Wed Jun 9 10:20:51 PDT 2004 T0200 DUE 4 Aug 2004 Wed Jun 9 22:29:54 PDT 2004 Kevin Karplus This looks like a barrel? try1 messed it up a bit, but it didn't have the strand constraints. Let's try again with lots of alignments and with the strand constraints. Thu Jun 10 12:32:26 PDT 2004 Kevin Karplus Looks more like a sandwich. There are beta hbonds in each of the alignment models that look pretty good---perhaps I should gather them as constraints and see if all the pieces can be put together consistently. There may be some conflicts, so bonus_constraints may be the way to go. I can get the hbonds by modifying the superimpose-best.under script. The hbonds are NOT all consistent---there are register shifts on the strands. What I'll do is include all the constraints for try3.costfn (using bonus_constraints) and see how the model looks when done. I'll throw out the inconsistent ones for try4---perhaps just selecting those hbonds that are kept in try3. Fri Jun 11 00:28:54 PDT 2004 Kevin Karplus The alignment in try3-opt2 looks completely wrong---the secondary structures don't line up with the predictions at all. I'll have to go through the hbonds one strand at a time, removing any that are ludicrous, and keeping the reasonable ones. For example, E154-W160 should be helix, rather than a strand. Perhaps G141-H148 belongs in its place, since it should be strand rather than helix. Fri Jun 11 09:42:26 PDT 2004 Kevin Karplus On try3, the N-terminal results from about Q12 to R118 look fairly good. It is alignments past there that are really suspect (and the first strand). Fri Jun 11 16:51:08 PDT 2004 Kevin Karplus On try4, the results are looking good for most of the structure. Strand R165-I170 needs to be placed yet. It probably runs parallel to T142-H148 141: gTVFVDfh 164: gRVAAVIgthth The next strand is probably also parallel: 141: gTVFVDfh 164: gRVAAVIgthth 187: gTAYQTDagf The conserved histidines H148, H173, H175 need to be close, I think. H66 and N65 are probably also close. Try4 did move these very close, so I may need to loosen the constraints on them a bit. Sat Jun 12 07:20:21 PDT 2004 Kevin Karplus In try5-opt2, strand G99-N106 seems to be messed up and K243-I253 has been wound into a helix. I've put in extra hbonds taken from try4-opt2, to keep this from messing up again. It is annoying to have to keep adding explicit Hbonds to keep the already built good parts from getting messed up in new tries. After try6, if most parts are good, I'll try reoptimizing from all existing models with high crossover. Sat Jun 12 19:32:12 PDT 2004 Kevin Karplus A lot of try6 is good, but A157-V169 is still messed up as is A189-G195. M158-A163 should probably be parallel to T142-A149, maybe something like 140> lgTVFVDfha 157> aMGWHLa 171< gIVAAVRg Also, looking at the histidines, it seems that there are enough for TWO metal-binding sites--perhaps at H46,H66,H69,H70, and at H148, H161, H173, H175. My interpretation may be wrong here, as there are a log of histidines on one face of the cleft in the top-scoring homology template (1hp1A). I've made enough changes to the cost function to make another run from the alignments. If it doesn't produce a convincing structure, I should go back to 1hp1A and 1ush to see how their beta sandwiches are structured, and try to map strands onto the structures. Sun Jun 13 08:43:00 PDT 2004 Kevin Karplus Try7 is getting better, but is still a bit messy E150-A163 should probably be a clean helix, but I took out the helix constraint because I had a conflicting strand constraint from the str2 network. Maybe I need to generate some more alignments to the domain that I believe to be correct, using different secondary structure tracks---perhaps I can get a better alignment that way. Since t2k and t04 alignments have different key residues, I should probably try alignments from both. It might also be a good idea to try to clean up the two alignments to get a better multiple alignment. The t04 one is more diverse, so (unless it is contaminated) may better reflect the structural constraints. H200 and H221 should probably have constraints on them to move them onto the same end of the sandwich as the other histidines. And maybe H161 DOESN'T belong with the rest. Sun Jun 13 19:57:56 PDT 2004 Kevin Karplus The comparative model in T0200 is turning out to be somewhat recalcitrant---I think I (or whoever takes over the lead on it) needs to spend some time staring at the structures for the templates it will be modeled on, and perhaps creating some hand alignments. I suspect that the neural net predictions of the local structure are not super quality for this target, since they disagree on some crucial pieces. Mon Jun 14 11:03:04 PDT 2004 Kevin Karplus I looked at the metaserver http://bioinfo.pl/Meta/status.pl and searched for T0200: Everyone agrees on the superfamily, and the most common template seems to be 1ush. (Note this is our second best hit, slightly behind 1hp1A.) Most of the alignments have trouble at the C-terminal end of the protein, but one to 2ushA that 3d jury selects out of the 3DPS predictions looks pretty good on the web page. Unfortunately, when expanded with all the missing insertions, it is pretty dubious: >T0200 .........MRVLFIGD........VFGQPGRRVLQNHLPTIRP.....QFDFVIVNMENSAGGFG.....MHRDAARGALEAGAGCLTLGNHAWHHKDIYPMLSEDTYPIVRPLNYADPGT.PGVGWRTFDVNGEKLTVVNLL.........GRVFMEAVDNPFRTMDALLERDDLGTVFVDFHAEATSEKE.........AMGWHLAGRVAAVIGTHTHVP..............TADTRILKGGTAY.QTDAGFTGPH.........DSII....GSAIEGPL.......QRFLTERPHRYGVAE.............................................GRAELN.............GVALH......FE................................................GGKATAAERYRFIED >2ushA yeqdktyklITVLHTNDhhghfwrnEYGEYGLAAQKTLVDGIRKevaaeGGSVLLLSGGDINTGVPesdlqDAEPDFRGMNLVGYDAMAIGNHEFDNPLVLRQQEKWKFPLLSANIYKSTGErLFKPWALFKRQDLKIAVIGLTtddtakyftDIEFRKPADEAKLVIQELQQTEKPDIIIAATHMGHYDNGEhgsnapgdVEMARALPGSLAMIVGGHSQDPvcmaaenkkqvdyvPGTPCKPDQQNGIwIVQAHEWGKYvgradfefrNGVLytpeIAENQQMIsllspfqNKGKAQLEVKIGETNlegdrdkvrfvqtnmgrlilaaqmdrtgadfavmsgggirdsieaGRGDISynkvlkvqpfgnvVVYADmtgkevIDyltavaqmkpdsgaypqfanvsfakkikgepvdpaktyrmatlnfnatDNKPVNTGFISW--- The first part is in pretty good agreement with the other alignments I've bene using, but they diverge pretty far toward the C terminus. I'll add this alignment to show-align.under to see whether it is any good. If it is, I'll toss it into the hopper with the rest. It seems to do ok up to about P199. In fact, it has a better first strand than in some of the alignments that sam-t02 identified, but I've already fixed that up in the models we're generating. In short, it doesn't look like the meta servers are going to provide much help on the C-terminal end. Fri Jun 25 18:30:27 PDT 2004 Kevin Karplus News from CASP: T0200 -- dimetal center-- two Manganese ions Mon Jul 19 15:49:08 PDT 2004 ggshack Try8 on caw. I'm trying a basic run trying out the RR constraints. Running using the bonus_constraints option. I'll take a look at what we get. I'm not so sure this isn't a barrel. Restarted without SCWRL. Tue Jul 20 10:00:25 PDT 2004 ggshack Sure keeps looking like a barrel. This time the bonus_constraints have left us with a big change in assigned helices and strands. This is a prime choice to try without bonus_constraints. Sure looks like a barrel... Doing TRY9 on caw, no bonus this time. Tue Jul 20 20:07:46 PDT 2004 ggshack Ok, try9 has possibilities. Sort of makes sense. Now I'll crank up the cost of break and see what we get. Also shoved phobic up a notch. Started TRY10 on caw. Wed Jul 21 11:07:44 PDT 2004 ggshack Try10 looks better with all attached but I don't like the remaining loops or at least a couple of the strands. I think I'd like to have Kevin look at it now. After looking at try10 with Kevin, I am defining a couple of new helices and one strand. Will give it another run. TRY11 now running on crow. For TRY12 I increase to: StrandConstraint T188 T192 20.608 HelixConstraint I208 L216 20.702 Thu Jul 22 11:04:21 PDT 2004 ggshack Try12 crashed but did get to opt1. It got rid of the faulty strand. I am going to enforce a couple of "new" helices 152-158+, 70-78+. Started TRY13 on crow. Kevin told me to get new alignments and best alignment hits for the 'parent' list: set MANUAL_TOP_HITS macro make -k extra_alignments then remake all_align.a2m.gz (by removing and doing a make??) Fri Jul 23 14:32:56 PDT 2004 ggshack Results of try13 were awful. Need to do what Kevin suggested. Ok, now I have a new 'all-align.a2m.gz.' I'll do TRY14 hopefully based on new frags and aligns. I'm going to back off of the higher enforced values that I had added for the earlier runs but I'm going to give the RR constraints higher values and use the 'bonus' option. Try14 started on croak Fri Jul 23 21:42:10 PDT 2004 Kevin Karplus With an unconstrained costfcn, try14-opt2 scores the best of any of the models. It has some problems: notably helices where strands should be and strands wound up into helices. One of the simplest fixes is to slide predicted strand T142-D146 where A133-R137 currently is sitting. I'll change the sheet constraints to do that. I think that mode sliding is probably needed, pushing out an insertion, so that the Cterminal strands come into the barrel and some of the dubious strands are pushed out. From ggshack@pacbell.net Sat Jul 24 00:04:23 2004 MIME-Version: 1.0 From: George Shackelford To: Kevin Karplus Subject: T0200 try15 Date: Sat, 24 Jul 2004 00:04:21 -0700 Content-Disposition: inline X-Spam-Checker-Version: SpamAssassin 2.63 (2004-01-11) on fs.cse.ucsc.edu X-Spam-Level: X-Spam-Status: No, hits=0.0 required=3.0 tests=none autolearn=no version=2.63 I'm glad to hear that try14 did well. I notice the changes for try15 corrects the use of earlier RR predictions and replaces the constraints from t2k with those from t04. I must be getting fuzzy in the head that I didn't notice those. I can't wait to see the results but it will be tomorrow morning. Thanks for helping out. - George From ggshack@pacbell.net Sat Jul 24 08:47:05 2004 MIME-Version: 1.0 From: George Shackelford To: Kevin Karplus Subject: T0200 try15 Date: Sat, 24 Jul 2004 08:47:02 -0700 Content-Disposition: inline X-Spam-Checker-Version: SpamAssassin 2.63 (2004-01-11) on fs.cse.ucsc.edu X-Spam-Level: X-Spam-Status: No, hits=0.0 required=3.0 tests=none autolearn=no version=2.63 It looks really nice, except that it's a barrel. The RR constraints say 'barrel' while the alignment structures say 'sandwich.' I've seen this barrel before. It just won't go away. - George Sat Jul 24 12:15:52 PDT 2004 Kevin Karplus The barrel is not looking too bad. We might want to cluster some of the conserved HIS residues. Hmm, we have constraints for that, but they are fairly weak. I don't like putting H161 in with H148,H173,H175---it's at the wrong end of the helix. I'd rather throw H66 in with H148,H173,and H175, and take it out of the H69,H70,H46 cluster. The strand I slid over, T111-N114, seems to be happier where it is, but I'd like to move one pairing a bit so that the conserved F144 pairs with V112. I created a "strands" file with the following definitions: define s1 1-7 define s2 31-36 define s3 60-63 define s3a 84-87 define s4 101-105 define s5 108-114 define s6 118-122 define s7 142-147 define s8 167-171 define s9 181-184 define s10 189-192 define s11 223-226 define s12 230-239 define s13 243-251 define beta s1 or s2 or s3 or s3a or s4 or s5 or s6 or s7 or s8 or s9 or s10 or s11 or s12 or s13 The sheet constraints for the TIM-barrel that we are currently favoring are SheetConstraint M1 I6 T188 D193 hbond T188 10 # s1 || s10 SheetConstraint R2 I6 D30 V34 hbond F5 10 # s1 || s2 SheetConstraint D30 M36 C60 H66 hbond I33 10 # s2 || s3 SheetConstraint T62 N65 I84 P87 hbond L63 10 # s3 || s3a SheetConstraint I84 P87 K109 V112 hbond V85 10 # s3a || s5 SheetConstraint T111 N114 V143 D146 hbond V112 10 #s5 || s7 SheetConstraint T142 D146 A167 G171 hbond V145 10 # s7 || s8 SheetConstraint V169 G171 A189 Q191 hbond I170 10 # s8 || s10 The problem with this model is that s4, s6, and s9 get mapped to helices, and that the constraints are not consistent with the slip of a TIM-barrel, unless I change the pairing: s10 gtAYQTDa s1 MRVLFIGdv s2 fdFVIVNMe s3 gCLTLgnh s3a pIVRpl s5 gEKLTVVNllg s7 gTVFVDm s8 grVAAVIGt s10 gtAYQTDa s1 MRVLFIGdv For try16, I'll try to form this TIM-barrel, but for try17, I'll try to reform it so that the skipped strands get used to form sheet. Sat Jul 24 13:35:18 PDT 2004 Kevin Karplus George had started to edit try16.under, but not gotten very far, so I've taken it over. From: George Shackelford To: Kevin Karplus Subject: Re: T0200 Date: Sat, 24 Jul 2004 16:23:15 -0700 In-Reply-To: <200407242038.i6OKcRoR005739@cheep.cse.ucsc.edu> Sorry, I had to do some errands and then I had a hardware problem. It's fixed now and my system is running again. I'll look at what you have for 16 and 17. I was going over the rr constraints and getting new ones that covered more ground, just in case that might hint at the sheet connections. Are you pursuing the barrel configuration? Is one of the the new trys for the sandwich? What would you suggest I try now so we don't duplicate efforts? - George On Saturday 24 July 2004 01:38 pm, you wrote: > Since I couldn't reach you by phone or e-mail, and you didn't appear > to have done any editing yet on try16.under, I took it over. > I'll also do a try17.costfcn and try17.under. > > If you want to try something new, start with try18. Date: Sat, 24 Jul 2004 16:31:02 -0700 From: Kevin Karplus To: ggshack@pacbell.net CC: karplus@soe.ucsc.edu In-reply-to: <200407241623.15682.ggshack@pacbell.net> (message from George Shackelford on Sat, 24 Jul 2004 16:23:15 -0700) Subject: Re: T0200 As the README says, try16 is working on the barrel, and try17 is trying something (anything!) else. It isn't specifically trying to get a sandwich, but some of the barrel constraints have been removed, and the anti-parallel sheets are treated as separate objects. I may not have anything consistent for try17. Go ahead and try to create a sandwich, if you have an idea how it should go. From: George Shackelford To: Kevin Karplus Subject: Re: T0200 Date: Sat, 24 Jul 2004 17:00:50 -0700 In-Reply-To: <200407242331.i6ONV2So007720@cheep.cse.ucsc.edu> T0200.try16-opt1.pdb is already there and if it's any indication, that doesn't seem to work. I don't believe this is a sandwich. I see some indication from pairs that it is s10 || s2. Will look at that. I don't like that the start of sequence is going to be part of the sheet. - George Sat Jul 24 17:03:25 PDT 2004 George Shackelford I see some indication from pairs that it is s10 || s2. Will look at that. I don't like that the start of sequence is going to be part of the sheet. also redefining I84-P87 to 85-88 connect 63 to 88 got s10 to match to s2. Now running TRY18 on peep. Sat Jul 24 19:49:58 PDT 2004 Kevin Karplus George says "Try17 doesn't look too bad. Some of the sheets are facing the wrong way." Sat Jul 24 20:39:39 PDT 2004 Kevin Karplus try16 is finally getting away from the TIM barrel (though I THOUGHT I was encouraging the TIM barrel. try17 is starting to look like a sandwich,though s9 is curled up into a helix. s8 is facing s1 (which is next to s10), so s9 should probably be facing s10, not hbonded to it. I'm not sure where s11, s12, and s13 go, but I can try to fix up s9. One possiblity: s9 ^v s8 and the final antiparallel sheet fits on the end of the sandwich, maybe bending to meet both sheets of the sandwich. I won't try to force this on try19, but if it looks feasible, I may try it on the next iteration. Sat Jul 24 21:16:50 PDT 2004 George Shackelford Try18-opt1.pdb is out and it looks good (except for the breaks). Still need to do some work on lining up the strands. I wish I had Jenny or Kevin's experience with barrels. Should start working on try20 based on what I see. Sun Jul 25 00:11:05 PDT 2004 George Shackelford Still working on getting the best alignments. Undertaker insists on putting M1-I6 as part of the barrel for 8 strands. I'll have to finish it tomorrow. Sun Jul 25 10:17:19 PDT 2004 George Shackelford I think I've got it worked out. Everything is now one residue offset thru eight residues. Trying Y190 Q191 T192 D193 A194 G195 V003 L004 F005 I006 G007 D008 F029 D030 F031 V032 I033 V034 L061 T062 L063 G064 N065 H066 I084 V085 R086 P087 L088 L110 T111 V112 V113 N114 V145 D146 F147 H148 A149 E150 A168 V169 I170 G171 T172 A189 Y190 Q191 T192 D193 A194 V003 as TRY20. Sun Jul 25 11:00:32 PDT 2004 Kevin Karplus The best scoring with the unconstrained costfcn are try14-opt2, try19-opt2, try15-opt2, try18-opt2, try8-opt2. try14 looks ok for a barrel try19 looks pretty good for a sandwich---it scores better than try14 with the try14 costfcn. try15 looks pretty good for a barrel try18 looks like our currently best barrel--scoring better than try15 or try14 with the try15 costfcn. try8 is a terrible barrel. Rosetta, perversely, likes try8-opt2.repack-nonPC best, then try14, try13, try18, try15, try19, ... but it really hates all of them. I'll set up a try21 (so as not to step on George's try20), polishing up the try15/try18 barrel. Nope---after playing with it for a while, it seems not worth my time to make a small improvement---I'll move on to other targets. Looking at the superposition of the best models so far, it looks like we could improve the sandwich by discarding s1 || s10, and adding s1 ^v s12 maybe M1-F239 through G7-N233 hbonding v3. We could add s10 || s13 or s10 || s8. I'll put together try21 with s1^vs12 and the s10 || s8 option and see what happens. Sun Jul 25 17:49:07 PDT 2004 Kevin Karplus Try21 did not succeed in replacing s10 by s12, and I didn't really expect that it would. To do that right would probably require going all the way back to the alignments with the templates and generating some new alignments that did what we wanted. The top scorers with unconstrained.costfcn are try14, try19, try21, try15, try18, try13, try8, try17, try3, ... With the "strands.costfcn", which has constraints for the strands and helices, but not for the sheets or predicted contacts, the order is try19, try21, try18, try15, try14, try17, try13, try9, ... Unless George's try20 comes up with something golden, I think we'll go with try19 top sandwich try18 best barrel try8-opt2.repack-nonPC Rosetta's favorite try1-opt2 automatic top model from alignment. Question: should we include try9, which seems to score best in several of the early score functions? Sun Jul 25 18:17:25 PDT 2004 George Shackelford I looked at try9 with 'near' and spacefill and it doesn't really look all that nice. I don't think I would submit it. Sun Jul 25 18:56:54 PDT 2004 Kevin Karplus Correction to above: Rosetta likes best try3-opt2.repack-nonPC which is a terribly mangled sandwich. I don't think try9 is all that bad---I wouldn't put it FIRST, but we might want to include one like it in our predictions. At least both the long hairpins are in the sandwich, something none of our current favorites have. From ggshack@pacbell.net Sun Jul 25 20:19:37 2004 MIME-Version: 1.0 From: George Shackelford To: Kevin Karplus Subject: try20 has finished Date: Sun, 25 Jul 2004 20:19:35 -0700 I looked at the T0200.try20-opt2.pdb and I am not overwhelmed by it. There is still a misalignment of a strand or two (about 147). I'd love to take a bit more time with it, but I figure that you need to pack things up. You can look at it and decide whether to go with it or try18. ------------------------------------------------------------ Sun Jul 25 20:39:44 PDT 2004 Kevin Karplus With the try20 score function, try19 score best, then try21, try18, and finally try20, so I'm not sure there is much point to including try20. With the unconstrained cost function the order is try14, try19, try21, try15, try20, try18, try8, ... With the strands cost function the order is try19, try21, try18, try20, try15, try14, try17, ... Sun Jul 25 22:53:54 PDT 2004 George Shackelford I am giving the barrel one more try: TRY22 running on peep. This time I am going to turn strand 145(etc) around. I think that might work. Can't hurt. Probably don't have enough time. Mon Jul 26 07:49:59 PDT 2004 George Shackelford Ok, try22 shows I've learned how to fix up barrels. I'm pleased. On the other hand when I look at rr.distinct, nothing is really working. I'm not pleased. I feel further from what I want. I'm going back to the sheets that Kevin first proposed and see if there is any nice adjustments I can make to improve those results. Mon Jul 26 13:10:06 PDT 2004 Kevin Karplus try18 or try22? maybe both (dropping the try3) (22 before 18) Maybe try21 instead of try19. Mon Jul 26 16:53:42 PDT 2004 Kevin Karplus I'm hard pressed to distinguish between try 21 and try19. The biggest difference seems to be that try21 brings the final hairpin in closer to the sandwich (causing clashes in the process). I think I'll stick with try19. I'll submit try19-opt2 best sandwich try22-opt2 best barrel try18-opt2 another good barrel, with different C-terminal sheet try1-opt2 full auto undertaker model 1 Thu Nov 18 22:10:33 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 37.7260 (it's the try1 version of our model 1: try19-opt1) best submit 37.7122 (model 1) model1 37.7122 auto 31.8224 align 34.9732 robetta best 35.8422 (robetta model2) robetta1 30.1387