22 May 2002 Kevin Karplus T0129 HI0817 is P44882 in Swissprot This sequence has a few homologs, but all are of unknown structure and function. The predictions from the CAFASP website are all over the map for fold recognition. The SAM-T02 server is giving bogus results, because the method for combining scores in best_scores is flawed (I fixed this in pce/casp5/scripts-human/best_scores) and because selecting alignments by the alignment score (ignoring the combined scores) is risky at best. I have not yet come up with a good way to select the alignments. Looking at the sequence logos I see some conserved glycines (in the second helix and at the end of the 5th helix. There are also two conserved tryptophans (W41 and W102). It might be worth checking predictions to see if these two tryptophans come close to each other. (Maybe I could add a constraint to undertaker to reward them for being close?) The prediction of a single strand (in almost all the 2ry structure predictors) is somewhat strange---perhaps the protein is part of a complex or dimerizes? 6 June 2002 Kevin Karplus I think that packing these helices may work. We may want to subdivide into domains around residue 54, since the short and the long helices may be almost separate subdomains. 7 June 2002 Kevin Karplus I did some runs today reoptimizing from the large pool of decoys already created. I'm still having trouble getting the large helices to form completely, but I'm afraid of turning up the helix-favoring score terms too much, as that might cause the whole thing to form a long helix. I like the first subdomain, but the longer helices are not doing well. I may have to add some weak constraints to try to orient them in a reasonable way. Run try16 is starting to look better. I started from all the existing decoys and did re-optimization. I'm now getting 3 of the 4 long helices and the straight strand, but they aren't packing right. I'm pretty sure they run parallel to each other, making a bundle. Perhaps I need to add constraints to align the helices in the orientations I think they should have, orientation P56-D76 helix4 21 long L86-F80 strand F93-I110 helix5 17 long L136-E120 helix6 17 long E145-F170 helix7 25 long We could put some real weak constraints on P56, F93, L136, E145 to hold one end of the proposed bundle together and on D76, I110, E120, F170 to hold the other end together. 8 June 2002 Kevin Karplus Adding the constraints helped somewhat in try17 , but the second long helix and the strand are still crumpled. Constraint 429 717 -50 10 20 // CA P56 CA F93 Constraint 1042 717 -50 10 20 // CA L136 CA F93 Constraint 583 852 -50 10 20 // CA D76 CA I110 Constraint 1327 852 -50 10 20 // CA F170 CA I110 Maybe we need some more constraints or slightly stronger weight. Let's add P56 L136 and D76 F170, and loosen the distances a bit. 9 June 2002 Kevin Karplus Stronger constraints didn't help with an SSBond in try18. The SSbond is almost (but not quite) formed, but the hoped-for helix bundle is not. Trying again without the SSbond term. 12 June 2002 Kevin Karplus Maybe I should try some constraints to keep the long helices straight (1.5 Angstrom/residue)? I don't need to go all the way to the ends of the helices--just to where the probability drops off sharply: G58-L74 Q95-G109 I124-L137 E145-H169 24 June 2002 Kevin Karplus try23, which started from scratch (using the old fragments and alignments), did not score as well as some of the earlier attempts. I'll have to look at it to see if what it is doing. I can always go back to starting from a selection (or all) of the previously generated decoys. Will start try24 with the NEW fragments from the STR HMM. 25 June 2002 Kevin Karplus Oops, noticed a bug in the constraints. I'm off by 1 in numbering CA Q135 (got N instead). I should rerun after fixing the bug, though it shouldn't make much difference, since I was just using a loose constraint to try to straighten the helices. 26 June 2002 Kevin Karplus With the new constraints, the best-scoring decoy is still T0129-try22-opt.pdb, which was from a run which looked at many decoys initially. Let's try that again for try25, starting from the best 100 decoys. Now try25-opt is the best-scoring, with a small, but noticeable gain over try22-opt (visually they look much the same). Each of the helices look individually good (except for a bend in the final one), but the "strand" is not straight and the last three helices don't pack against the 4th helix. I like the way the first 4 helices pack. This structure REALLY makes me want to go in and manipulate the structure by hand! Let's try making the bundling constraints tighter. Instead of -50, 12, 25, let's make them -50, 8, 15 27 June 2002 Kevin Karplus It looks like run try26 failed before completion. Perhaps I should make another try, pciking up any good starts that have been found. Run try27 had no problems and is beginning to look fairly good. The fifth helix is still messed up. Perhaps now I need to start playing with the constraints, moving the fifth helix (D76-F93) around a bit to make it fit in better. 14 July 2002 Kevin Karplus I tried running undertaker from the best 25 or so conformations with new fragments from the improved fragfinder. The scores improved almost immediately, on the first generation of the first iteration. Initial improvements seem to be mainly clash reduction, but then the number of clashes increases as the score continues to drop, and the Crossover operator becomes the most frequently successful one. In try28-opt, helix 5 is still messed up. It may be that the local minimum around the current solution is too deep to allow fixing the helix, and I may want to try a run without the initial alignments or conformations, just fragments. On the other hand, the bundling is now getting good enough that I might want to try DALI (or other structural aligner) to see if there is anything similar in PDB, so that I could have a better starting point. I was unable to get the DALI web site to respond today, so I sent try28-opt to VAST instead---VAST did not find any structure neighbors. Another possibility is to try to refine the constraints that are inducing the bundling---I may have chosen bad distances that are making it hard for the helices to bundle. Let's try adding a constraint for v60 and L86 (the 4th helix and the "strand"), and make the length constraint for helix 5 punish shortening more severely. I'll try rerunning without the initial conformations (to avoid local minima), but with the alignments and new fragments. Try29 seg-faulted after about 1.3 hours---very frustrating, the scores so far were not very promising. 28 July 2002 Rachel Karchin One of the CAFASP servers, MODELLER, has a nice prediction for this target, with the four long helices forming a clear bundle. The parent for this prediction is 1gz5A. It differs from what we have in our current predictions in that the orientation of helix 4 is reversed. orientation: D76-P56 helix4 21 long F93-I110 helix5 17 long L136-E120 helix6 17 long E145-F170 helix7 25 long the strand is part of a turn connecting D76 and F93 I'm going to work on changing the constraints to favor this conformation and see whether the prediction improves. more on 1gz5A. Some sequence homlogs: 1ggqB, 1eqlA, 1flmA The ligand binding histidines in the three homologs are aligned to lysine 92 and his 141 in 1g5zA, according to FASTA/SAS annotation at: http://www.biochem.ucl.ac.uk/cgi-bin/sas/annotate.cgi In the MODELLER structure these residues are aligned with E141 and V92. V92 is at the N-cap of helix 5 and E141 is at the C-cap of helix 6. 29 July 2002 Rachel Karchin A few other mainly alpha proteins with four-helix bundles from viruses: 2tmvP (tobacco mosaic virus) 1aa7A (influenza virus matrix protein) No changes to constraints yet, but I'm rerunning undertaker (since it's been improved considerably since the last run). 11 Aug 2002 Kevin Karplus I re-ran "make" with the new template library. Still no great scores (best is E-value 9 for 1brwA). The fragment library has been redone, so a new undertaker run may have slightly different fragments to search. I'm not convinced that the 4-helix bundle is better than the 4 helices plus an extended strand. It might be useful to change the "bundling" constraints to try to bring hydrophobics that might line up close to each other. 11 Aug 2002 Rachel Karchin Reran undertaker (try31) with changed constraints -- tried orienting helix 4 (the first long helix) in opposite direction. new best is: T0129.try31+T0129-5cpp-2track-...+T0129-1brwA-2track-local-adpstyle1.pw.a2m.gz:1brwA-scwrl.60.40.pdb Second best is T0129.try31-opt.pdb This run was done BEFORE the fragment library was rebuilt. Fri Aug 16 12:37:22 PDT 2002 Kevin Karplus With the current score function, using try30-constraints, try30-opt-scwrl scores best. The best of the robetta models (with this score function) is robetta1, which packs the helix K121-G138 nicely against N144-F166. I wonder what would happen if we tried optimizing from the robetta models? 17 Aug 2002 Kevin Karplus New best is try32-opt-scwrl. It has clearly borroed a lot from Robetta models, as it has some of the bad peptide planes---but it has fixed a lot of them. Maybe we should reoptimize it, with cross-over encouraged from other models and lots of fragment insertion. We can try turning constraints off, also, to let this one pack on its own now. 17 Aug 2002 Kevin Karplus try33-opt is the new best score, but still has a lot of the robetta structure and some bad breaks. We should probably copy the ends (1-53 and 151-182) from the best structure we've made without robetta (say try17-opt), and let OptSubtree try to get them back into place. Made this as decoys/T0129.cut-and-paste.pdb Starting 2 runs---try34 and try35. try35 has OptSubtree turned way up, so that it can get in before SpecificFragment insertion pulls the stuff back together a different way. Both are failing because the random conformation is better than the cut-and-paste one. Killed them both and started try36 with the random conformation removed. 18 Aug 2002 Rachel Karchin try36-opt-scwrl.pdb has moved helix-6 and helix-7 out of the bundle. best score is ?? I'm going to do a new run, starting with best scoring conformation, which has the desired conformation of a four-helix bundle surrounding a single strand, and add constraints to pack hydrophobic residues selected from the helices to hydrophobic residues selected from the strand. CA F80 CG L74 CA L86 CG L59 CA F80 CG L108 CA L86 CZ F93 CA F80 CD I124 CA L86 CD I137 CA F80 CZ F106 CA L86 CG L148 -- per kk suggestion, wait on this, try cut and paste with try37-opt-scwrl 18 Aug 2002 Kevin Karplus MISTAKE-- the top score was for try33-opt, not cut-and-paste. (there was an emacs backup file for cut-and-paste which was just a copy of try33-opt). Try33-opt has too much of the robetta noise in it for my taste. I have an idea, using a new feature of undertaker (not installed yet) to do a better job of fixing up the cut-and-paste. This is the "no_initial_fill" option for OptConform, which does not try doing a bunch of insertions to fill the initial pool. This should let the OptSubtree operations fix up the huge breaks before the insert fragment operators start being applied. From: Rachel Karchin Date: Sun, 18 Aug 2002 13:12:19 -0700 To: karplus@soe.ucsc.edu Subject: T0129 Kevin, I've put a fair amount of time into T0129, and hoped to spend the day trying out some ideas (and in the process climbing some of the undertaker learning curve). You clearly have come up with new ideas about how to proceed -- and we can't try out stuff at the same time with the current setup. Do you want me to move on to some new target(s)? Is there a window today where I can pick up on T0129? Let me know how I can most make myself useful to the team. -- rk Date: Sun, 18 Aug 2002 14:02:56 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu In-reply-to: <15711.65443.395913.50818@grunt.cse.ucsc.edu> (message from Rachel Karchin on Sun, 18 Aug 2002 13:12:19 -0700) Subject: Re: T0129 I just finished the latest t0129 run, and am rescoring all the decoys. try37-opt-scwrl does not score quite as well as try33-opt (with the current scoring function), but I really like the way it looks (other than N91-A118). I'd be glad to hand T0129 over to you for a while, as I'm time-sharing between 13 targets right now, and am getting quite stressed. I'll probably want to do some more work on it in a a couple of days---getting rid of the last remnants of robetta's bad peptide planes. What I'd probably do next is to cut out a good helix for the N91-A118 region from some other model, and combine it with the rest of try37-opt-scwrl to make a new cut-and-paste model, then optimize that using the new "no_initial_fill" option of OptConform, with OptSubtree turned way up the way it is for try37. If you want to do some ordinary optimization, remember to turn OptSubtree back down to about 0.03. You may want to turn InsertSpecificFragment up to about 5 again. -------------------------------------------------- 18 Aug 2002 Rachel Karchin try37-opt-scwrl does look good,except for helix-5 (roughly from N91 to A118 according to the STR secondary structure prediction). so where to find a good helix-5 for a cut-and-paste? try25-opt has a nice helix-5, but it extends from N89 to L108. The STR prediction shows very weak signal after G109 . . . Not very convincing that the helix should extend to A118. However the N91 to A118 region of try37-opt-scwrl is a very broken helix, so let's try pasting these residues from try25-opt. T0129-try38.cut-and-paste.pdb is try37-opt.scwrl plus N91-A118 from try25-opt. Now starting try38 with parameters per Kevin's suggestions. 19 Aug 2002 Rachel Karchin try38-opt-scwrl is now best scoring and has a good version of helix-5. There is a small piece of helix (P115-K119) that is detached from helix 5, but this is in line with the STR secondary structure prediction, which is predominantly T and G here. Now we need to fix helix 6 (G122-G138), with the same approach. This helix looks good in try25-opt, so new cut and paste is T0129-try39.cut-and-paste.pdb -- this is try38-opt.scwrl with G122-G138 pasted in from try25-opt. new best is try39-opt-scwrl. Helix-4 and Helix-7 are now broken. Also loop between Helix-6 and Helix-7 is broken. I can't find an unbroken Helix-7 in our decoys library, but there is a good one in the Modeller prediction. I'll try one more cut-and-paste run before doing polishing and turning up the reduce breaks parameter. try40.cut-and-paste.pdb is try39-opt-scwrl with E145-F170 from t129-modeller-1gz5A. 20 August 2002 Rachel Karchin try39-opt-scwrl is still scoring better than the best decoy from try40 (try40-opt-scwrl) and Helix-7 is still broken. what operations are dominating the score function? from correlation plots: dry6.5 Next run will begin with try39-opt-scwrl and use polishing parameters suggested by Kevin. New best is T0129.try41-opt. What is dominating the score? dry5 break Helix-4 and Helix-7 are still broken. undertaker-try41.log shows that JiggleSubtree and OptSubtree are dominating the operations. Reduce Break is not getting used. I'll adjust initial weights and run again. From karplus@bray.cse.ucsc.edu Tue Aug 20 16:55:09 2002 Date: Tue, 20 Aug 2002 16:55:06 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: T0129 Forget ReduceBreak---it doesn't seem to work on any of the targets. The way to shrink gaps right now with with JiggleSubtree, Optsubtree and CloseGaps. We'll have to work on creating better break-reduction operators. From rachelk@grunt.cse.ucsc.edu Tue Aug 20 17:01:05 2002 MIME-Version: 1.0 From: Rachel Karchin Content-Type: text/plain; charset=us-ascii Date: Tue, 20 Aug 2002 17:01:02 -0700 To: Kevin Karplus Subject: Re: T0129 In-Reply-To: <200208202355.g7KNt6909410@bray.cse.ucsc.edu> X-SpamBouncer: 1.4 (10/07/01) X-SBPass: NoBounce X-SBClass: OK Okay -- I'm about to start try42. I'll adjust parameters. How high should the three below be set? -- rk Kevin Karplus writes: > > Forget ReduceBreak---it doesn't seem to work on any of the targets. > The way to shrink gaps right now with with JiggleSubtree, Optsubtree > and CloseGaps. We'll have to work on creating better break-reduction > operators. From karplus@bray.cse.ucsc.edu Tue Aug 20 17:07:17 2002 Date: Tue, 20 Aug 2002 17:07:15 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu Cc: karplus@soe.ucsc.edu Subject: T0129 It looks to me the try41-opt is beginning to look pretty good. The helix E120-L137 seems to be rotated a bit. Perhaps G111-K119 needs to be moved away and perhaps a weak constraint added for L99-V128? From rachelk@grunt.cse.ucsc.edu Tue Aug 20 17:12:58 2002 MIME-Version: 1.0 From: Rachel Karchin Content-Type: text/plain; charset=us-ascii Date: Tue, 20 Aug 2002 17:12:56 -0700 To: Kevin Karplus Subject: Re: T0129 In-Reply-To: <200208210007.g7L07FF10090@bray.cse.ucsc.edu> X-SpamBouncer: 1.4 (10/07/01) X-SBPass: NoBounce X-SBClass: OK Kevin Karplus writes: > > It looks to me the try41-opt is beginning to look pretty good. The > helix E120-L137 seems to be rotated a bit. Perhaps G111-K119 needs to > be moved away and Yeh, this piece is not really predicted as helix -- but where should it go, and how does one move stuff with undertaker? > perhaps a weak constraint added for L99-V128? According to rasmol, these 2 residues are 17.36 A apart right now. What do you mean by weak constraint? -- rk From karplus@bray.cse.ucsc.edu Tue Aug 20 17:51:00 2002 Date: Tue, 20 Aug 2002 17:50:57 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: <15714.55358.992757.448533@grunt.cse.ucsc.edu> (message from Rachel Karchin on Tue, 20 Aug 2002 17:01:02 -0700) Subject: Re: T0129 Leave ReduceBreak around 1 to 5, JiggleSubtree and OptSubtree around 0.03 to 0.5, CloseGaps around 0.1 to 0.2. Make InsertFragment and InsertSpecificFragment be around 5-10. To move things around, I wrote little single-use perl scripts--see t0130/move*perl t0139/move-residues.perl t0141/move*perl t0156/move-residue.perl You could also just cut-and-past that region in from a different decoy. A weak constraint might be something like constraint L99.CG V128.CB 2 4 10 (The large "max distance" makes the well rather shallow.) 20 August 2002 Rachel G111-K119 will be moved $x-5 $y-5 $z-8 Wed Aug 21 11:10:29 PDT 2002 Kevin Karplus There was a typo in define-score.script that caused try42 not to run (the try42.constraints file name was mistyped). I'm running try42 again. 19:51 Kevin Karplus The try42 run was not very successful---try41-opt still scores better, even with the try42.constraints. We can try once more to optimize from try42, starting from try42.1.40 and try42.7.40, and allowing a lot more insertfragment. From karplus@bray.cse.ucsc.edu Thu Aug 22 10:28:38 2002 Date: Thu, 22 Aug 2002 10:28:36 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu Cc: karplus@soe.ucsc.edu Subject: T0129 I think it is time to try optimizing try43-opt-scwrl without constraints (or with only helix length constraints). Will you set this up and run it? We'll probably then want to take the best models from several lineages, and do a run with CrossAndInsert set high, to see if we can get any useful crossover. Then we'll have to pick the best 2-5 different models to submit to CASP5, since it is due by Monday. If there is time, and we have a reasonably good backbone to work from, we could turn up OneRotamer to try to pack the residues better. From karplus@bray.cse.ucsc.edu Thu Aug 22 10:30:38 2002 Date: Thu, 22 Aug 2002 10:30:36 -0700 From: Kevin Karplus To: karplus@soe.ucsc.edu CC: rachelk@soe.ucsc.edu In-reply-to: <200208221728.g7MHSaF00560@bray.cse.ucsc.edu> (message from Kevin Karplus on Thu, 22 Aug 2002 10:28:36 -0700) Subject: Re: T0129 You could toss try43.3.50 try43.9.50 try43.6.50 ... into the next run also, as they score better than try43-opt when read in again. 23 August 2002 Rachel Karchin try44 dumped core. bug fixed by Kevin. 23 August 2002 Rachel Karchin Try45 (try43-opt-scwrl optimized with only helix length constraints)-- best scoring is T0129.try45-try43-3.8.50.pdb This looks good, except that the strand in the middle of the bundle is now broken. Other breaks seem mainly to be in loop regions on this one. Problem: this decoy has the C-terminal short helix poking into the bundle. next best are: T0129.try45-try43-3.1.50.pdb T0129.try45-try43-2.4.50.pdb T0129.try43.3.50.pdb 23 August 2002 Kevin Karplus Since Rachel seems to have given up for the night, I'll do a longer run with a slightly tweaked score function (to move the local minima around a bit), to see if anything happens. The new score function mainly increases the penalty for breaks and clashes. 24 August 2002 Rachel Karchin try46 is still running. todo: run with C-terminal helix moved out of the bundle with cut and paste? current best new decoys: try46-opt.pdb --broken strand, c-terminal helix in the way, broken helix 4 try46.4.100 -- broken strand and c-terminal helix in the way From rachelk@grunt.cse.ucsc.edu Sat Aug 24 08:59:14 2002 MIME-Version: 1.0 From: Rachel Karchin Content-Type: text/plain; charset=us-ascii Date: Sat, 24 Aug 2002 08:59:12 -0700 To: karplus@soe.ucsc.edu Subject: T0129 X-SpamBouncer: 1.4 (10/07/01) X-SBPass: NoBounce X-SBClass: OK I can work on it today. How do you want to coordinate efforts? I can do some model inpsection to pick good ones. You've mentioned using multiple good models and using CrossAndInsert. Then perhaps, a polishing run. -- rk From karplus@bray.cse.ucsc.edu Sat Aug 24 09:36:47 2002 Date: Sat, 24 Aug 2002 09:36:45 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: <15719.44368.824898.454320@grunt.cse.ucsc.edu> (message from Rachel Karchin on Sat, 24 Aug 2002 08:59:12 -0700) Subject: Re: T0129 You work on t1029 today---I'll look at your notes tomorrow (probably afternoon), and make the final decisions. 24 August 2002 Rachel Karchin I'm going to do a cut-and-paste run, with try46-opt and the C-terminal helix moved out into outer space (just slid along the z-axis). What would be really nice here is an operator that would rotate this piece 180 degrees. $z+20 for 170-182 I'm also trying a search for structural neighbors of: try45-try43-3.8.50 (unbroken central strand) and try46-opt using the DALI server. Nothing back from dali yet. On try47, best scoring decoy is still try46-opt try47-opt is number four. The c-terminal helix is still poking into the bundle, but the central strand is unbroken on this one. I'm going to do the CrossAndInsert run (try48) , selecting a few of the best models: try46-opt try47-opt try43.8.50 try25-opt try45-try43-3.8.50 From: Dali Server To: rachelk@soe.ucsc.edu Subject: Dali-Server Date: Sun, 25 Aug 2002 02:10:39 +0100 FSSP FAMILIES OF STRUCTURALLY SIMILAR PROTEINS, VERSION 1.0 (Apr 1 1995) CREATED Sun Aug 25 02:10:34 BST 2002 for dali on sputnik2-node55.ebi.ac.uk METHOD Dali ver. 2.0: Holm, L., Sander, C. (1993) J.Mol.Biol. 233,123-138 DATABASE 3241 protein chains PDBID 4445-A HEADER COMPND SOURCE AUTHOR SEQLENGTH 182 NALIGN 9 WARNING pairs with Z<2.0 are structurally dissimilar ## SUMMARY: PDB/chain identifiers and structural alignment statistics NR. STRID1 STRID2 Z RMSD LALI LSEQ2 %IDE REVERS PERMUT NFRAG TOPO PROTEIN 1: 4445-A 4445-A 36.6 0.0 182 182 100 0 0 1 S 2: 4445-A 1fpo-A 3.8 4.5 86 171 5 0 0 8 S chaperone protein hscb (hsc20) Mutant 3: 4445-A 1br0-A 2.9 3.7 78 120 5 0 0 7 S syntaxin 1-a fragment Mutant 4: 4445-A 1quu-A 2.6 4.1 73 248 7 0 0 6 S human skeletal muscle alpha-actinin 2 fragment 5: 4445-A 1qu2-A 2.5 4.0 96 917 7 0 0 8 S isoleucyl-trna synthetase (isoleucine--trna ligase, il 6: 4445-A 1ocr-C 2.5 9.3 83 261 6 0 0 7 S cytochrome c oxidase (ferrocytochrome c:oxygen oxidored 7: 4445-A 1brw-A 2.3 3.7 68 433 13 0 0 3 S pyrimidine nucleoside phosphorylase (pynp) 8: 4445-A 1ile 2.2 3.8 82 821 9 0 0 8 S isoleucyl-trna synthetase (ilers) biological_unit 9: 4445-A 1htj-F 2.1 3.7 85 182 12 0 0 9 S kiaa0380 fragment Mutant 1fpo not in t2k library or our pdbaa file (so can't make alignments with our existing scripts) 1br0 " " 1quu " " 1qu2 " " 1ocr " " [[CORRECTION: 26 Aug 2002 Kevin Karplus The chains are there---the names are in uppercase in pdbaa, so the grep search didn't find them. You need grep -i.]] 1brwA already found by SAM-T2K 1ile is in t2k library and I'm now building pairwise alignments to it. 1htjF " " Want to do another try reading in these alignments and seeing if they create some good decoys. 1fpo with RMSD 4.5 A from T0129 has a completely different structure. It has a helix-turn-helix motif pointing out at a right angle from a bundle of 3 helices. 1br0 is an NMR structure. It is a three-helix bundle of very long helices. Again, nothing like our existing decoys for T0129. from try48, best scorers are now: T0129.try48-try46.4.20.pdb lots of breaks T0129.try48-try46.2.20.pdb lots of breaks maybe I'm just tired, but these last two look a bit better packed to me in spacefill mode than try46-opt. to try and get some new stuff into the pool, try49 will attempt to use the new alignments built from 1ile 1htjF and old alignments from 1brwA try49 is still running current best is try49-try48.6.20. seems to have fewer breaks and has an unbroken central strand. try50 should be a polishing run 25 August 2002 polishing run -- parameters copied from T0131 (another target with low CAFASP consensus, "hand model", which was submitted already) Date: Sun, 25 Aug 2002 21:40:37 -0700 From: Kevin Karplus To: rachelk@soe.ucsc.edu Subject: T0129 You chose a rather long polishing run for T0129. (T0131 was only half as long, so ran much faster). I was hoping to see some notes from you on which DIFFERENT models are worth submitting. I assume we'll pick the best scoring one (with some socring function) as the first model, but what other different topologes do we have that are any good? I'd like to submit the 5 best representatives of the "most likely" folds, and I haven't been looking at them enough to know which ones are reasonable, and which are just minor variations on the same idea. I expect that try50 will be done by 10 or 10:30 tonight. If you're still awake and on-line then, could you give me a list of the "best" different folds for T0129. (I'll probably submit the results of try50, unless it is terrible, as teh first model---what I want are ideas for models 2-5.) ------------------------------------------------------------ 25 Aug 2002 22:15 Kevin Karplus The polishing run try50 just finished. I'm rescoring everything with NO constraints (not even helix length), to try to choose the number 1 model. I hope that Rachel will choose numbers 2-5 for us tomorrow. The best score is try50.3.100. It looks like the polishing run did too many JiggleSubTree and not enough CloseGap or InsertFragment. I'll do another short polishing run starting with the best results from each run from try33 (arbitrarily chosen point) on, with CrossAndInsert set high to make it possible to swap partial solutions. From: Rachel Karchin Date: Mon, 26 Aug 2002 08:32:56 -0700 To: Kevin Karplus Subject: Re: T0129 Kevin, the 25 or so models that I have downloaded onto my local machine are very much variations on the same theme. I'm not sure that we have several distinct "good" topologies on this one. I can come up to school later this morning and attempt to look at some of the over 1200 decoys we've got for T0129, in search of some distinct variations, but over my ISDN line this is too many files to download and view locally. -- rk ------------------------------------------------------------ 26 Aug 2002 Kevin Karplus The best scoring is try51.1.80. I think we've gotten a bit wedged into a local minimum. I'll submit this one, but it's tood bad that we could not explore more of conformation space---we spent too much time polishing this fold, which is probably not quite right. In particular, I don't like the way the helix L148-F166 is positioned---I think that stripe of E150,E153,E154,E157 should clearly be on the surface, not at the bottom of a cleft. Raising the wet6.5 weight should help look for a solution which does not bury the glutamic acids so thoroughly. Hmm---try51.1.80 still scores best, even with wet6.5 set quite high. For a somewhat different one, how about try28-opt? or try25-opt? There is also the robetta-based try32.0.60, but I'd rather not use anything that still has the obvious marks of robetta (T0129 had incorrectly rebuilt peptide planes in robetta). Submitted try51.1.80 as first model. I'll wait to hear from Rachel to submit other models. From: Rachel Karchin Date: Mon, 26 Aug 2002 13:02:28 -0700 To: karplus@soe.ucsc.edu Subject: T0129 I like these two: try49-try48.6.20 (has unbroken central strand) try25-opt (we've borrowed several helices from this one) -- rk ------------------------------------------------------------ Mon Aug 26 13:37:52 PDT 2002 Kevin Karplus model2 try25-opt model3 try28-opt model4 try49-try48.6.20 26 November 2002 Kevin Karplus The best model for the whole chain is try37-opt-scwrl at 11.4905 RMSD (10.9273 CA). RMSD whole chain _1 _2 best decoy 11.4905 8.37342 8.26092 model1 try51.1.80 14.1133 11.6593 13.4236 model2 try25-opt 15.1425 12.2253 16.8285 model3 try28-opt 13.3083 10.4632 9.32523 model4 try49-try48.6.20 14.0837 11.5761 13.4475 28 November 2002 Kevin Karplus Looking at superpositions, it looks like the best overall model is model 3, which is also best for domain 1 (1-77) and domain 2 (78-171). The first domain was right except for the first helix, which did not bundle with the second and third as we predicted.