Tue Jul 6 08:05:27 PDT 2004 T0226 DUE date 24 Aug 2004 Tue Jul 6 13:49:26 PDT 2004 Kevin Karplus SAM-T04 gets a very strong match to 1c7qA in SCOP class c.80.1.2. This is also the top hit for t2k in the w0.5 scoring, but the score is much weaker. I think the difference is the inclusion or non-inclusion of the family on one iteration of the iterated search. The t2k sequence logo looks a bit more convincing---there seems to have been some drift in the t04 alignment. Tue Jul 6 20:55:44 PDT 2004 Kevin Karplus The try1-opt2 model has some good features (a bit of beta sheet), but has a lot of mis-placed parts. I think that we can probably do much better by including more alignments in the initial search---some of them are probably more complete than the ones in the t2k.undertaker-align selection. Although the template seems fairly clear, we're not picking up a good alignment on try1. Wed Jul 7 07:05:44 PDT 2004 Kevin Karplus We may want to use somewhat shorter runs in future---try2 took a while to complete. Even so, try2 is not looking great. There are some pieces of beta sheet (a strand-helix-strand and a strand-helix-strand-helix-strand) but the structure as a whole does not fit together. The sheets we do have may need to be tweaked a bit---I think some strands may be off by 1 or 2. We need to see whether George's predicted contacts will help any with getting overall topology. We also need to check whether any of the fold-recognition hits are reasonable---perhaps looking at the top t04 hits as well as the top t2k hits. Wed Jul 14 14:11:27 PDT 2004 Kevin Karplus This may be a classic parallel beta sheet with alpha helices packed on each side. The hard part will be figuring out the strand order. It looks like some, but not all of George's T0226.134.rr.constraints may be useful. Marcia Soriano has requested being the lead on this target for a while. We can jump in and help her out later if it turns out to be needed. Wed Jul 21 11:08:04 PDT 2004 Marcia Soriano For try3, I changed a couple of strand contraints by 1 as suggested. I based this upon comparison of the sequence logos. StrandConstraint E101 I104 5 Changed to: V102 I104 StrandConstraint L216 G222 6 Changed to: A217 G222 When running undertaker, I made a typo. I meant to type in the command try10.log, but instead somehow typed ytu3.log. Thu Jul 29 11:32:45 PDT 2004 Marcia Soriano I finally took a look at the try3-opt2.pdb. It looks like trash. I think I will need someone to help me out here. I might play around with the SheetConstraints, but would prefer to consult with someone a little more experienced before I do so. Wed Aug 11 15:38:20 PDT 2004 Marcia Soriano I worked on try4.costfcn. I increased the weights on the SheetConstraints to 5.0 and rescaled several parameters (contraints, hbond_geom, hbond_geom_backbone, hbond_geom_beta, and hbond_geom_beta_pair) to the new standard (0.1 becomes 1, 1.5 becomes 15, etc.). I commented out SheetContraints from try1-opt2 and a couple of them from align1 and added SheetConstraints from try2-opt2.sheets. See changes below. # from try1-opt2 # SheetConstraint A40 L42 F66 L68 hbond A41 5.0 # from align1 # SheetConstraint A34 G37 A81 A84 hbond Y35 5.0 # SheetConstraint A41 G43 A84 T86 hbond A41 5.0 SheetConstraint L83 T86 E101 I104 hbond A84 5.0 SheetConstraint L164 F167 P215 A218 hbond L166 5.0 SheetConstraint P165 F167 L190 L192 hbond F167 5.0 # from decoys/try2-opt2.sheets SheetConstraint R14 G16 L21 L19 hbond G16 5.0 SheetConstraint P38 L42 L65 E69 hbond H39 5.0 SheetConstraint P38 G43 A89 V94 hbond A40 5.0 SheetConstraint P215 V219 A242 A246 hbond L216 5.0 I have started to run try4 and will check on it tomorrow. Aug 13 17:51 2004 Bret Barnes So, I'm going to try a run on this one. I came up with some sheet constraints after looking at the str2 t2k predictions. It seems there are probably two sheets both parralel sheets with anti-parralel strands on the outside of the first sheet. Bellow are the constraints that I added. # Sheet 1 SheetConstraint T9 D13 L21 L17 5 SheetConstraint L17 L21 H39 G43 5 SheetConstraint H39 G43 L65 E69 5 SheetConstraint L65 E69 A81 E85 5 SheetConstraint A81 E85 I104 V100 5 # Sheet 2 SheetConstraint L166 S169 L216 V219 5 SheetConstraint L216 L220 A240 V244 5 try6 is running on meow. Tue Aug 17 14:27:50 PDT 2004 Marcia Soriano Thanks Bret for helping me out. I was working on the try5.costfcn, and have not yet finished it. My intent was to work off of try2, but I really was not sure where to go with it. I am going to look at it tomorrow after looking at T0239. Tue Aug 17 23:49 2004 Bret Barnes Ok, so I took another look at this one. After reviewing the best trys so far and the robetta models I have concluded on some strand groupings that I think are resonible (by noticing the number of times each pairing occurred in other structures). However I'm not going to try to connect any of the different sets of strands. Instead I'm going to toss in the George's RR constraints, since those havn't been used yet and hope that they will constribute to the correct pairing of strands. I also lowered the break by 5 (from 20 to 15) and raised the constraints by 5 (from 10 to 15) to try to get undertaker to put the sheets together. If this provides a more organized structure we should probably do a structure alignment search with VAST. Running try7 on meow. Wed Aug 18 11:46:48 PDT 2004 Kevin Karplus It would probably be worthwhile to create a "strands" rasmol script with a bunch of commands like define s1 8-12 define s2 41-42 ... to give names to the different strands. That way we can talk about which strands a parallel or anti-parallel, and can do selection in rasmol: select s1 or s2 color green select (s3 or s4) and backbone ... I looked a bit a try7-opt2. There are some good bits (H39-A41 || F66-L68), but some terrible ones as well (L252-M256 is strongly predicted to be helical, not strand). It looks like Bret and Marcia have NOT been using the (make T0226.do7) target, because no repacking or rescoring was done. I'll run through the tries and repack the missing ones, but PLEASE use the "do" targets in future. Since this protein is 290 long, it is quite likely to be multiple domains. It might be worthwhile to try to guess the domain boundaries (look at the domain predictions on the casp6 servers web site). The t2k and t04 multiple alignments disagree CONSIDERABLY about what the conserved residues are. The t04 alignment seems to get conservation for about G80-E200, while the t2k one gets a quite different signal in N151-P170 and gets conservation for L220-T290. This strikes me as the sort of problem we would expect if we had multiple domains and were getting contamination of the model by neighboring domains. It may be very important to break this target into domains to reduce the contamination. Wed Aug 18 14:04:03 PDT 2004 Marcia Soriano I went to the CASP website and took a look at domain predictions. Some servers detected only one domain,others two domains, and one predicted three. The servers that predicted two domains came up with inconsistent breaks. Servers that predicted 2 domains: 3D-JIGSAW-recomb, break at 99-100. BAKER-ROBETTA-GINZU, break at 124-125. BAKER-ROSETTADOM, break at 106-107. baldi-group-server, break at 158-159. Servers that predicted 1: cubic-chopper DomPRED DomSSEA Dopro SSEP-Align Server that predicted 3: Sternberg_Phyre, break at 20-23 (blanks at 21,22), and 71-79 (blanks at 71-78). Wed Aug 18 15:42:04 PDT 2004 Marcia Soriano Martina helped me figure out how to break the target into domains and create subdirectories. Domain 1 is from 1-122 and domain 2 is from 108-290. Try1 for both domains are running on squawk. In the region of 108-122, there is a helix. We decided to create a domain break in this area in order to use the helix as a linker for the domains. Thu Aug 19 0:01 2004 Bret Barnes I saw how the domains were broke into two. I'm going to try using the same constraints that I used for try7 (but divided obviously) for try2 on both domains and see how that works out. try2 (1-122) running on meow. try2 (108-290) running on woof. Thur Aug 19 13:07 2004 Bret Barnes So, both try2s (in each domain) don't seem to look that good (all helix). However I noticed that robetta model 4 is divided exactly into two domains (infact the domains we want). So, I'm going to cut that pdb file in half and then re-score with it and add it to the decoys. Robetta model 1 also looks good, and 2 and 3 are ok (all two domains). Robetta-model4 did not score well with the try2 cost function, but that is pretty predictable since the Robetta model doesn't really match out secondary structure predictions (which are not that good anyways). So, for try3 in 1-122 I'm going to try to use the Robetta-model4 (1-122) sheet constraints and see what happens, since try2 doesn't accomplish anything. SheetConstraint V102 H103 A84 E85 hbond A84 SheetConstraint L83 E85 L67 E69 hbond L68 SheetConstraint L65 L68 T9 V12 hbond Y10 SheetConstraint E8 V12 G37 A41 hbond H39 102: VH :103 83: LAE :85 65: LFLLE :69 8: ETYLV :12 37: GPHAA :41 try3 running on meow. I'm going to start looking at the 108-290 half of Robetta-model4 and see what I can do with that. Thur Aug 19 18:27 2004 Bret Barnes try3 (1-122) did not improve much, so I'm going to see if doing a VAST search on robetta-model-4 (1-122) will help improve our alignments. ID: VS60932 PASSWORD: T0226B Thur Aug 19 21:07 2004 Bret Barnes Ok, it seems the VAST results hit best against 1dqrA (new), 2pgi (already a main hit), 1qv9A (new) and several others that are not top hits. I'm going threw the README file to see how to add to the alignments. As for the (108-290) domain it seems that our the second domain of Robetta 4 matches most of the other trys. The t04 and t2k predictions seem to agree a little more at this end of the protein also. I'm going to come up with a set of constraints after reviewing all the trys so far and robetta models. Then maybe do another VAST run. I'll probably get to most of this tomorrow because my parents are comming into town tonight, but I'll work until they get here. Fri Aug 20 18:59 2004 Bret Barnes So, I'm still in the process of trying to add the VAST alignments to 1-122 (I haven't done it before). In the mean time I'm running a VAST alingment for the 108-290 domain on robetta-model4 (the one I like the best so far). The additional alignments should help a lot since the alingments we have are fairly weak. VAST ID: VS60962 PASSWORD: T0226B ----------------------------------------------------------------------------- Date: Sat, 21 Aug 2004 02:07:29 -0700 (PDT) From: Martina Koeva To: Bret D Barnes Cc: Martina Koeva Subject: Re: T0270 (different subject!) > > Also, when you want to add a VAST alignment to a target's domain (T0226 > in paticular) you just put the alignment into domain directory I assume, > right? Assuming that you have two subdirectories with the main target directory (because you have two domains), if you find an alignment through VAST to, say, domain 1 (1-122 in your case), you can do two things: 1. If the hit that you get from VAST is to a template that we are already using, then you would be able to find the subdirectory with the same name in your domain directory. For example, if domain 1 (1-122>) has a hit to a structure 1dqrA (from what I can see, it is the 4th hit in the order of P-values), since we already have 1dqrA as a template, we have a directory 1dqrA (under T0226/1-122/1dqrA). In that case, just take the alignment from the VAST website, save it and edit it, so that the dashes are replaced by dots (which makes it a valid a2m file) and put it in the appropriate directory: T0226/1-122/1dqrA. There are additional instructions in the main README file that Kevin sent on adding VAST alignments. 2. If the hit that you get is not to a template that we already have, such as 2pgi (again from looking at your VAST run), then create a directory called 2pgi in the domain directory: cd T0226/1-122 mkdir 2pgi Then follow the same procedure - get the alignment from the website, save it in the mFASTA format, edit it to a valid a2m format and put it in the directory: T0226/1-122/2pgi. You would have to follow the additional instructions from the main README file again, namely: 1.) Add 2pgi to the MANUAL_TOP_HITS in the Makefile for T0226. 2.) Run make extra_alignments, etc. (see main README) If these comments are confusing, let me know. If you would like me to show you how to do it for one of them, I can do that over the weekend (I will most likely be going to lab) - just let me know in advance. Enjoy your weekend and your parents' stay! Martina --------------------------------------------------------------------------- Sat Aug 21 10:29:41 PDT 2004 Kevin Karplus I looked at 1-122/ try1,2,3. All of them still look like trash. For 108-290, try2 failed because of /projects/compbio filled up. I'll run a try3 with the same costfcn, but starting from existing models, since the try2-opt1 model looks feasible. Actually, I'll rewrite 108-290/try3.costfcn, using more up-to-date cost function components, and getting the sheet constraints from visual inspection of try2-opt1. The sheet constraints are consistent with the ones in try2. Sat Aug 21 13:46:11 PDT 2004 Kevin Karplus 108-290/try3 does not quite form the desired sheet, but looks pretty good. There seems to be 4-strand sheet forming, though only a 3-strand sheet was specified. Burial patterns look pretty good also. Before running 108-209/try4, I'll add the MANUAL_TOP_HITS to 108-290/Makefile and make extra_alignments and all-align.* I'll also make extra_alignments for 1-122. Hmm. It looks like the t2k alignment was damaged there. Maybe I'd better remove it and redo the make! Sat Aug 21 15:23:47 PDT 2004 Kevin Karplus INTERESTING! The damage to T0226.t2k.a2m seems to be coming from the a2m script itself. No--it is coming from the lack of a new-line at the end of the T0226.a2m file. I'll fix that and remake everything, which could take quite a while. Of course, it won't really help, since T0226 remains a complete ORFan in this region. Sat Aug 21 18:27:10 PDT 2004 Kevin Karplus I started 1-122/try4 with what I would normally do on try2---similar to the default try1 run but with constraints taken only from t04.dssp-ehl2.constraints. When that is done. I'll look at the results and see if there are any sheet constraints I want to try to add, not being influenced by the ones previously tried by Bret. Since this region is an ORFan, we may not be able to come up with a convincing model. At some point I'll give up on the first domain, and slap whatever we have back together with the second domain, which does seem to be making some progress. 108-290/try4 is looking pretty goos, but the sheet is not quite forming. Perhaps I should send it to VAST to get a good alignment to a known structure, so that we could start over with a better alignment---one in which the sheets do form. JOB: VS60975 PASSWORD: T0226try4 Sat Aug 21 19:14:06 PDT 2004 Kevin Karplus The best of the VAST alignments is for 2pgi (one of our templates). I put the alignment into 108-290/2pgi/2pgi-T0226-try4-vast.a2m and made 108-290/show-2pgi.under to show some of the 2pgi alignments. Unfortunately, our vast alignment only picks up the helices, not the strands. The 2pgi structure has two sheets. One has 4 parallel strands (2,1,3,4) with a extra strand antiparallel to 4 (actually coming from much earlier in the sequence---before the other domain). The other sheet has 5 parallel strands (2,1,3,4,5). The strands that we are seeing trying to form in try4 are ordered (2,1,3,4) so this seems like a decent match. The problem is one of getting the strand phase right, and the vast alignment doesn't seem to help here. Perhaps one of the other vast alignments will cover some of the strands. The 1jeoA alignment covers some of the gaps in the 2pgi alignment. I looked at the 1jeoA structures from alignment, and while they did not look great, they did suggest a slightly different alignment of the strands, which I am trying in 108-290/try5 Sat Aug 21 19:49:46 PDT 2004 Kevin Karplus 1-122/try4 has finished, and it scores better than earlier runs (at least with the try4 costfcn), though it was starting from alignments. It looks fairly good for an ORFan match---the secondary structure prediction is not a good match, but that prediction is notoriously poor when you have only one sequence in your multiple alignment. Given that 1-122 is an ORFan, I'm not going to try to get a better model than this one. This may be trash, but anything we generate is likely to be, and it actually looks more convincingly packed than many of the comparative models we've built. Of course, it may end up not fitting with the second domain, though they both have helices in the overlap region. Sat Aug 21 21:33:10 PDT 2004 Kevin Karplus 108-290/ try5-opt2 looks pretty good. I'll make try6 have sheet constraints that correspond to the sheets that appear to be forming there, and do a polishing run from the existing models. If that seems to be converging, I'll do another polishing run with an unconstrained costfcn and try putting the domains back together. I updated the MANUAL_TOP_HITS to include all the weak hits in both the subdomains and am doing make extra_alignments and make all-align*, so that we have a good set of alignments to work from when we try optimizing the stuck-together domains. Sun Aug 22 07:23:31 PDT 2004 Kevin Karplus Although 108-290/ try6-opt2 has started forming sheets, it has done so by opening up the helix bundling, exposing a lot of hydrophobics. I think that try4-opt2 looks better. For try7, I'll try solidifying the sheets that try4 almost forms. Sun Aug 22 10:13:17 PDT 2004 Kevin Karplus In 108-290/ try7-opt2, we're still not forming the sheet, and L216-L221 is badly broken up. I don't think I'm making more progress---maybe I should do one unconstrained polishing run (try4 and try7 are the current favorites), then put the two domains together, since there will undoubtedly be more problems then. Sun Aug 22 17:05:46 PDT 2004 Kevin Karplus Let's try putting everything together now: 1-122/try4-opt2 108-290/try8-opt2 The superposition in superimpose-domains.under creates two-domains.pdb, and the clashes are not bad. I'll make a chimera by crossing over at R116-F117 and try doing an optimization run. Sun Aug 22 19:51:42 PDT 2004 Kevin Karplus try8-opt1 just finished and is the new best with the unconstrained costfcn, handily beating try1-opt2. Perhaps try8-opt2 will be good enough to submit. Mon Aug 23 10:51:23 PDT 2004 Kevin Karplus The try8-opt2 model does not look too terible. There are a few bad breaks: T0226.try8-opt2.pdb.gz breaks before (T0226)V91 with cost 15.7959 T0226.try8-opt2.pdb.gz breaks before (T0226)P38 with cost 5.0311 T0226.try8-opt2.pdb.gz breaks before (T0226)F172 with cost 3.66513 T0226.try8-opt2.pdb.gz breaks before (T0226)P165 with cost 3.47099 T0226.try8-opt2.pdb.gz breaks before (T0226)A217 with cost 2.84762 T0226.try8-opt2.pdb.gz breaks before (T0226)L216 with cost 2.6915 T0226.try8-opt2.pdb.gz breaks before (T0226)G37 with cost 2.02827 T0226.try8-opt2.pdb.gz breaks before (T0226)G222 with cost 0.830911 and a lot of bad clashes: self-bump: 0.321762 Ang (T0226)F172.N and (T0226)F172.CB threshold= 2.24843 cost= 0.999991 other-bump: 1.12368 Ang (T0226)V91.N and (T0226)L119.CD2 threshold= 3.1578 cost= 0.99797 other-bump: 1.12083 Ang (T0226)L221.O and (T0226)P223.CD threshold= 2.83689 cost= 0.996197 other-bump: 1.25182 Ang (T0226)E150.O and (T0226)P152.CD threshold= 2.83689 cost= 0.992618 other-bump: 1.27892 Ang (T0226)L216.O and (T0226)A242.O threshold= 2.47194 cost= 0.980821 other-bump: 1.45093 Ang (T0226)M80.CB and (T0226)G222.O threshold= 2.77262 cost= 0.979463 other-bump: 1.47585 Ang (T0226)G222.N and (T0226)Q254.NE2 threshold= 2.68166 cost= 0.972213 other-bump: 1.80492 Ang (T0226)V30.CG2 and (T0226)Y35.CD2 threshold= 3.20339 cost= 0.968005 other-bump: 1.93072 Ang (T0226)G222.N and (T0226)Q254.CD threshold= 3.2554 cost= 0.95648 other-bump: 1.68576 Ang (T0226)G79.O and (T0226)G224.CA threshold= 2.83765 cost= 0.956043 other-bump: 1.86609 Ang (T0226)T27.CG2 and (T0226)L82.CD2 threshold= 3.09357 cost= 0.951823 other-bump: 2.04456 Ang (T0226)V30.CG1 and (T0226)P36.CD threshold= 3.33936 cost= 0.947323 other-bump: 1.87024 Ang (T0226)E150.C and (T0226)P152.CD threshold= 3.00942 cost= 0.942391 other-bump: 1.89486 Ang (T0226)Y112.CE1 and (T0226)L252.O threshold= 2.95529 cost= 0.93052 other-bump: 2.11781 Ang (T0226)Y112.CE2 and (T0226)V131.CG1 threshold= 3.14043 cost= 0.905945 other-bump: 2.05209 Ang (T0226)V91.N and (T0226)L119.CG threshold= 3.01312 cost= 0.900213 other-bump: 2.07085 Ang (T0226)L52.CD1 and (T0226)L68.CD2 threshold= 3.03525 cost= 0.899137 other-bump: 1.90368 Ang (T0226)P170.CB and (T0226)V219.O threshold= 2.77262 cost= 0.895232 other-bump: 2.23414 Ang (T0226)Y112.CD2 and (T0226)V131.CG1 threshold= 3.20722 cost= 0.885741 self-bump: 1.79156 Ang (T0226)F172.N and (T0226)F172.CG threshold= 2.54256 cost= 0.877604 other-bump: 2.16402 Ang (T0226)L221.CB and (T0226)L258.CD1 threshold= 3.05865 cost= 0.874575 other-bump: 2.16854 Ang (T0226)L19.CD1 and (T0226)R90.CB threshold= 3.05865 cost= 0.872992 other-bump: 2.28366 Ang (T0226)V91.CA and (T0226)L119.CD2 threshold= 3.17578 cost= 0.861744 other-bump: 1.84704 Ang (T0226)S54.O and (T0226)D58.OD1 threshold= 2.54817 cost= 0.85496 other-bump: 2.18773 Ang (T0226)L221.C and (T0226)P223.CD threshold= 3.00942 cost= 0.852409 other-bump: 2.16959 Ang (T0226)G222.CA and (T0226)Q254.NE2 threshold= 2.9768 cost= 0.850111 other-bump: 2.22409 Ang (T0226)G222.N and (T0226)Q254.CG threshold= 3.01312 cost= 0.83826 other-bump: 2.35789 Ang (T0226)G213.CA and (T0226)L216.CB threshold= 3.17553 cost= 0.832413 other-bump: 2.31406 Ang (T0226)M80.N and (T0226)G224.N threshold= 3.11081 cost= 0.830565 other-bump: 2.41295 Ang (T0226)M80.CA and (T0226)P223.C threshold= 3.23479 cost= 0.827728 other-bump: 2.31164 Ang (T0226)L160.CD2 and (T0226)P165.CB threshold= 3.07655 cost= 0.820058 other-bump: 2.4348 Ang (T0226)G79.C and (T0226)G224.CA threshold= 3.23479 cost= 0.818153 other-bump: 2.39096 Ang (T0226)G222.CA and (T0226)Q254.CG threshold= 3.13521 cost= 0.803289 other-bump: 2.11778 Ang (T0226)V148.O and (T0226)N151.CB threshold= 2.77262 cost= 0.801419 Mon Aug 23 15:04:52 PDT 2004 Kevin Karplus At this morning's meeting, we agreed to try to pull a robetta domain1 prediction, polish it up and paste it into the current models. robetta model1 looks ok, as does robetta model4. I truncated these to 1-122 and put them in 1-122/decoys/from-robetta-model1.pdb and 1-122/decoys/from-robetta-model4.pdb It looks like Marcia did the same, calling the result robetta-model4-1-122. Neither scores well with the unconstrained costfcn, not being sufficiently densley packed. I'll try polishing just them for 1-122/try5. 1-122/try5 uses a cost function that includes constraints from the 2ry prediction. 1-122/try6 uses an unconstrained cost function. Mon Aug 23 16:53:38 PDT 2004 Kevin Karplus It looks like try5 is trying to slide the predicted helix out of the strand and slide some predicted strand in. Marcia is setting up sheet constraints in try7-try9 to encourage this. try6-opt2 just squished the robetta model together a bit. try6 scores almost as well with the unconstrained costfcn as try4, and try5 is not that far behind. I have great hopes for the tries that Marcia is starting. Mon Aug 23 17:53:10 PDT 2004 Marcia Soriano I am currently running try7-try14. All of them have sheet constraints. Try7 running on squawk: SheetConstraint A89 V94 T64 E69 hbond F66 2 Try8 running on squeal: SheetConstraint A89 V94 G63 L68 hbond F66 2 Try9 running on coo: SheetConstraint A89 V94 E62 L67 hbond F66 2 Try10 running on woof: SheetConstraint A89 V94 E61 F66 hbond F66 2 Try11 running on cluck: SheetConstraint R90 G95 T64 E69 hbond F66 2 Try12 running on caw: SheetConstraint V91 F96 T64 E69 hbond F66 2 Try13 running on chirp: SheetConstraint V92 R97 T64 E69 hbond F66 2 Try14 running on crow: SheetConstraint R93 P98 T64 E69 hbond F66 2 Mon Aug 23 20:10:25 PDT 2004 Kevin Karplus Based on the opt1 results and the tiny changes in the cost functions, none of Marcia's runs are doing a thing to get the desired change in the strands. The problem may be that the strand and helix constraints are not quite right for the robetta-derived models. Perhaps we should just keep HelixConstraint G75 A84 and the added sheet constraint. I'll try a quick run with changing the try8 costfcn this way (since try8-opt1 was the best of the robetta-based -opt1 runs with the unconstrained cost fcn. This will be try15. Mon Aug 23 20:32:35 PDT 2004 Kevin Karplus 1-122/ Foo, I left one of the try4 models in the try15 read list, so had to kill it and start over. I removed the SCWRL from the all-align to make it run faster. Try15-opt1 has managed to wind up the helix, and there is a chance that try15-opt2 will start getting the strand in place. Stiffening the constraints and removing all constraints except the critical ones seems to have made a difference. If I had time, I'd probably start another set of optimization runs from try15-opt2, with different alignments of the strand. Mon Aug 23 21:45:09 PDT 2004 Kevin Karplus Well, 1-122/ try15-opt2 has the helix nicely wound up, but doesn't have the strand in place at all. The bad breaks make it score poorly on the unconstrained cost function, and there isn't really time to polish it (and all its siblings) with its own cost function. What I'll do instead is a single, short, polishing run for all the robetta-derived models (excluding our own try4) with the unconstrained costfcn, and then take the scunched robetta model, however bad it really is, and stick it back onto the full model for a full optimization run. The try16 run looks like it is going to polish up try6 (the squished robetta model), so I'll try assembling that onto a full model to see if there will be any problems. Mon Aug 23 22:01:11 PDT 2004 Kevin Karplus The script superimpose-domains-2.under created two-domains-2.pdb, which was crossed over between Y114 and L115 to make decoys/two-domain-chimera-2.pdb . Mon Aug 23 23:16:23 PDT 2004 Kevin Karplus I redid superimpose-domains-2 and the rest using 1-122/ try16-opt2. This chimera does not score quite as well as the one from our 1-122/ try4-opt2 domain, but almost the same. I'll reoptimize the whole model the way I did for that domain (except that I won't SCWRL the alignments), then submit both in the morning. Mon Aug 23 23:20:34 PDT 2004 Kevin Karplus What I'll submit in the morning are try8-opt2 our best try9-opt2 with first domain optimized from robetta model try1-opt2 full auto try1-opt2.repack-nonPC best rosetta energy T0226-1c7qA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template Tue Aug 24 07:32:54 PDT 2004 Kevin Karplus I decided to swap the order of the first two try9-opt2 with first domain optimized from robetta model try8-opt2 our best try1-opt2 full auto try1-opt2.repack-nonPC best rosetta energy T0226-1c7qA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template since try9-opt2 has a similar first domain to try1-opt2---I'm not very convinced by the try8-opt2. I think that secondary structure prediction for the first domain by itself was rather poor. I'm including try8-opt2 on the off chance that it really is the better solution. Thu Nov 18 23:34:49 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 29.1171 (this was also align2) best submit 28.0896 (this was also model1) model1 28.0896 auto 24.9261 align 26.3666 robetta best 26.6495 (robetta model4) robetta1 24.7822