Thu Jul 11 12:49:42 PDT 2002 11 July 2002 Kevin Karplus T0152 appears to be scop domain d.108.1.1, with 1i21A or 1i12A as a probable template. The CAFASP servers agree, though 1qsmA seems to be a more popular template. In try1-opt.pdb, the beta sheet seems to have been pulled apart, though it is intact in model 4 of T0152.t2k.undertaker-align.pdb.gz, so we can easily figure out which Hbonds to add. Perhaps we should try rerunning undertaker with the new fragments before we start adding constraints, though. 12 July 2002 Kevin Karplus Most of the beta sheet is intact in try2-opt, but the strand C162-F167 has pulled slightly away from the sheet and E190-D192 and M198-E204 are floating in space. The problem is that in the templates, the last strand is inserted into the sheet of dimerization partner, and the 2nd-to-last strand is on the other side of the inserted strand. Note: C180 and C185 come quite close. Should they be treated as a disulphide? They are NOT conserved in the multiple alignment---in fact C185 is usually F or Y. 23 July 2002 Kevin Karplus Adding the SSbond and putting in the new pred_alpha cost function, new best-scorer is still try2-opt. Let's add constraints to try to make up for the missing strand We want K186-E190 to be held in place near R164-F167. The two strands run roughly parallel, so let's add CB constraints based on the distances in model 4 of T0152.t2k.undertaker-align.pdb (from T0152.t2k-2track-undertaker.a2m:1i21A). CB distances 10.876 R164 K186 -7.678 60.674 20.345 -9.929 50.594 16.936 12.149 I165 F187 -10.506 59.136 23.169 -6.926 48.673 18.139 8.692 M166 L188 -8.746 56.085 24.487 -4.207 51.167 18.940 10.441 F167 G189 -9.461 53.287 26.913 -1.542 49.676 21.145 8.326 D168 E190 -7.918 49.986 27.919 -0.805 47.675 24.241 Since a strand is fairly straight, these distances still allow for rotation about either strand. Let's add one constraint from the next strand over---say P124 13.888 P124 K186 -12.682 56.292 29.298 -9.929 50.594 16.936 Most of the alignments agree with this arrangement of the strands, but models 27 (T0152-1yghA-2track-protein-STR-global-adpstyle1) and 28 (T0152-1qstA-2track-protein-STR-global-adpstyle1) have a different placement of the final strand. Since these scored much worse, I'll ignore them in the scoring. With the constraints added, try1.12.20 is now the best scoring structure. It looks pretty bad though, so let's run a new optimization from the beginning with this score function. 23 July 2002 Kevin Karplus try3-opt-scwrl is now the best scoring, but it looks terrible. We need some constraints to pull the sheet together. 24 July 2002 Yael I have added CB constraints to keep the sheet intact (as in model 4 of T0152.t2k.undertaker-align.pdb.gz). IMPORTANT! by mistake I haven't changed undertaker.script so try4 was rewriting some of the try3 intermediate decoys (before I noticed the mistake). All the intermediate try3 were renamed to *try3a*. The T0152.try3-opt.pdb and T0152.try3-opt-scwrl.pdb were not changed and correspond to the original try3 run. All try4 decoys correspond to the run4 with the sheet constraints. 25 July 2002 Yael After run4 TO142.try4-opt.pdb and T0152.try4-opt-scwrl.pdb score the best Though don't look much better then try3. 27 July 2002 Kevin Karplus Let's try zipping up the hairpin started by T193-R197. The hbonds would be 191-A199 189-201 187-L203 Also, the reason that the try4.constraints worked so badly, is that adjacent strands were being forced to 10.876 apart instead of 5.12 for the CB, and the register of the strands seemd quite different from the T0152-1i21A alignment. Fixing the distance on the correct constraints, and removing the mis-registered constraints makes try3-opt-scwrl the best (rejecting try4). Hmm---the constraints I had added for R164-K186... are ALSO on the wrong atoms. I must have gotten them from an incomplete alignments that used a different atom numbering. Let's fix those also. After fixing those constraint errors, recoring the existing decoys puts try1-opt on top. Let's try re-optimizing from there. The try5 run seems to be getting some better sheet structure, even in early iterations. There looks like a strong possibility that strand G128-I133 is paired in parallel with R164-F167, though they aren't in 1i21A. The try5-opt-scwrl structure looks good in itself, but I think that the structure has folded in on itself, where the strand is missing. The gap for leaving the strand is still more or less there, but a loop seems to have crossed the gap and pulled the hairpin in. The hairpin seems to be off-register by 2, pairing G189 and L203, instead of G189 and Y201. Strand Y43 through V47 has gotten detached from the sheet at the other end, but should be possible to pull back into place with constraints. Hbond G44-G96 and R46-L94. Question---perhaps this is NOT a domain-swapped structure, and the extra strand of the hairpin should be on the other side, replacing the missing strand from the dimerization partner??? This would be an interesting avenue to explore. Even after adding all these changes (try6.constraints), try5-opt-scwrl still scores the best. Let's try optimizing for them in try6. 28 July 2002 Kevin Karplus Lowest cost is now try6-opt or try6-opt-scwrl (the scwrl is best in the log file, but on reading the PDB file back in, the unscwrled one scores better). I don't like the way the hairpin is paired or oriented. It is pairing F187 with A205, but I thing F187 should pair with L203. It is pairing P169 with E190, but I want F167 with Y201. I added back in the constraints on the separated strands (R164-K186..D168-E190), and increased the weight of the constraints cost function. try-opt still scores best, but now the constraints are clearly doing badly. Let's try again without a starting conformation, using the new score function. The try7 run starts with 1cjwA, then adds 1seiA alignments. The 1seiA alignments only cover the C terminus, so are probably trying to get the second sheet. When we're done with this run, it might be good to do a superposition of the model with T0152-1i21A.pdb, and add some constraints to keep the sheets properly positioned, if they have moved with a hinge motion. The beta sheets look well-formed even after just the first iteration of try7, but I may want to remove the extra spacing constraints on the separated strands, since they may be constraining the sheet too much. There are some bad breaks still, which I hope will get cleaned up in the superpool iteration. 29 July 2002 Kevin Karplus A power failure yesterday morning brought down the machine around 10:58am (that may have been when the UPS was exhausted), so the try7 job did not finish. The runs were all scoring better than try6-opt (except for iteration 2), so I'll start try8 with the try7.*.40.pdb starting points. I'll also take out the "spacing" constraints on the edges strands of the last sheet, relying on just the hbond constraints. Try8 is getting better scores, even on the first iteration, but I may want to play with the constraints for G128-A132, as that strand seems to be badly placed (I changed Yael's constraints, based on one alignment to 1i21A, but I think now that I was even more wrong). From karplus@bray.cse.ucsc.edu Mon Jul 29 11:39:39 2002 Date: Mon, 29 Jul 2002 11:40:25 -0700 From: Kevin Karplus To: yael@biology.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: (yael@biology.ucsc.edu) Subject: Re: t0152 The problem with less stringent incorrect constraints is that they are only useful when conformations are a long way from being correct, as they will drive the conformations toward the specified "desired distance". It is better, I think, to try different sets of constraints, and see which end up with reasonable structures. I'm not sure that we will want to start try9 from the best of the try7 or try8 conformations---we may want to start over from the initial alignments, as I did for try7. Why don't you start editing try9.constraints, looking mainly at the strand G128-A132. We want to close the gap D121-A122, which probably means pairing G128 with something later than L106. Here are some possible alignments of that strand: 110< WYLELYGGDT | | | | | 127> LGLHAAIA 110< WYLELYGGDT | | | | | 127> LGLHAAIA 110< WYLELYGGDT | | | | | 127> LGLHAAIA 110< WYLELYGGDT a definite possibility | | | | | 127> LGLHAAIA 110< WYLELYGGDT | | | | | 127> LGLHAAIA currently specified---too far off I'd suggest trying the L108-L127, L106-L129 pairing first (for try9). The try8-opt run, though the best-scoring with the try8.constraints, is not very good. The second sheet is broken and the G128-I133 strand seems misaligned. July 29 Yael As kevin suggested I am trying to get the last strand of the first sheet to stick to the sheet by pairing L108-L127 up to G104-A131 otherwise keeping the register of the sheet as in try7 and try8. I took out from the constraints the constraints on strand 162-167 since it didn't seem to help in try8. As suggested by kevin I am runing try9 from initial alignments not using conformations from try8. 30 July 2002 Kevin Karplus With the try9 constraints, the lowest cost is still T0152.try8-opt.pdb. In try9, the first sheet did not form fully, but the second sheet formed an interesting structure---the first strand of it is parallel to the second strand and hbonded to it on the same side that the third strand is antiparallel: 1 3 >>> <<<<<<< >>>>>>>>>>> 2 There isn't really room, so strands 1 and 3 bump into each other. I don't see why Yael took out the constraints that held Y43-47 antiparallel to P93-W98. For try10, I'll restore those constraints, and fix the typo in the CB for E107 (it came out as CA), but leave the constraints for the last strand of the first sheet as in try9. I'll also try starting from try6-opt, try7.6.40.pdb, try7.4.40.pdb, try7.1.40.pdb, try7.3.40.pdb, try8-opt, and try9-opt. From yael@darwin.UCSC.EDU Tue Jul 30 11:09:22 2002 Date: Tue, 30 Jul 2002 11:08:45 -0700 (PDT) From: yael@biology.ucsc.edu To: Kevin Karplus Subject: try10 I think try10 looks much better (it is also scoring better) though the first sheet still sems to have problems. I will try changing the register of 103-106 to its partners (slightly). The second sheet looks also better I will try now to force the first strand to be intact with the hairpin. Yael BTW I took out the HB constraints on 43-47 to 93-98 since we had the CB constraints for these strands already but it does look like the hb constraints help. Yael Mandel-Gutfreund Department of Chemistry and Biochemistry University of California, Santa Cruz Phone : 1-831-459-5128 e-mail : yael@darwin.ucsc.edu From yael@darwin.UCSC.EDU Tue Jul 30 11:22:32 2002 Date: Tue, 30 Jul 2002 11:21:56 -0700 (PDT) From: yael@biology.ucsc.edu To: Kevin Karplus Kevin, I was just thinking that since the sheet constraints (on the first sheet) are breaking the structure will it be worth while to try taking out all the constraints on the sheet and running undertaker based on the last conformation before trying a new set of constraints? Yael Yael Mandel-Gutfreund Department of Chemistry and Biochemistry University of California, Santa Cruz Phone : 1-831-459-5128 e-mail : yael@darwin.ucsc.edu Date: Tue, 30 Jul 2002 11:38:48 -0700 From: Kevin Karplus To: yael@biology.ucsc.edu You can TRY doing a run without constraints, starting from a decent model. My experience has been tha tour lack of Hbond cost functions causes sheets to explode if there are no constraints, but that doesn't ALWAYS happen. I suspect we'd make more progress by tweaking the constraints and adjusting some of the parameters of the score function. For example, I'll replace radius_fit by phobic_fit to get better "hydrophobic collapse", and add dry5 to get more care in tight packing. With those small changes, try10-opt is still the best, but try8-opt and try9-opt are not terrible scores. One could start with all of them as starting points, adding in try6-opt also. July 30 Yael Running try11 after tweaking the beta sheet constraints relative to try10. All constraints are specified and documented in try11.constraints July 30 Yael T0152.try11-opt.pdb now scores the best though it looks nearly identical to try10 execpt for one break that was closed. Tweaking the constraints didn't seem to change anything. Maybe we should put stronger constraints for sheet hbonding. July 31 Yael Run try12 after adding the hbonding constraints to try and get the two sheets intact. Increased the closegap probability to 0.1 and the cost for breaks to 7. T0152.try12-al4.8.30.pdb scores the best now. It doesn't look much different compared to try11-opt. Maybe should try and run it again without reading the previous conformation? Thu Aug 1 10:29:33 PDT 2002 Kevin Karplus try13, as expected, scores much worse than try12. I think that we need to look at each of try8 through try13 and pick what features we like. We may have to trim the constraints back to eliminate ones that we can't seem to satisfy. I've submitted tr12.al4.8 to VAST to try to get a better starting alignment that we can work from. Unfortunately, VAST does not see one of the strands of the second sheet, so will be handicapped in finding the alignment. I grabbed two alignments from vast, and edited them somewhat: T0152-1bo4A.vast.fasta T0152-1qsmD.vast.fasta Let's take out all constraints for residues 120-135, since we don't seem to have a good idea where that strand goes---none of the alignments we've created agree with our notion of where that strand is. I think we want to make the two sheets into one sheet. Perhaps we need to add some parallel constraints between the last strand of the "first sheet" and the first strand of the "second sheet". The "second sheet" is well-modeled by the 1seiA alignment, which hbonds R163-E204 I165-A202 F167-L200 L188-L203 E190-Y201 D192-A199 T194-R197 leaving R164 and M166 free to hbond with the previous strand. In T0152-1i21A, R164.N is Hbonded to D121.O, but this is near the bad mid-strand break in that alignment, so is suspect. We could try to match the CB spacing across 2 strands: W110.CB-M166.CB and ALA112.CB-R164.CB I've put together a new set of constraints in try14.constraints, which do not attempt to constrain the last strand of the first sheet, but have the other constraints that we have some confidence in. With these constraints, the best-scoring decoy is try7.30.40 and the second best is try13-opt-scwrl. Let's try a new run starting from a few of the best scoring conformations, and including the new vast alignments. Date: Thu, 1 Aug 2002 12:23:13 -0700 From: Kevin Karplus To: yael@biology.ucsc.edu CC: karplus@soe.ucsc.edu Subject: Re: t0152 I've just started another run for T0152 (try14). It will probably take several hours. I agree that try13 did a beter job on the second sheet---it looks like we messed up the constraints on it in the past. I started over, getting the constraints from the good alignment to 1seiA, and I think it will do better now. There is really only one sheet, I'm pretty sure, but we're having trouble getting the constraints right on the connecting strand. I left them off of this run, to see if we can get it to form naturally. I did add some bridging CB constraints for the strands on either side of the one we're having trouble aligning. (Look at the full 1i21.pdb file and see how the sheet is formed.) Date: Thu, 1 Aug 2002 12:51:16 -0700 (PDT) From: yael@biology.ucsc.edu To: Kevin Karplus Subject: Re: t0152 In-Reply-To: <200208011923.g71JNDL21495@bray.cse.ucsc.edu> Thanks for the update, I have notice you have started to run try14 and I think the constraints in try14.constraints are very reasonable. I totally agree that there is one sheet. I was thinking maybe it would be possible to get a one full antiparallel sheet (which I now noticed is the way the sheet is also formed in 1i21) if the strand 43-47 is flipped a bit. To me it seems more likely then having a parallel connection between the two halfs of the sheet. Yael Yael Mandel-Gutfreund Department of Chemistry and Biochemistry University of California, Santa Cruz Phone : 1-831-459-5128 e-mail : yael@darwin.ucsc.edu Date: Thu, 1 Aug 2002 19:00:14 -0700 From: Kevin Karplus To: yael@biology.ucsc.edu cc: karplus@soe.ucsc.edu Subject: Re: t0152 Are you sure that 1i21 is a completely antiparallel sheet? I just switched at home from working on my MacOS X laptop to my OS9.2 desktop, and the SFTP client available for OS 9 is terrible---it won't let you specify a file by name---you have to download the directory it is in, which takes FORWEVER for the pc/data/pdb directory. I also find RasmMac's imitiations quite frustrating after using the UNIX Rasmol (scripts have to be in the directory as the executable, so far as I can determine). When I download a PDB file directly from the PDB web site, I can't get RASMac to recognize it! OK, finally managed to download 1i21.pdb. There is DEFINITELY a pair of parallel strands: G195-K130, I97-I132 in chainB. 1 Aug 2002 Kevin Karplus There IS a pair of parallel strands in 1i21.pdb. try14-opt is looking fairly good. It hbonds Y125 with L108---we could add that as a constraint, plus 123 with W110, which would transitively give us CB constraints on 123-M166 and 121-R164. I'm not certain of these though---maybe I'd better look at it again the morning and discuss it with Yael. Date: Thu, 1 Aug 2002 23:12:24 -0700 (PDT) From: yael@biology.ucsc.edu To: Kevin Karplus Subject: Re: t0152 > > OK, finally managed to download 1i21.pdb. There is DEFINITELY a pair > of parallel strands: G195-K130, I97-I132 in chainB. > There is a pair of parallel strands in 1i21 but that is not the pair that is connecting the two "two sheets". I myself hate the raswin at home, I don't get a command window at all so it is really hard to work with it but it looks like the parallel strands are at the edge of the sheet. Looking now at try14-opt it doesn't look easy to get a one sheet structure and even more impossible to get the two parts in antiparallel conformation. try14-opt looks actually not too bad Yael Fri Aug 2 01:46:34 PDT 2002 Kevin Karplus I looked at 1i21 again, and I'm certain that the pair of parallel strands corresponds to the interface betwee the two "sheets" that we have been modeling. I will proceed on that assumption. The proline P124 probably marks a bulge or other disruption on the last strand of the first antiparallel sheet. If D123 and Y125 make the H-bonds, then P124 could be in the strand, I guess, but it worries me. I've put together some new constraints, to try to get the sheets joined: // Add hbonding parallel constraints Y125-L108 and D123-W110 // WARNING: may be off-register by 1. Constraint 859 1014 2 2.7 3.5 // L108.N Y125.O Constraint 865 1004 2 2.7 3.5 // L108.O Y125.N Constraint 879 995 2 2.7 3.5 // W110.N D123.O Constraint 891 990 2 2.7 3.5 // W110.O D123.N // Add CB constraints on parallel strand pair // WARNING: may be off register by 1 or 2. Constraint 978 1305 3 5.12 7 // CBs D121 R164 Constraint 986 1316 3 5.12 7 // CBs A122 I165 Constraint 991 1324 3 5.12 7 // CBs D123 M166 Constraint 999 1332 3 5.12 10 // CBs P124 F167 I made the last constraint a bit looser, since P124 can't Hbond with the N atom. I also removed an Hbonding constraint on P93.N, since it makes no sense---proline N can't Hbond. From karplus@cse.ucsc.edu Fri Aug 2 09:34:15 2002 Date: Fri, 2 Aug 2002 09:34:14 -0700 From: Kevin Karplus To: yael@cse.ucsc.edu CC: karplus@cse.ucsc.edu Subject: another thought I'm beginning to think that we may have the second sheet trying to end up in the wrong place---perhaps the residues H130-I133 should be aligned with the second sheet. From karplus@bray.cse.ucsc.edu Fri Aug 2 14:49:18 2002 Date: Fri, 2 Aug 2002 14:49:16 -0700 From: Kevin Karplus To: yael@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: T0152 Although I don't think T0152 is finished, I'm going to submit a model (try15-opt) today. We can replace it if we come up with a better one. I think we should investigate moving the final sheet so that G128-I133 is parallel to R164-F167. This would require creating try16.constraints, removing the constraints at the end of try15.constraints: // Add CB constraints on parallel strand pair // WARNING: may be off register by 1 or 2. Constraint 978 1305 3 5.12 7 // CBs D121 R164 Constraint 986 1316 3 5.12 7 // CBs A122 I165 Constraint 991 1324 3 5.12 7 // CBs D123 M166 Constraint 999 1332 3 5.12 1 // CBs P124 F167 // Note: last possibly looser, since proline can't hbond with N atom. and adding CB constraints for the register we want to try. From yael@darwin.UCSC.EDU Fri Aug 2 14:55:41 2002 MIME-Version: 1.0 X-Authentication-Warning: darwin.UCSC.EDU: yael owned process doing -bs Date: Fri, 2 Aug 2002 14:55:03 -0700 (PDT) From: yael@biology.ucsc.edu To: Kevin Karplus Subject: Re: t0152 In-Reply-To: <200208021909.g72J9GZ01955@bray.cse.ucsc.edu> Content-Type: TEXT/PLAIN; charset=US-ASCII X-SpamBouncer: 1.4 (10/07/01) X-SBPass: NoBounce X-SBClass: OK try15-opt is now scoring the best though I try14 still looks better to me. The doesn't seem to be an improvment in getting the sheets attached. Now that I could really look at the structures of 1i21 and the alignments I agree that the sheets should probably be connected by a parallel interaction (BTW there is a typo in the README file the constranits Y-125-L108 and D123-W110 should be antiparallel) . To me it seems very likely that they should be connected through 120-124 164-167. In try15 it looks like the constraints helped somehow to get these two parts closer but the sheet didn't form (though there is a very week signal for a strand in the region 121-124). It is possible that H130-I133 (first sheet!) could be aligned with the second sheet... though it will leave a very long loop floating. Otherwise there could be a strand in-between 116-122 but these would move helix 144-155 position and give an all antiparallel sheet. I just saw your e-mail would you like me to work on try16 ? Yael Yael Mandel-Gutfreund Department of Chemistry and Biochemistry University of California, Santa Cruz Phone : 1-831-459-5128 e-mail : yael@darwin.ucsc.edu Fri August 2 Yael Runing try16 after changed the constraints for aligning H130-I133 to R164-F167 creating a parallel strand interaction. If that works well it may be helpful to add constraints to get strand 127-135 aligned antiparallel to 102-107 (3rd and 4th strand of sheet 1) 4 Aug 2002 Kevin Karplus The try16 run looks poor to me---the first sheet has been pulled apart without getting any closer to the second sheet. Date: Mon, 5 Aug 2002 11:29:39 -0700 (PDT) From: yael@biology.ucsc.edu To: Kevin Karplus Subject: Re: t0152 Hi Kevin, Unfortunately we had problems with our DSL connection during the weekend so I could not do much work. I saw try16 and it does look very poor. I am not sure if it is worth while trying to get the two sheets together. I will look at it again later today (must finish something now) Let me know if there is anything you would like me to try. Hope you are enjoying the conference Yael Yael Mandel-Gutfreund Department of Chemistry and Biochemistry University of California, Santa Cruz Phone : 1-831-459-5128 e-mail : yael@darwin.ucsc.edu From karplus@bark.cse.ucsc.edu Mon Aug 5 19:17:43 2002 Date: Mon, 5 Aug 2002 19:17:42 -0700 From: Kevin Karplus To: yael@biology.ucsc.edu Cc: karplus@soe.ucsc.edu In-reply-to: (yael@biology.ucsc.edu) Subject: Re: t0152 I'm not sure what to do with t0152. I submitted try14-opt before leaving, just in case. I'm not sure what else to try to improve the model. I suspect that there is one sheet, but I don't know how to form it.