Fri Aug 16 16:04:52 PDT 2002 t0193 Fold recognition, probably c.2.1 1 Sept 2002 Kevin Karplus 15:55 Best result from first run is try1-opt-scwrl. Model seems to be based on 1cf2O, one of several c.2.1.3 templates. There are some alignment problems, as V182-V187 is predicted to be a parallel strand but is not part of sheet, and R195-I200 is predicted to be helix, but is aligned in sheet. Perhaps we should turn up pred_alpha2, add "heavy-str" alignments, and start over from scratch. 3 Sept 2002 Kevin Karplus 10:02 try2-opt has best score, but has mangled the sheet. We need to figure out where the sheet really belongs and add constraints to hold it together. Sep 6, Fri 10:46 AM Look at the try2-opt Add constraints to pair the beta strands 104> FELRGFFD 80> WGLCIVG 142> IALLT 166> GILN 188< VNEVA Sep 6, Fri 4:01 PM Undertaker has taken a long time. Check on the log. The machine has crashed on me because of my constraints. I set constraints for beta-carbon of G residues which does not have beta-C. Re-run try3. Sep 7, Sat 12:25 AM Oscar Hur I do not see the try3-opt file (pdb). Look at the log and find that the template.atom cannot open. Sun 8 Sept 2002 Kevin Karplus 10:17 Fixed the Template.atoms permission. (Oscar, you should send me e-mail when that happens---I read my e-mail more frequently than I scan all the README files.) Looking at the projected pairings above---they don't seem to match what was found in try1. predicted strands s1 24 qgvHRTsse ? s2 78 nrkwGLCIVGmg ^ s3 102 gesfELRGFFDvdpe ^ s4 115 pekvgRPVrg v s5 123 rgGVIEHvd ^ s6 138 griEIALLTvpr ^ s7 163 agikGILNfap ^ s8 171 fapVVLEVpke v s9 180 keVQVENVDfl ^ Pairing in try1 s5 s3 s2 s6 s7 H129-F110-I85-L145-I168-L196 V130-F111-V86-L146-L169-S197-K180 I think that the s5 through s7 parallel connections are worth preserving. and s7-s8-s9 is probably a simple antiparallel connection s7 163> agikGILNfa s8 179< pVELVVp s9 180> keVAVENVDfl I extracted the hbonds for the first part of the sheet from try1-opt, then added my guess for the antiparallel connection as try4.constraints. I'll try running Oscar's try3.constraints, also, but I have less confidence in them---I don't see how he chose the pairing he did. 13:32 Kevin Karplus try4 did not do a great job. s5-s3-s2-s6 are ok, but s7 has started to drift, and nothing after than looks good. It MIGHT be the case that s1 should run parallel to s5. try3 (Oscar's idea) seems to be working better. try3-opt has across the sheet G167=I143=L83=L107= L169=L145=I85=F110 and antiparallel V122=G125 We can almost make the parallel F111-V126 connection. Can I make a new set of constraints combining Oscar's ideas with what I saw in try1? Make s8 not be a strand, but just an extended turn, and line up s9 antiparallel to s7? s4 s5 s3 s2 s6 s7 s9 R120-I127-D112- P121-V126-F111-V86-L146-L170-A183 G125-F110-I85-L145-I169-V184 17:18 try5 is beginning to look vaguely protein-like. Perhaps we should add some helix length constraints to help them fold up a bit faster. Trying them (together with current strand constraints) in try6.constraints. After that, we'll either have to think about the sheet, or try running VAST. Maybe I should start a vast run now from try5-opt, and see if it finds anything. job VS31343 password casp5t0193 The VAST run finds a good homology to 1b7gO, with 1.3 Ang over 107 residues and 13.1% sequence ID. This homology is basically an entire domain, but does not include the N-terminal end of the target, just from GLCIVG on. There is also a similar homology for the same domain to 1bmdA (2.8 Ang over 91 residues with 16.5% id). Unfortunately, VAST does not show an overview screen of all the hits that shows what PARTs of the sequences are covered---going into CN3D for each is tedious. There is one that has a preceding domain that might match---1lnqA, which is not in our template library, nor in our old version of FSSP. Let's add it to the library and force scoring with make 1lnqA.new_t2k_lib_pwise_alignments Other possible domains before the one we have reconstructed are from 1ejjA, 1a7aA (fssp rep 1b3rA), 1ep1B (fssp rep 1ep2B), 1f0kA, 3btoA (fssp rep 2ohxA). All of these, or their very similar FSSP representatives are already in our library, but none came up in the top scores. We'll try try7 from scratch with the try6.constraints and these alignments added to the pool. (oops, the 1lnqA t2k build is not done---the alignments for it are even trashier than one would expect---why was no error message seen?) 21:47 try6 is done. It has 4 parallel strands, with scattered helices, but that's about it. We can submit it (or one of the closely related ones) if we have to, but I hope we can do something with the N domain. 22:39 1lnqA build is done. I'll remake force-extras and extra-alignments. 9 Sept 2002 Kevin Karplus 11:32 The try7-opt run is not bad, and (though it scores worse) seems to have packed the last few strands better than try6. Let's do a from-scratch run with the other conjectured alignment for strands s7-s8-s9, and see how it runs (try8). 14:20 Try8 looks so-so: the s7-s8-s9 region is not assembled and s4 has gotten detached. try6-opt scores best with the try8.constrants, as well as with the try6 constraints. For try9, let's try making strand s9 PARALLEL to s7, as try7 seems to be trying to do. 17:20 try9 looks like a mess, but scores best with try9.constraints, try8.constraints, try6.constraints, try6-opt-scwrl scores best with no constraints So far I like try7-opt best, though it doesn't seem to score well. What do I try next??? 20:37 Maybe try putting V182-V187 parallel to K166-N170, letting F189-A199 go back to being a helix. Hmm---that's exactly what try9-opt was TRYING to do. We could add Hbonds to G167 and L169. 10 Sept 2002 09:01 try10-opt does not look any better. Let's try working on some other strands for a while. How about the weakly predicted strand 56> gtRGVGYTvp? It should run the same direction as 77> lnrKWGLCIVGmg, but we currently have strands on both side so that (though with bad breaks at the end of the strand). Let's insert the strand at 60 between the strand at 80 and the one at 145 164> gikGILNfap 138> pgrIEIALLTvp 56> gtRGVGYTvp 77> lnrKWGLCIVGmg 102> gesFELRGFFSvdp (Hmm, it looks like the restarting of the condor queue by the cluster administraotrs restarted the try10 job, wiping out the try10 log for the completed job. Damn condor for a pice of incompetently written code!) I've killed the job, but the overwritten log is gone. I'll start try11 with the extra strand in mid-sheet, starting from just the alignments. 10 sept 2002 13:39 try11-opt is not great, but it shows some promise. Let's dink with the alignment a bit to match what try11-opt seems to be trying to do: // 164> gikGILNfap // 138> pgrIEIALLTvp // 56> gtRGVGYTvp // 77> lnrKWGLCIVGmg // 102> gesFELRGFFSvdp // 124> gVIEHv // 123< rVPRGvk 15:07 Started an N-terminal only run in t0193-1-80. It looks like there is a pretty good hit to fold a.4.5. 15:16 try12-opt is tiny change over try11-opt, because only 20 generations were run, not 200. Trying again in try13. 16:03 just discovered that the run in t0193-1-80 had failed for lack of a define-score.script file. Foo---I'm too tired to be doing this. 16:30 t0193-1-80 try1-opt looks ok. Let's do one polishing run with it, then cut-and-paste it into the best full chain. From: "rph" To: Subject: more unexciting alignments Date: Tue, 10 Sep 2002 16:54:50 -0700 In t0193-1-80 T0193-double2track-EBGHTL-local-adp5.a2m & .pa has 2-track model-2-track model alignments to all the pdb hits except 1j5yA as /cse/faculty/karplus/pcem/pdb/1j/1j5yA does not exist. Again, not sure if it will help at all. Richard ---------------------------------------------------------------------- Tue Sep 10 17:35:55 PDT 2002 Kevin Karplus In 1-80 try2-opt looks reasonable---I'm not sure about the helix bend around T13---we have strongly predicted helix I12-E22 and fairly strong E5-A23. Let's try cutting-and-pasting it into try13-opt, copying just through L77 (to avoid the curling in of the C-terminus of 1-80). I then moved the inserted section so it stood out a little ways from the rest of the structure but in a reasonable place to slide in. I'll try try14 from there with try12 constraints, and try15 rom there with the try10 constraints. From: "rph" To: Date: Tue, 10 Sep 2002 17:40:18 -0700 rph.a2m in the t0193 has for 1b1b, 1bi0, 1bia the model, the sequence aligned as a model guide seqnece, and the sequence as it aligned to the target model. Mostly, the mod-model likes longer runs. did shift 1 internal delete a few positions in 1bi0. ---------------------------------------------------------------------- From karplus@bray.cse.ucsc.edu Tue Sep 10 18:09:21 2002 Date: Tue, 10 Sep 2002 18:09:19 -0700 From: Kevin Karplus To: rph@soe.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: Subject: Re: I think the guide sequences are incorrectly assigned. It looks like you picked up the lower-case letters from stride-mixed.seq, and not just the upper-case ones. If you want to use all the letters, try the stride.seq sequence, which is the upper-case subset of the stride-mixed.seq ------------------------------------------------------------ Tue Sep 10 20:43:46 PDT 2002 Kevin Karplus try15 does not look horrible. It could use a little polishing, but looks feasible. We may want to add strand at R28 T29 next to V182-E185. Which orientation? maybe pair H27 with V182 and S30 with V184. Hmm that's not consistent. How about 33< esSTRHvg 180>kevAVENVdf try14 DOES look horrible. 12 Sept 2002 00:15 Kevin Karplus try16-opt, which is based on try15-opt, but tries to attach the strand in the Nterminal domain to the C-terminal sheet, manages to get a much lower score,but has twisted the edge strand severely out of the plane of the sheet. Try17 will try do do the same attachment, but as a parallel edge (which is more consistent with the prediction, anyway). 12 Sept 2002 08:56 Kevin Karplus Try17 has really messed up the strand around E185. Have I messed up the constraints I think I want 139> griEIALLTvpr 163> agikGILNfa 180< kpvELVVp 181> eVAVENVd 24> qgvHRTss Currently, try17.constraints is requesting 139> griEIALLTvpr 163> agikGILNfa 181> eVAVENVd 24> qgvHRTss For try18, let's concentrate on 180< kpvELVVp 181> eVAVENVd 24> qgvHRTss and not worry about how it connects to 163> agikGILNfa For try19, let's go back to the try12 constraints (which did not attempt to attach 24> qgvHRTss), and re-optimize from try15-opt, try16-opt, and try17-opt. For try20, let's just omit the constraints on strand 24> qgvHRTss from try17.constraints, and reoptimize from try15-opt, try16-opt, try17-opt. 11 Sept 2002 11:23 Sigh, I don't really like any of try18, try19, try20. perhaps the best is try20, which might make some sense if I reverse the strand E185-D188, hooking it up between E181-A183 and G167-N170, and anti-parallel to both. To get it placed right, I might have to flip around everything from V184-G211. Let's try for 139> griEIALLTvpr 163> agikGILNfa 190< lfDVNEV 179> pkeva 24> qgvHRTs Date: Wed, 11 Sep 2002 16:49:27 -0700 From: Kevin Karplus To: karplus@soe.ucsc.edu, rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, learithe@cats.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu Subject: t0193 and t0194 I'm running out of ideas for t0193. The best I've come up with so far is probably T0193.try21-opt. Anyone willing to look at it (and any of try15-opt through try20-opt) and give me one more idea to try? ---------------------------------------------------------------------- best models scored with try21.constraints: try21-opt try18-opt try20-opt try17-opt try20.constraints: try20-opt try17-opt try16-opt try10-opt try18.constraints: try18-opt try20-opt try17-opt try21-opt try17.constraints: try20-opt try17-opt try16-opt try18-opt try16.constraints: try16-opt try17-opt try20-opt try18-opt try12.constraints: try13-opt try14-opt try12-opt try11-opt try11.constraints: try13-opt try14-opt try12-opt try11-opt try10.constraints: try20-opt try17-opt try16-opt try10-opt no constraints: try18-opt try20-opt try21-opt try17-opt Current thinking: model 1: try18-opt model 2: try20-opt model 3: try21-opt model 4: try17-opt model 5: try16-opt From: Rachel Karchin Date: Wed, 11 Sep 2002 17:25:44 -0700 To: Kevin Karplus Subject: Re: t0193 and t0194 Kevin Karplus writes: > > I'm running out of ideas for t0193. > > The best I've come up with so far is probably T0193.try21-opt. > Anyone willing to look at it (and any of try15-opt through try20-opt) > and give me one more idea to try? It does look good. Perhaps some of the predicted strand around 119-126 is part of the sheet? but you probably considered that already . . . ---------------------------------------------------------------------- 11 Sept 2002 Jonathan Casper 5:53pm I'm still not sure what I like from 15-opt through 20-opt, but whatever happened to try7-opt? It had a nice conformation for the sheet that I liked. The closest approximation I could find was in try17 or try18, but they didn't seem quite as good. I know that try7 scored poorly, but is there time for some final optimization? Date: Wed, 11 Sep 2002 17:58:50 -0700 (PDT) From: Jonathan Casper To: Kevin Karplus Subject: Re: t0193 and t0194 It's outside your suggested range, but I really liked the sheet from try7-opt. Is there time for another run off of try7, or else time to get constraints for that sheet and apply them to another decoy? -Jonathan ------------------------------------------------------------ I liked the sheet in try7-opt also. If I remember right, try7 was optimized with the try6.constraints, and I made up try8.constraints based on try7 With try6.constraints: try20-opt try18-opt try21-opt try17-opt try8.constraints: try18-opt try21-opt try20-opt try8-opt Let's look for the .rdb file in which try7-opt has the lowest rank, and reoptimize using those constraints. It looks like that would be try12.constraints on line 24, followed by try10 (with try7-opt-scwrl on line 26) and try11 and try20 (line 26). 20:00 The try23 run looks fairly good. With try12.constraints try12-opt now scores second only to try13-opt, but the sheet does not look as good as the one in try7-opt. Let's go with model 1: try18-opt model 2: try20-opt model 3: try21-opt model 4: try7-opt model 5: try16-opt Enough thinking---I'm going submit.