Thu May 25 08:58:54 PDT 2006 T0301 Make started Thu May 25 08:59:07 PDT 2006 Running on shaw Thu May 25 16:18:52 PDT 2006 Kevin Karplus The best-scores.rdb file lists a lot of 2-domain d.21.1.2 domains, with 2gkeA scoring best, even though BLAST gets no hits. The try1 run picks out 2g6yA as its favorite template (after initially favoring 2azpA). Interestingly, neither of these score particularly well in the fold-recogniton. 2azpA appears in undertaker-align (as the 10th model), and 2g6yA scores poorly enough not to even appear in undertaker-align. Is try1 picking up something good here? or is it getting confused into picking up trash? Thu May 25 18:45:37 PDT 2006 Kevin Karplus The try1-opt1 model looks pretty good, but the C-terminal domain is better modeled than the N-terminal domain. Perhaps we need to do subdomains. M-D183 and F178-F395 might be good subdomains. Thu May 25 18:50:39 PDT 2006 Kevin Karplus OK, I started the two subdomain predictions. With any luck, we'll get good models for each and will be able to superimpose them on try1-opt2 and make a chimera. Thu May 25 21:46:43 PDT 2006 Kevin Karplus The M1-D183 prediction gets somewhat weak hits (best is 2gkeA at E-value 0.085), but they are still to d.21.1.1 domains. Fri May 26 11:57:54 PDT 2006 Kevin Karplus It looks like it might be worth making a chimera with M1-P150 copied from the M1-D183 as well as copying sheet constraints from M1-D185 (at least through P150). I made a chimera1 which has M1-V149 from the M1-D183 try1-opt2 and the rest from the whole chain try1-opt2. For the C-terminal half, it is not clear whether the subdomain model is better or worse than the domain in the whole model. Fri May 26 13:39:00 PDT 2006 Kevin Karplus I'll make a chimera2 that has the whole model up to D200, then from F178-F365 try1-opt for L201 on. I'll start try2 with sheet constraints from the two subdomain models. Note: I had to fix a sheet constraint from the second model, as undertaker occasionally seems to take a hairpin and convert it into one long sheet constraint. I wonder where *that* bug is. Fri May 26 14:22:35 PDT 2006 Kevin Karplus Fixed bug in undertaker so that hairpins no longer get merged into long sheet constraint, but they may still have one pair too many residues. Fri May 26 14:48:45 PDT 2006 Kevin Karplus I picked up the templates that any of the try1 runs used (whole protein or subdomains) and put them in MANUAL_TOP_HITS and made alignments for them (some of which had not come up in the whole-chain top hits). I made the alignments that were missing and made read_alignments to get undertaker scripts for reading the alignments. For try3 I'll use the same costfcn as try2, but start from alignments rather than from the chimeras. Fri May 26 16:23:44 PDT 2006 Kevin Karplus For try4, I should do the same as try3, but add in all alignments from the subdomains. Fri May 26 18:05:19 PDT 2006 Kevin Karplus try2-opt1 looks pretty good, so I expect try2-opt2 to do well. try3-opt1 looks like it is really lost---not much hope for it. The scores for try4 look comparable to those for try3, so I expect it to do poorly also. Fri May 26 21:46:42 PDT 2006 Kevin Karplus Sure enough try2-opt2 looks pretty good, and try3 and try4 look terrible. Tue May 30 09:10:33 PDT 2006 Kevin Karplus I accidentally stomped on the Makefile and README file this morning, using new-target, but I restored them from my computer at home. I need to fix new-target to warn people before making changes! Sat Jun 10 09:00:20 PDT 2006 Kevin Karplus I'm scoring the models with unconstrained.costfcn, and try2-opt2 still scores the best, though repacking does reduce clashes with only a small total increas in cost (mainly from the "dry" packing terms). I should probably do a polishing run on try2, to try to close gaps, but I'll have to be careful not to separate the two domains. Sat Jun 10 09:52:12 PDT 2006 Kevin Karplus I took the sheet constraints from try2-opt2 (except for from F178-F395/decoys/try1-opt2.sheets SheetConstraint A346 H351 G359 G354 hbond F349 1 SheetConstraint A360 S363 K373 T370 hbond E361 8 ) and added some constraints on H57 to hold the domains together: Constraint H57.NE2 H313.ND1 -10 5 8 1 Constraint H57.NE2 H314.NE2 -10 6 9 1 Constraint H57.NE2 H351.ND1 -10 9 14 1 to make try5.costfcn, which scores try2-opt2.repack-nonPC best. I'll do a polishing run for try5 on orcas, including fragments (and a few alignments) from the subdomains as well as the whole protein. Sun Jun 11 21:01:16 PDT 2006 Kevin Karplus try5-opt2 now scores best (both with try5.costfcn and Rosetta repacking), but still has some pretty bad breaks: T0301.try5-opt2.pdb.gz breaks before (T0301)K70 with cost 9.41183 T0301.try5-opt2.pdb.gz breaks before (T0301)P181 with cost 7.46926 T0301.try5-opt2.pdb.gz breaks before (T0301)S69 with cost 5.0552 T0301.try5-opt2.pdb.gz breaks before (T0301)T355 with cost 3.1167 T0301.try5-opt2.pdb.gz breaks before (T0301)K264 with cost 1.83677 T0301.try5-opt2.pdb.gz breaks before (T0301)R248 with cost 1.81326 T0301.try5-opt2.pdb.gz breaks before (T0301)Q364 with cost 1.69404 T0301.try5-opt2.pdb.gz breaks before (T0301)I131 with cost 1.50402 T0301.try5-opt2.pdb.gz breaks before (T0301)A346 with cost 1.47929 T0301.try5-opt2.pdb.gz breaks before (T0301)H82 with cost 1.41118 T0301.try5-opt2.pdb.gz breaks before (T0301)V156 with cost 1.01317 I'll try yet another polishing run, with breaks turned up even higher. (try6 running on lopez) I'll also score the server models, which we have not done yet. I'll use the try6 costfcn, which has no constraints, but has high break and clash costs. Sun Jun 11 21:32:31 PDT 2006 Kevin Karplus 5 servers score better than try5-opt2 on the try6 costfcn: ROBETTA_TS 5,1,4 PROTINFO_TS1 RAPTORESS_TS1 All look pretty similar at first glance. Tue Jun 13 22:16:40 PDT 2006 Kevin Karplus I adjusted the cost function so that try6-opt2 scored a little better than the best server model (now ROBETTA_TS5) and am starting a polishing run (with high crossover) including all the models---including the server models. I'm hoping the the crossover will pick up the better parts of different models and make something that is overall good. (running as try7 on orcas) Wed Jun 14 07:31:10 PDT 2006 Kevin Karplus try7 crashed, because of 1uocA:undertaker: ChainsResiduesAndAtoms.cc:1185: void ChainsResiduesAndAtoms::add_chain_from_pdb(const char*, void*, std::vector >, char, int): Assertion `biomt.size() == trans_number-1' failed. Since that assertion is not *necessary* for anything, I will remove it. I will have to look at 1uoc.pdb.gz and figure out what the correct assertion is. Wed Jun 14 09:17:59 PDT 2006 Kevin Karplus try7 crashed again, with a less useful assertion failure: # iteration 0 # filling pool to have 531 conformations undertaker: XYZpoint.h:57: void XYZpoint::unitize(): Assertion `mag>0' failed. This probably results from a bad model from one of the servers. Tracking down the problem could take a while. Maybe I should eliminate any server models that haven't a chance of being involved in the crossovers, since the very high-cost models are the ones most likely to have serious flaws that would crash undertaker. Thu Jun 15 00:04:43 PDT 2006 Kevin Karplus Try7 crashed again: # iteration 1 # filling pool to have 449 conformations undertaker: Segment.cc:85: int Segment::OK() const: Assertion `N_atoms[0] != N_atoms[1]' failed. undertaker is not set up to handle models as bad as the server models! I'll clean up try7 to include only a few of the top servers. Thu Jun 15 06:44:19 PDT 2006 Kevin Karplus try7-opt2 cleaned up try6-opt2 a little bit, but *still* has bad breaks: T0301.try7-opt2.pdb.gz breaks before (T0301)K70 with cost 10.4123 T0301.try7-opt2.pdb.gz breaks before (T0301)S182 with cost 5.83876 T0301.try7-opt2.pdb.gz breaks before (T0301)S69 with cost 3.99813 T0301.try7-opt2.pdb.gz breaks before (T0301)T355 with cost 3.09077 T0301.try7-opt2.pdb.gz breaks before (T0301)K264 with cost 1.81185 T0301.try7-opt2.pdb.gz breaks before (T0301)R248 with cost 1.7879 T0301.try7-opt2.pdb.gz breaks before (T0301)Q364 with cost 1.48286 T0301.try7-opt2.pdb.gz breaks before (T0301)I131 with cost 1.42886 T0301.try7-opt2.pdb.gz breaks before (T0301)H82 with cost 1.35671 T0301.try7-opt2.pdb.gz breaks before (T0301)V156 with cost 0.987475 T0301.try7-opt2.pdb.gz breaks before (T0301)N366 with cost 0.918614 T0301.try7-opt2.pdb.gz breaks before (T0301)A346 with cost 0.822846 try8 started on cheep, using ONLY server models (not any of the try*models), in the hopes that we can produce a polished model that is slightly better than the ROBETTA models. Thu Jun 15 07:28:24 PDT 2006 Kevin Karplus try9 started on peep, using only server models, but specifically excluding the ROBETTA models, to get a different solution. The cost function was also tweaked a bit to get more polished models. Thu Jun 15 07:44:54 PDT 2006 Kevin Karplus Prepared the first submission with the following text added to the method file: For T0301, we did separate quick predictions of two subdomains: M1-D183 and F178-F395. We made a chimera of the two, crossing over betweeen D200 and L201. Optimization of this chimera did not quite succeed in closing all the gaps, the worst of which (before K70) would require extensive loop redesign to close. We would probably have done better to polish the two subdomains separately a bit before creating the chimera. Several of the server models scored better with our cost functions than all but the most polished of our own models, so we tried polishing up the better server models for some of our submissions. There is considerable agreement between our model and the best server models, with the differences mainly in loop closure and surface rotamers. Model 1 is try7-opt2, our best-scoring model refined from the chimera of the two domains. Model 2 is ROBETTA_TS5, to be replaced by a model polished from several servers. Model 3 is PROTINFO_TS1, the top-scoring non-ROBETTA server model in our cost function. It will be replaced by a model polished from several of the top-scoring models, excluding the ROBETTA models. Model 4 is the automatic prediction generated for domain M1-D183. Model 5 is the automatic prediction generated for domain F178-F395. Thu Jun 15 07:47:09 PDT 2006 Kevin Karplus Preliminary submission mailed. try8 seems to be optimizing ROBETTA_TS5 as expected. try9 seems to be optimizing RAPTORESS_TS1 which has remarkably few clashes. Thu Jun 15 08:52:49 PDT 2006 Kevin Karplus try9 is increasing the clashes but decreasing breaks in RAPTORESS_TS1. CrossOver is not having much effect, but the gap closing operators and the tweaking operators seem to be making modest improvements. try8 is similarly working only on ROBETTA_TS5, reducing both clashes and breaks (though clashes and breaks remain much higher than in try9). Thu Jun 15 09:22:15 PDT 2006 Kevin Karplus The try8 costfcn prefers try8-opt2, try7-opt2, try6-opt2, try5-opt2, try9-opt2.repack-nonPC The try9 costfcn prefers try8-opt2, try7-opt2, try9-opt2.repack-nonPC Since try7 is *ours*, I'll put it first, even though the next two are probably better. Thu Jun 15 09:34:47 PDT 2006 Kevin Karplus I've resubmitted with try7-opt2 try8-opt2 # from ROBETTA_TS5 try9-opt2.repack-nonPC # from RAPTORESS_TS1 M1-D183 try1-opt2 F178-F395 try1-opt2 try8 and try9 agree fairly closely (except on rather disordered loops) except for the alignment of E343-A365 and R25-V72. Strand D301-L307 is also shifted by one residue. They are clearly based on different alignments to the same template. I think that the thing to do next is to do further optimization of the two domains separately, then remake the chimera and reoptimize. Thu Jun 15 09:56:04 PDT 2006 Kevin Karplus In M1-D183, I did make extra_alignments and make all-align.* to update some of the missing alignments. I started try2 (with constraints from try1-opt2) on orcas. Thu Jun 15 10:07:40 PDT 2006 Kevin Karplus I started F178-F395/try2 on orcas also. Thu Jun 29 11:02:58 PDT 2006 Kevin Karplus I don't like the F178-F395/try2 I think it has messed up E227-A270. The M1-D183/ try2-opt2 may be slightly better than the try1-op2, but I'd have to see them in context to be sure. Hmm---looks ok out to about G153, maybe E158, but after that it conflicts with the second domain. The M1-D183/ try1-opt2 has similar problems, though not so severe. The ROBETTA and RAPTORESS alignments (in try8 and try9) have a more reasonable connection between the domains for E158-F178 I'll make a new chimera: M1-D183 try2-opt2 M1-E158 try8-opt2 E158-F178 try7-opt2 F178-D199 F178-F395 try2-opt2 D199-A225 try7-opt2 A225-P276 F178-F395 try2-opt2 P276-T372 try7-opt2 T372-F395 No, this is not compatible. try8-opt2 puts E177 where F178-F395 try2-opt2 puts D200. There are about 32 extra residues that have to go somewhere. Thu Jun 29 12:57:49 PDT 2006 Kevin Karplus How about select (1-158 and */4) or (158-170 and */2) or (170-178 and */5) or (178-195 and */6) or (195-225 and */7) or (225-276 and */1) or (276-372 and */7) or (372-395 and */1) from the current best-models.pdb.gz 4=M1-D183/try2-opt2 2=try8-opt2 5=M1-D183/try1-opt2 6=F178-F395/try1-opt2 7=F178-F395/try2-opt2 1=try7-opt2 This is still not completely consistent, but undertaker may be able to rearrange things into a consistent model. I may also want to do a subdomain for D165-F395, to see if there is more C-terminal domain that I am missing. Thu Jun 29 13:14:06 PDT 2006 Kevin Karplus I created D165-F395/ and started it on the farm cluster with para-trickle-make -quick 'make -k >& make.log; gzip -9f make.log' Thu Jun 29 13:28:15 PDT 2006 Kevin Karplus I created chimera-many.pdb.gz from REMARK 44 1-158 from M1-D183/T0301.try2-opt2.pdb REMARK 44 159-169 from T0301.try8-opt2.pdb REMARK 44 170-177 from M1-D183/T0301.try1-opt2.pdb REMARK 44 178-194 from F178-F395/T0301.try1-opt2.pdb REMARK 44 195-225 from F178-F395/T0301.try2-opt2.pdb REMARK 44 226-275 from try7-opt2 REMARK 44 276-371 from F178-F395/T0301.try2-opt2.pdb REMARK 44 372-395 from try7-opt2 Thu Jun 29 13:41:21 PDT 2006 Kevin Karplus chimera-many has lots of clashes and breaks, but I will try optimizing it anyway, to see if it can make a decent fold after cleaning up the problems. Thu Jun 29 13:48:20 PDT 2006 Kevin Karplus try10 to polish chimera-many started on the farm cluster. Thu Jun 29 22:20:46 PDT 2006 Kevin Karplus try10 failed to clean up F169-S182, so the try10-opt2 model is fairly useless. Maybe we need to Proteinshop that region to make a strand and a loop. Maybe rotating the two domains with respect to each other would make room for a longer beta strand. From: "Firas Khatib" To: "Kevin Karplus" Subject: Re: T0301 Cc: thiltgen Not yet, I am coming up to lab this afternoon (around 3pm) to work on 301 & 311. Can you mention in the README which model you think is the best one to start with Proteinshop? I was going to attempt with try7 and maybe try9, is that right? --Firas On 7/4/06, Kevin Karplus wrote: > Anything done yet on T0301? I don't see any new README notes. Date: Tue, 4 Jul 2006 11:55:36 -0700 From: Kevin Karplus To: bort CC: thiltgen Subject: Re: T0301 I don't know what model is best to ProteinShop. You might want to look at chimera-many.pdb try7-opt2 is ours, so is probably more honest to start with than try8 (based on ROBETTA_TS5) or try9 (based on RAPTORESS_TS1). Those three models: try7, try8, and try9 are probably our best bets as the basis for the whole protein. Kevin ---------------------------------------------------------------------- Tue Jul 4 15:06:01 PDT 2006 Firas Khatib Grant and I are going to Proteinshop try7,8,and 9 now Tue Jul 4 15:42:11 PDT 2006 Grant Thiltgen I'm looking at using proteinshop to fix try7 right now and trying to get the helix into a sheet in that region seems like it could end up ruining the protein more than fixing it. The main problem appears to be the helix at the end of the last beta strand. While I'm attempting to fix it using proteinshop, I am going to start try11. Try11 is going to work on clearing he helix out of the end of the barrel at the end of the protein. I am also going to keep all the strand constraints, but add a higher constraint to get that region to form a strand in the middle of the protein. Try11 started on whidbey Tue Jul 4 16:07:20 PDT 2006 Firas Khatib I am trying to fix the sheet (around residue 174) on try8 with Proteinshop. I took Robetta's model 5 (so that I could see hbonds) and tried to do this. Proteinshop has a lot of trouble flattening the sheet with destroying the rest of the chain. The only wasy I was able to flatten it and get some hbonds was to chop of residues 1-169 and then make a sheet from 170-176 and try to line it up with the sheet next to it: 213-216. I then catted this model with the 1-169 half of Robetta5 and am going to run Undertaker to try to heal the gap and I will add strand constraints in hope that this will help. this will be try12 running on vashon Tue Jul 4 17:03:44 PDT 2006 Kevin Karplus I'm not sure what you mean by "flattening the sheet". The existing sheet was good and should not be mucked with. The problem is attaching the edge strands in a way that allows the two halves to be a contiguous model. Tue Jul 4 18:39:27 PDT 2006 Firas Khatib Grant was working on try7 with Proteinshop and made this in the Proteinshop directory: T0301.try7-opt2-newsheet.pdb I tried to straighten the two strands at 174 & 179 and will see if Undertaker can save it because it has MANY clashes. My attempt is called salvageNewSheet.renum.pdb and will use try11's constraints and will run as try13. Hopefully try11 or try13 will work, since try12 will probably have messed up Robetta's model 5. I changed the following sheet constraints from try11: SheetConstraint (T0301)V149 (T0301)K143 (T0301)E174 (T0301)D180 hbond (T0301)H148 10 SheetConstraint (T0301)E174 (T0301)D180 (T0301)S377 (T0301)V371 hbond (T0301)E174 10 to SheetConstraint (T0301)T144 (T0301)H148 (T0301)A172 (T0301)L176 hbond (T0301)H148 10 SheetConstraint (T0301)A172 (T0301)L176 (T0301)A374 (T0301)T370 hbond (T0301)I174 10 try13 is running on shaw Date: Tue, 4 Jul 2006 17:57:47 -0700 From: "Firas Khatib" To: "Kevin Karplus" Subject: Re: T0301 Cc: thiltgen Hey Kevin, I am afraid that Grant and I are confused about how to improve try8. If you open try8-opt2 and look at the residues that you had mentioned in the README (clean up F169-S182). In rasmol F169-S182 does not look like a nice sheet, it's kind of bumpy, and we understood that by making that into a nice sheet it would make a better barrel. Can you specify exactly which strands you are referring to when you say: "The problem is attaching the edge strands in a way that allows the two halves to be a contiguous model." Sorry about this misunderstanding. We are both in lab if you want to call us so it can be crystal clear what it is we should be doing! 459-1023 ---Firas Date: Tue, 4 Jul 2006 18:50:49 -0700 From: Kevin Karplus To: bort CC: thiltgen Subject: Re: T0301 --text follows this line-- Sorry about the confusion. I have to go to dinner at a friend's house tonight, so I don't have time to try to figure out what to do with T0301. I'll try to look at it again tonight and when I get back. We can get it fixed tomorrow, maybe. I remember that there was a longer predicted strand (or pair of strands) in our target than in the alignments we were getting. I was thinking that we could make the first half of that become the edge strand of the first sheet, and the second half become the edge strand of the second sheet. It may require rotating the second domain to have enough room to do this. (Current models had insertions that had to hang out somewhere, but rotating the second domain might allow the insertion to stretch out ---not straight, but following the curve of the existing sheets. Kevin ------------------------------------------------------------ Wed Jul 5 16:58:59 PDT 2006 Firas Khatib well, try11, try12 and try13 are garbage. I will try to take try8 into Proteinshop and get 286-294 to form a sheet with the 343-347 strand! Grant will start from try7 and try to fix that one. I managed to get the yellow 286-294 area moved, this is called try8MoveCrappyYellow.try.renum.pdb and I will input it as try14 using try8's constraints. Wed Jul 5 18:48:18 PDT 2006 Firas Khatib try14 has these added constraints to the try8.helices and try8.sheet #these are the sheet constraints that I tried to line up in Proteinshop SheetConstraint G342 R348 A297 L303 hbond L303 10 hbond A297.O G342.N 3 SheetConstraint A290 P284 A297 L303 hbond A290 10 I hope they work! Try14 is running on shaw Wed Jul 5 18:54:55 PDT 2006 Grant Thiltgen I adjusted the 1w61A alignment by hand. I aligned residues 368-395 of the target to the last residues of the 1w61A chain. I did this to try to get the last strand in the barrel that we weren't getting. I am going to start a new run, try15. I am hoping that it will be able to put in the alignments that are missing. I am also adding in sheet constraints to the regions of interest to make sure that they are going to align properly. Wed Jul 5 19:01:10 PDT 2006 Firas Khatib so, for clarity, try14 is based off Robetta_5 and tried to move an insert to form a nice sheet (286-294 with 343-347) If it is somewhat successful, it might be worth submitting since it is going to be different than our 4 other models. There will probably be no time to polish it, though. Wed Jul 5 19:41:10 PDT 2006 Grant Thiltgen I also had to remove the last two gaps at the end of the protein in order to get the strands to line up. I think I got the alignment right and the constraints in order, so I'm running try15 on vashon. For try15 I'm using the T0301-1w61A-handalign.a2m file in the 1w61A directory to make the initial model. I believe I have aligned the sheets up correctly in the alignment, but it's really hard to say. I'm hoping that the modifications I made might create a better model than we had before. I chose to go with 1w61A because it had the best alignment of the templates that had a similar number of residues to our target. Wed Jul 5 21:29:34 PDT 2006 Kevin Karplus I'm not sure Grant got the sheet constraints right for the alignment he provided, as he is getting terrible scores in try15. He doesn't say where the sheet constraints came from. They should have been from the alignment itself, using PrintAlignmentSheets, as in show-align.under from casp7/starter-directory. I made such a script in extract-sheets.under, and used it to extract constraints both from the standard t2k 1w61A alignment and Grant's hand alignment. They *are* somewhat different. I'll use the hand alignment to set the constraints for try16.costfcn. (The other alternative is to pick an alignment consistent with the constraints Grant extracted from try7-opt2.) Wed Jul 5 21:51:56 PDT 2006 Kevin Karplus try16 started on cheep. Wed Jul 5 22:40:13 PDT 2006 Kevin Karplus try14 finished, and it scores well with its own costfcn, but not with unconstrained. Rosetta does not like repacking it as much as it likes try8. It may actually be a slight improvement, though the backbone has not completely closed. If we submit try8 or try9, we should probably do the ones that Rosetta likes best: decoys/T0301.try9-opt2.gromacs0.repack-nonPC.pdb.gz decoys/T0301.try8-opt2.gromacs0.repack-nonPC.pdb.gz It doesn't look likely (at the moment) that either try15 or try16 will accomplish anything useful---the cost function is improving too slowly, and try7-opt2 scores much better at either cost fcn than what we are likely to achieve with try15 or try16. I'd be pleased to be wrongabout this pessimistic assumption. Wed Jul 5 23:17:14 PDT 2006 Firas Khatib yeah, it doesn't seem like try14 was able to make a sheet where I was hoping it would. Kevin, can you do a preliminary submission (excluding try15 and try16) and if some miracle occurs, we can resubmit? Thu Jul 6 00:31:27 PDT 2006 Firas Khatib well, try16 did score better than try15 (according to its costfcn), but as you predicted it did not do better than try7. try15 and try16 both score very poorly with the unconstrained costfcn, especially compared to try7. Last time we submitted with: try7-opt2 try8-opt2 # from ROBETTA_TS5 try9-opt2.repack-nonPC # from RAPTORESS_TS1 M1-D183 try1-opt2 F178-F395 try1-opt2 should we submit: try7-opt2 try8-opt2.gromacs0.repack-nonPC.pdb.gz try9-opt2.gromacs0.repack-nonPC.pdb.gz M1-D183 try1-opt2 F178-F395 try1-opt2 or do you want the last 2 to be try14 & try16 ? If Kevin doesn't submit before he leaves for SF, I will submit at 9am tomorrow when I launch the makes for the new targets. Thu Jul 6 07:14:58 PDT 2006 Kevin Karplus Submitted Model 1 is try7-opt2, our best-scoring model refined from the chimera of the two domains. Model 2 is try8-opt2.gromacs0.repack-nonPC, which was optimized from a large number of server models, but which chose to work mainly on ROBETTA_TS5. Model 3 is try9-opt2.gromacs0.repack-nonPC, which was optimized from several of the top-scoring server models, excluding the ROBETTA models. It chose to focus on RAPTORESS_TS1, which had remarkably few clashes. Model 4 is try2-opt2 for domain M1-D183. Model 5 is the automatic prediction generated for domain F178-F395. Wed Mar 21 14:13:11 PDT 2007 Kevin Karplus The assessors set the domains as D1:1-182,378-395 D2:187-377 We did better on domain 2, with the try13-opt1-scwrl model as our best, and the model 1 was quite good (model5 was fine in GDT but was missing too much of the sequence, so was downgraded in our cost function, even though what was missing was not part of the real domain we were comparing to---it should have come out better than model2, not worse.