5" to 20, "dry6.5" to 30, and "dry8" to 20. We also lowered "sidechain" to 4, set "n_ca_c" to 7, raised "bad_peptide" to 12, soft_clashes" to 50, and "break" to 150. try55 is currently running on shaw. Sat Jul 22 11:36:28 PDT 2006 Cynthia Hsu try42-opt2 is our second highest scoring model with an unconstrained costfunction, after try46, which has really bad "n_ca_c", _bad_peptide", and "break" scores. try42-opt2 was an attempt to close the breaks that had existed in try32-opt2. try32-opt2 had been a polished version of try21, which was a chimera between the first try of the subdomain from residues K140-L188 (which had formed a three-helix bundle) and a barrel-shaped core from try19. We had attempted to close the breaks in this model in an effot to have a better model from which to begin ProteinShopping (try36-shifted), although no ProteinShop was used in generating this model, and the large hydrophobic patches on the subdomain and barrel are still exposed. try52-opt2 was a polished version of try31-opt2 (repacked). In try31-opt2, we attempted to close the breaks in that had existed try30-opt2, which was an Undertaker optimized version of the ProteinShopped model, try22-shifted7. The ProteinShopped model was an attempt to manipulate two hydrophobic patches, one on the three-helix subdomain and one on the barrel core, so that they would bury each other, which caused the breaks seen in the model. try22 was a polished chimera of the three-helix subdomain of residues K140-L188 and the barrel-shaped core. T0342.try55-opt2.gromacs0.repack-nonPC.pdb.gz was the highest scoring model in Rosetta, and the try55-opt2.pdb.gz file scored the third best (out of unique folds) in an unconstrained costfunction. We polished try26-opt2 in this model, which was an Undertaker optimization of try22-shifted2, a ProteinShopped model. As with other ProteinShopped models, this was an attempt to manipulate the hydrophobic regions of the three helix bundle so that it covered the hydrophobic patches on the barrel, but unlike try22-shifted7, the non-hydrophobic regions of the three-helix bundle were left protruding into the environment, instead of being aligned parallel to the barrel. try22 was a polished chimera of the three-helix subdomain of residues K140-L188 and the barrel-shaped core. ProteinShop/try36-shifted.pdb was our favorite of our Protein-Shopped models, based on try36. It achieved our goal of aligning the hydrophobic regions of the three helix subdomain from residues K140-L188 to bury the hydrophobic patch on the barrel, but because of severe clashes, could not be optimized in Undertaker. try36 was the final result of attempting to reduce the large gaps in the try22 model. try22 was a polished chimera of the three-helix subdomain of residues K140-L188 and the barrel-shaped core. We were considering, as the fifth model, of submitting one of the following: try22 (gromacs repacked), because it was the starting point for so many of our models and is still Rosetta's second favorite. There was a polishing run (try51) but this still failed to score as well with Rosetta, by a difference of about 8. We are also concerned, as try42-opt2 will have a similar appearance to try22, as both of them were highly polished version of the same chimera (try21). try46 scored very very well, but in RasMol, appears virtually identical to try52. try40 was a polishing run of try39-opt2, which had been an Undertaker optimization of try36-shifted, the ProteinShopped model based on try36 (chimera with breaks from try22 shifted). There were no extremely large regions that were exposed when they shouldn't have been, but the three helix bundle disintegrated. try53, since it was a polished version of our try12 model, which was our favorite in terms of appearance out of all of the models we generated before the formation of the chimera. However, it scored very poorly, considerably worse than even try1. try19, since that was the second favorite (according to Prof. Karplus) submitted during the preliminary submission, and consisted mostly of work done by Undertaker (in other words, no chimera-formation or ProteinShop), which produces a distinct fold. try9, given that that is the highest scoring (by a significant margin) out of all of our pre-chimera and ProteinShop models (comprable even to our most recent models) and has a distinct fold. However, we do not particularly like the overall appearance of this protein. try16, which was our second favorite models that scored poorly (however, still better than try12). superimpose-best.under was updated to include all of the aforementioned models. Also, I intend to put try42 through a polishing run, to improve the packing. Sat Jul 22 13:10:28 PDT 2006 Cynthia Hsu A polishing run was performed on try42, with the following changes to the cost function: "dry8" raised to 23 and "dry12" raised to 5.5. "n_ca_c" was raised to 6 and "bad_peptide" to 11, and "soft_clashes" were raised to 80. try56 is currently running on orcas. Sat Jul 22 15:10:13 PDT 2006 Cynthia Hsu Given that Prof. Karplus has not yet responded to my email, I've decided that I'll go through the possible targets for the fifth model to submit and try to select a couple that I think are worth polishing: ReadConformPDB T0342.try22-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0342.try40-opt2.pdb ReadConformPDB T0342.try53-opt2.pdb ReadConformPDB T0342.try19-opt2.pdb //ReadConformPDB T0342.try9-opt2.pdb This has protruding C terminal helices that are exposed. ReadConformPDB T0342.try16-opt2.pdb C terminus situated over the barrel in a strange T-shaped conformation; exposes some residues but could benefit from closer packing. I found myself curious as to whether or not try16 could benefit from tighter packing, so I decided to attempt another run on it. For try57,I raised "dry8" to 20 and "dry12" to 6. "sidechain" was lowered to 3 to allow the helices to move a little bit more, and I lowered the constraints on the three terminal helices to 0.5, in case they might need to unwind to fit over the barrel the way I wanted them to. try57 is currently running on peep. Chances are this will not produce the desired result, but I'm not sure where my time would be better spent. At the moment, I am also leaning towards try22 as the fifth choice model. It scores higher than the others, and if you look at it superimposed, it is actually a little different than the other top models. Sat Jul 22 15:40:33 PDT 2006 Cynthia Hsu try56 succeeded in scoring better than both try46 and try42. The best-models were updated accordingly; try9 was also commented out. Sat Jul 22 18:48:57 PDT 2006 Kevin Karplus I was going to comment on the models, but I was so tired I fell asleep for a while. I'll try again after supper. Sat Jul 22 20:47:03 PDT 2006 Kevin Karplus The barrel is essentially the same in all models, only the alpha helices differ. We want to include a few of the 3-helix bundles, and one or two other possible locations for the helices. Sat Jul 22 21:01:18 PDT 2006 Kevin Karplus I think that try40-opt2 makes for an interesting positioning of the helices (though one could wish that they were packed in tighter). try16-opt2 is also a possiblity. try22-opt2 is too similar to try56-opt2 to include both. I'll submit try56-opt2 best scoring with try56 costfcn (rosetta likes it best after gromacs has reoptimized it) try55-opt2.gromacs0.repack-nonPC (rosetta's second favorite) try52-opt2 a different positioning of a 3-helix bundle try40-opt2 a non-bundle version of the C-terminus try16-opt2 a different non-bundle version of the C-terminus Sat Jul 22 21:21:00 PDT 2006 Kevin Karplus Submitted with comment We got fairly easy fold recognition for the N-terminal barrel, but we had more difficulty with the C-terminal helices, so we did a separate prediction for K140-L188, resulting in a 3-helix bundle that we tried to attach to the N-terminal barrel. The first three models contain this bundle in different locations. The next two are different C-terminal possiblities without the bundle. ProteinShop was used to tweak the three-helix bundle and to try different locations for it. All models were optimized with undertaker. Model 1 is try56-opt2, the best-scoring model with a cost function that emphasizes removing breaks and clashes. Rosetta also likes this backbone (at least, after doing some optimization with gromacs to remove small clashes and repacking the sidechains). Model 2 is try55-opt2.gromacs0.repack-nonPC, which rosetta likes best after try56-opt2.gromacs0.repack-nonPC (too similar to model 1 to be worth submitting). It has the 3-helix bundle in a totally different place. Model 3 try52-opt2, yet another location for the 3-helix bundle. The linker region is not quite long enough for this position to work. Model 4 is try40-opt2, a non-bundle placement of the C-terminal helices that seems to do an almost reasonable job of burying the hydrophobic faces of the helices, though one would wish for tighter packing. Model 5 is try16-opt2, a different placement of the C-terminal helices. ------------------------------------------------------------