Fri Jun 2 09:25:22 PDT 2006 T0312 Make started Fri Jun 2 09:26:16 PDT 2006 Running on cheep.cse.ucsc.edu Fri Jun 2 13:52:58 PDT 2006 Kevin Karplus weak hit to 1xv2A (7e-03 with the hmms) This template only scores E-value 41.2 with blast of PDB, 24th in the BLAST hit list, with 40% id over 20 residues (at least the hit is to residues in the middle, not to the HIS tag at the end). Fri Jun 2 18:11:31 PDT 2006 Kevin Karplus The top hit to 1xv2A actually looks fairly good and try1-opt2 polishes it up a bit. We need to fix the alignment for Y53-L73, but after that the model should be ok. Tue Jun 20 13:35:31 PDT 2006 George Shackelford Navya and I are working on T0312 now, and we're picking up on the problem with residues E71-L73. We've picked up the sheet constraints from try1-opt2.sheets, put them in the costfcn, and replaced the sheet between L73 and 44-39 with: SheetConstraint (T0312)H39 (T0312)V46 (T0312)N78 (T0312)E71 hbond (T0312)I40 1 This should adjust the problem. Try2 running on peep. Wed Jun 21 09:28:22 PDT 2006 Kevin Karplus try2-opt2 did not score quite as well as try1-opt2 and has the same constraint values. The top server models (Pcons6_TS3, Pmodeller6_TS1, ROBETTA_TS1) pretty much agree up with try1 and try2 to L120, aside from the loop Y54-P69, so I think that the goal here will be to come up with different plausible C-termini and different loops for Y54-P69. Wed Jun 21 13:36:35 PDT 2006 Kevin Karplus There are very bad breaks in try2-opt2: T0312.try2-opt2.pdb.gz breaks before (T0312)L73 with cost 17.2516 T0312.try2-opt2.pdb.gz breaks before (T0312)A37 with cost 3.50798 T0312.try2-opt2.pdb.gz breaks before (T0312)E60 with cost 3.38951 T0312.try2-opt2.pdb.gz breaks before (T0312)P69 with cost 2.29557 Wed Jun 21 14:04:30 PDT 2006 Kevin Karplus Neither try1.costfcn nor try2.costfcn made any attempt to fix the bad breaks in the strand. I've set up try3 to try to get a decent strand here (based in part on Pcons6_TS3). I'll also see if the long loop could fold over to match the rest of the sheet. Wed Jun 21 15:44:56 PDT 2006 Kevin Karplus With the try3 costfcn, the SAM_T06_server_TS1 model scores best after try3-opt2. The strand we want is *not* being formed though, as the strand id upside-down from what we requested. The choice is between trying to flip the strand over (say with ProteinShop) or changing what we want. # What the SAM server has is SheetConstraint S48 I51 E109 F106 hbond S48 1 SheetConstraint V50 G52 F106 H104 hbond G52 1 SheetConstraint S48 Y53 P69 K64 hbond A49 1 which matches the constraints in try3.costfcn. Let's do another try, this time polishing from existing models (including the SAM_T06_server_TS1 model). Wed Jun 21 16:03:00 PDT 2006 Kevin Karplus try4 uses the sheet constraints from SAM_T06_server_TS1.sheets and optimizes from existing models (including SAM_T06_server_TS1.pdb). [Running on orcas] try5 will have the same costfcn, but starting from alignments (focussing mainly on 1xv2A and 2fug7, which were used in earlier tries). [Running on whidbey] Wed Jun 21 16:23:29 PDT 2006 George Shackelford (Navya) Navya and I noticed that the following sheet isn't working.: SheetConstraint S48 Y53 P69 K64 hbond A49 1 It doesn't allow enough between 69 and 70 and P69 is a proline. It probably doesn't play well with the sheet. We're trying this: SheetConstraint S48 Y53 M67 V62 hbond A49 5 try6 running on peep? Wed Jun 21 18:03:25 PDT 2006 George Shackelford (Navya) So we got something that has a couple of bad breaks. They look fixable but Kevin suspects part of the problem comes from missing residues in the actual crystal structure of 1xv2A. Thus we use VAST to find 1xv2A and have it form an alignment. Alignment with 1xv2 from VAST 10 20 30 40 50 60 ....*....|....*....|....*....|....*....|....*....|....*....| VS11_A 7 EVGKGFLLRLDYGKDLVRQIEEFLEEKGiHAAHISAIGAVRSAVIGYYDqek-------- 58 1XV2_C 83 KASKTFPLQQLSQDDVFAQIKNEXLSEN-LFSAVKIYGTFKHXHVRXXPaqqppytrlid 141 70 80 90 100 110 120 ....*....|....*....|....*....|....*....|....*....|....*....| VS11_A 59 --keYVKKELMEPleILSLSGNVSMKDSK----PFCHIHVLLgkdgevygGHLFSAEVFA 112 1XV2_C 142 sarrQPEEKRQDI--RGAIVGFFTPELFHgvgsAGFHIHFADder--aygGHVLDFEVDD 197 ....*.... VS11_A 113 CEVFVLPLS 121 1XV2_C 198 VVVEIQNFE 206 The mFASTA version (which is actually more useful) >pdb|VS11|A Chain A, ------------------------------------------------------------ ----------------mkvfefEVGKGFLLRLDYGKDLVRQIEEFLEEKGiHAAHISAIG AVRSAVIGYYDqek----------keYVKKELMEPleILSLSGNVSMKDSK----PFCHI HVLLgkdgevygGHLFSAEVFACEVFVLPLSgeaperafdeqtglflwlehhhhhh---- -- >gi|61679628|pdb|1XV2|C Chain C snaxtnvlyqhgtlgtlxagllegtatinellehgnlgiatltgsdgevifldgkayhan ehkefielkgdekvpyasitnfKASKTFPLQQLSQDDVFAQIKNEXLSEN-LFSAVKIYG TFKHXHVRXXPaqqppytrlidsarrQPEEKRQDI--RGAIVGFFTPELFHgvgsAGFHI HFADder--aygGHVLDFEVDDVVVEIQNFEtfqqhfpvnnetfvkakidykdvaeeire ae Ouch. We've got an alignment with chain C not chain A. We're going to do a MANUAL_TOP_HITS to get chain C so our residues actually match up. Then we're going convert the mFASTA to an a2m format and put it in as our FragmentAlignment. Wed Jun 21 18:30:36 PDT 2006 Kevin Karplus 1xv2C is exactly the same sequence as 1xv2A. VAST just chose a different representative. There is *no* need to do anything with 1xv2C---just change the id to 1xv2A. Changing to a2m format is trivial---just replace the "-" with "." Take a close look at the alignment---you may want to move things to close any gaps so that the sequence of our target matches the sequence of the template in the strands. Wed Jun 21 19:35:29 PDT 2006 Kevin Karplus try6-opt2 scores best with the unconstrained costfcn, then try5, try4, try1, try3, try2 But try6 has too many residues in the loop from E68-I72. L66 should pair with V50 or I51, not A49. We had requested L66 with I51 in the try4/try5 costfcn, but neither try got it. The folding of the sheet is best in try4, even though the strand pairing did not quite form. From: George Shackelford To: karplus Subject: about using 1xv2C Date: Wed, 21 Jun 2006 23:28:10 -0700 I had thought that the actual sequence of residues was important to alignment. Since 1xv2C has the same structure but not the same sequence as 1xv2A, I thought that would throw off the use of the alignment. - George Date: Thu, 22 Jun 2006 04:34:40 -0700 From: Kevin Karplus To: ggshack Subject: Re: about using 1xv2C 1xv2C has exactly the same sequence as 1xv2A, at least according to the dunbrack-pdbaa reduced set of PDB sequences. It is true that they have different numbers of atoms in the PDB file, but only by 7 atoms, and they have the same holes in the structure where we care. Actually, there aren't holes there---just GTFKHMHVRMMPAQQ has MSE instead of MET for the M's, and rasmol doesn't seem to understand that convention. ------------------------------------------------------------ Thu Jun 22 05:16:09 PDT 2006 Kevin Karplus I tried making a hand alignment to 1xv2A that had breaks in the loopy regions, rather than in mid-strand (1xv2A/T0312-1xv2A-hand1.a2m). I also put in 1xv2A/T0312-1xv2A-VAST-try6.a2m based on the VAST alignment that George got to try6-opt2, though I don't like it much. I created incomplete models from the alignment with show-hand-align.under, which also creates T0312.hand-align.sheets I'll do an optimization run with just the T0312-1xv2A-hand1 alignment and the sheet constraints for it. (try7 started on shaw) Thu Jun 22 06:19:35 PDT 2006 Kevin Karplus For try8, I used the same alignment as try7, but added on extra sheet constraint to try to get the long hairpin to curl like the rest of the sheet. (try8 started on lopez) Thu Jun 22 07:59:52 PDT 2006 Kevin Karplus try7-opt2 looks pretty good and scores best with the unconstrained costfcn. The only part I don't like is D55-M67, which I think should be a hairpin parallel to the current last hairpin. try8 attempted to get that, but failed, tearing apart the end of the existing sheet in try7. I'll try again with try9, which has strengthened the sheet constraints on either side of the new strand-pairing I'm trying to make. (try9 started on lopez) It might also be worthwhile to try to do some thing with the first 10 residues---perhaps folding them back as another strand? Thu Jun 22 08:46:47 PDT 2006 Kevin Karplus try9 seems to be improving try7's score substantially, so I hope it will do the right thing. I'll wait until it is done before trying to fuss with the first 10 residues. Thu Jun 22 08:56:03 PDT 2006 Kevin Karplus I scored the server models with the try9 costfcn. The best servers (with this assumption of sheet topology) were Pmodeller6_TS2, ROBETTA_TS5, SAM_T06_server_TS1, HHpred2_TS1-scwrl, ... Thu Jun 22 10:46:24 PDT 2006 Kevin Karplus try9-opt2 is the new best with both try9.costfcn and unconstrained.costfcn. I think the N- and C-termini are wrong, but I'm not sure which way to move them. There is a large patch predicted to be buried (buried against another sheet in 1xv2A), which is probably an oligomer interface that the the current positions of the termini are blocking. The patch is not as hydrophobic as it seems, since there are some polar residues in it (S74, S76, H39, H91, H104). I'm going to try making a chimera: 1-52 try9-opt2 53-60 try4-opt2 61-67 ROBETTA_TS5 68-121 try9-opt2 122-end try4-opt2 This will have some bad breaks, but it should polish up to something useful. Thu Jun 22 11:38:03 PDT 2006 Kevin Karplus Running try10 from chimera-try9-try4 on cheep, with breaks and clashes turned up. Thu Jun 22 13:22:58 PDT 2006 Kevin Karplus try10-opt2 scores best with the try9 and try10 costfcns, but comes in second to try9-opt2 for the unconstrained costfcn. Thu Jun 22 13:42:12 PDT 2006 Kevin Karplus Starting try11 polishing from all models, but with moderately strong constraints still in place (this will probably just tweak up the try10-opt2 model, but there may be some crossover). When it is done, I'll do one more polishing run with no constraints. It looks like not much is happening in try11 (16 generations with no change in the best model). Thu Jun 22 13:54:59 PDT 2006 George Shackelford I've looked at the pair of hairpin turns that Kevin is using to finish the sheet. I'm concerned that this construct is not the most stable. I still like the awkward but workable solution that Navya and I had on try6. The reason we allowed for the extra residues in the loop from E68-I72 is because it would break otherwise. Using the 'near' script it appears pairing L66 with V50 is possible and desirable. We're trying this: SheetConstraint S48 Y53 E68 K63 hbond A49 5 I think I'll give it a spin in try12 (based on try6). try12 running on peep. Thu Jun 22 15:10:31 PDT 2006 Kevin Karplus I'm not convinced that try6 is a good basis for anything, but I'm willing to see some alternatives. I see that try12 is basing its start on the VAST alignment of the try6 model to 1xv2A. I think this alignment is wrong in several places, because the try6 model was wrong. It looks like the README file shrank a bit, as I *thought* I had put some more notes in here about why 1xv2C was not needed. Perhaps I stepped on those changes by mistake. Or perhaps I accidentally put them in the wrong README file. They aren't in any other README file, so I guess they're gone. They were pretty much a repetition of what is already earlier in the README file, so it's no great loss, but I will have to be more careful in my editing. I'm going to try a polishing run (no constraints) as try13 to (try to) close the remaining breaks in try11-opt2. Thu Jun 22 16:01:03 PDT 2006 George Shackelford Try12 came out looking like try10 in trying to form two hairpins on the end of the sheet, so my "fix" to S48-E68 went nowhere. I don't see any way to get "back" to try6's formation easily and I have enough on my plate. Thu Jun 22 16:49:50 PDT 2006 Kevin Karplus The preliminary submission is For the preliminary submission, we are ended up taking 1xv2A as our primary source. Alignment was difficult, and we ended up hand-tweaking the alignment to get gaps in the right places. We also tried to use undertaker constaints to form E68-Y53 into a hairpin and wrap it around like the other hairpins. This is not yet successful, possibly because of residual misalignment. Model 1 is T0312.try13-opt2.pdb, our best-scoring model, optimized from a chimera of try4-opt2 and try9-opt2. Model 2 is T0312.try9-opt2.pdb, which started to form a hairpin we wanted. It was optimized from a hand-tweaked alignment to 1xv2A, as non of the automatic alignments seemed very good. Model 3 is T0312.try8-opt2.pdb, an earlier attempt to for the E68-Y53 hairpin---one that tore the sheet apart a bit to satisfy constraints. Model 4 is T0312.try4-opt2.pdb, which used the sheet constraints from our server (SAM_T06_server_TS1) in an early attempt to get a decent model. There are some aspects of this model that seemed useful to save and were included in the chimera for model 1. (Notable the 53-60 and 122-end). Model 5 is T0312.try1-opt2.pdb, which was generated with no manual intervention. Mon Jul 10 16:42:25 PDT 2006 Kevin Karplus I'd better start looking at this again, as Navya is sick and George has other targets that are more pressing. I made a try14.costfcn that tries to get the sheets and helices of try13-opt2, plus the extra hairpin I mentioned above. I made the constraints pretty strong, and am trying again from the alignments. Mon Jul 10 17:10:58 PDT 2006 Kevin Karplus try14 started on cheep. I'll also do a run with the same costfcn, but starting from the existing models (which are better polished, and so may be harder to move). Mon Jul 10 17:14:55 PDT 2006 Kevin Karplus try15 started on lopez Mon Jul 10 20:51:48 PDT 2006 Kevin Karplus try14 did not come close to making the desired hairpin. try15 made such tiny changes to try13 that they are barely noticeable (though rosetta thinks that try15-opt2.gromacs0.repack-nonPC is the best model so far). Of course, not many of the other models have had gromacs0.repack-nonPC models made---so I'll make a few now and see if rosetta changes its favorite. Mon Jul 10 21:02:40 PDT 2006 Kevin Karplus Even after filling in some missing files, rosetta still likes try15-opt2.gromacs0.repack-nonPC best. I'll try making the constraints stronger and start from the gromacs0.repack-nonPC models. Even better, maybe I should deliberately break the chain before Y53 and after E68 and after E57 and before K63, to make gaps for the operators to play with. Mon Jul 10 21:28:02 PDT 2006 Kevin Karplus I made such a model (from try15-opt2.gromacs0.repack-nonPC) as scrambled-loop.pdb.gz Mon Jul 10 21:39:55 PDT 2006 Kevin Karplus try16 started on cheep, trying to fix up scrambled-loop.pdb Mon Jul 10 22:25:41 PDT 2006 Kevin Karplus try16-opt1 has very bad breaks, but is beginning to get the hairpin strands in the places I wanted to see them. Mon Jul 10 23:28:17 PDT 2006 Kevin Karplus try16-opt2 got very good costs with the try16 costfcn, but still had terible breaks and clashes. The final hairpin is almost where I want it, though, and gromacs relieves most of the clashes and breaks (though at pretty high cost to the sidechains). I'll try reoptimizing the try16-opt2.gromacs0.* models with the try15 costfcn to try to fix up the breaks and clashes (and clean up the damage done by gromacs). try17 running on cheep. Tue Jul 11 10:49:16 PDT 2006 Kevin Karplus try17-opt2 is pretty much what I was looking for. It could use a little polishing. Rosetta likes the gromacs's version of it best for repacking. Probable current submission: try17-opt2 < try16-opt2 < try15-opt2 < try13-opt2