Tue Jun 8 08:56:26 PDT 2004 T0197 DUE 2 July 2004 Tue Jun 8 09:31:09 PDT 2004 Kevin Karplus This looks like a distant fold-recognition target, or maybe even ab initio. I'll want to rerun the fold-recognition once we've added new neural nets for t04, since the e-values are not real strong. The top folds are b60.1.1, d63.1.1 and d32.1.[23]. Tue Jun 8 12:19:04 PDT 2004 Kevin Karplus The try1 model has bits and pieces of a structure, but no real consistency. For try2, I'll add helix and strand constraints from dssp-ehl2.constraints, and any .mi constraints that look plausible. I also added a strand constraint for the HIS tag, to make it be in extended conformation. Tue Jun 8 16:55:26 PDT 2004 Kevin Karplus In try2, I'm beginning to see a beta sheet. I should try improving the constraints to try to encourage the topology that I'm starting to see. Not right now though---I have to go to a school end-of-year picnic. Tue Jun 8 22:25:30 PDT 2004 Kevin Karplus Some of that beta sheet I was seeing was illusion---the initial his tag ended up forming a beta hairpin with the next few residues---an unlikely real structure. Also R55-H58 is predicted to be helix, but is being stretched out into a strand. Perhaps the first strand G8-K16 should run anti-parallel to either Y63-F67 or to an unpredicted strand around E40-E45. There are a lot of D and E residues lined up on one side of the hairpin. Perhaps E10 and E12 line up antiparallel with E84 and E82. And A91 and V94 on the helix should pack on the other side of the sheet, perhaps near Y63 and L65 respectively. Of the try2 constraints, I like V92-I102 (helix against back of sheet) I like K62-G86 replace with K62.O hbonded to G86.N I don't like F67-E77 way out of register I don't know about N74+N75 - N116 N74+N75 are on a hairpin turn, not clear whether N116 belongs nearby. This one goes with N75-T119. I'm not sure about D87-T105---both seem to want not to be buried. E69-F78 are close, but on opposite sides of sheet. Perhaps use hbond I70.N-F78.O I'm not sure about P49--F78 or P49-E69. It looks feasible if the helix packs on the back of the sheet. I like K68-E80 but replace with antiparallel Hbonds. I don't know about L71-R165. The beta-hbond terms were probably too strong in try2, since the somewhat weakly predicted helix at R55-H58 was unwound into a strand. Maybe I should add a weak helix constraint for P49-H58 (replacing the L51-G56 constraint). For try3, I'm moving to a try3.cost file that includes both the cost function and the constraints. This should make it easier to update the score-all.rdb file when needed. Wed Jun 9 10:40:19 PDT 2004 Kevin Karplus try3 crashed with an assertion failure. I'll have to try to debug this! Unfortunately, it ran for over two hours before crashing, so debugging will be a major pain. The place that failed (copy_from_frag) is mainly used in constructing single-segment AlignedFragments records from fragments, so I don't know why it is even being called! I added a couple of extra assertions, and will try running again. Wed Jun 9 14:29:48 PDT 2004 Kevin Karplus try3 completed, and the results look ok---the long hairpin is still all we have, but it looks like the Nterminal strand may be trying to pair up with Y63-Y79. Of course, Q39-P47 may be an intervening strand parallel to Y63-Y79. I'll try setting up the N-terminal strand antiparallel to Y63-Y79, with I13 and F15 lining up with L65 and Y63 (and Hbonding). Wed Jun 9 16:24:28 PDT 2004 Kevin Karplus Rats! try4 also crashed after an hour, with the same copy_from_frag assertion failure. Fri Jun 11 16:17:01 PDT 2004 Kevin Karplus The metaserver at bioinfo and be queried at http://bioinfo.pl/Meta/status.pl (search for the target name). The 3d Jury consensus seems to be that 1q0cA, 1f1yA, 1eilA, and 1dhy are the appropriate templates. (d.32.1) I have a few hits from this superfamily (1cjxA, 1ewjA, and 1ecsA), but they don't result in full-length alignments like 3D-jury is claiming. They don't generally run to the N terminus but start after 80. I don't like the turn at K107-K108-K109, which is predicted to be strand, but I'm not sure how to fix the problem---who does the strand pair with? Sat Jun 12 09:28:21 PDT 2004 Kevin Karplus I need hbonds for T119, E121, N123 probably with D133 and E135. Let's guess that N123 is the unpaired one. Sat Jun 12 18:07:02 PDT 2004 Kevin Karplus In try5, I'm now getting two sheets, which look mergeable into a single sheet. The strand around G99-K108 is not hydrogen bonding on either side---I might need to play around with the phase of that strand. It almost makes the antiparallel connection to R110-L115, but it really misses the parallel connection to the parallel sheet. It looks like V42-P47 should also be a strand of the parallel sheet. Wed Jun 16 11:12:48 PDT 2004 Kevin Karplus I think I need to add some Hbond constraints to pull the sheet together. It looks like I should pair V81, F83, I85 with I102, A104, I106, with the Hbonds being on the V81,F83,I85. Problem: the good hairpin centered at L115-V118 hbonds the buried side of the strand R110-L115, so orienting the mixed strand K101-K109 seems a bit difficult. I want the same residues (Q103, T105, and K107) for hbonding on both sides of the strand. Another problem-- I was thinking of packing the helix against the hydrophilic side of F78-I85. Not good thinking! It might make more sense to have F78-I85 as an edge strand and put K62-E69 against K101-L115. Nope---I'd really like a parallel connection between Y63, L65, F67 and preceding strands: 10: EVEIKFK 30: npEFVRYEeqe ?? 60: lkkYYLTFKEILde Wed Jun 16 21:18:01 PDT 2004 Kevin Karplus try6 died with an uncompleted error message: Error: Atom which is most likely from AlignedFragments::copy_from_frag getting a bad atom number from somewhere. I'm going to skip try6 and do try7 without scwrling the alignments. I think there may have been some trouble because I killed a SCWRL that was running much too long. Thu Jun 17 12:30:59 PDT 2004 Kevin Karplus oops---there were SCWRL commands in included files. I've cleaned that up by making some changes to the scripts and the Makefile (introducing NO_SCWRL macro). Don't forget to remove the frozen SetSeed in the next run! Thu Jun 17 17:07:20 PDT 2004 Kevin Karplus try7 finally finished without error. I'm seeing a good hairpin and a fair 4-strad sheet, which I think I want to join into a single sheet. The big question is whether the hairpin that is trying to form centered at K107-R110 is real, or whether there is just a bulge there. A bulge to match the one in the adjacent strand might be nice, as it would produce a long edge that could be matched by the long preceding hairpin. I think I want to have parallel strand with V81 and F83 matching A104 and I106 or perhaps 2 later. I can Hbond off of E77, Y79, V81, F83. and I85. 76: eEFYEVEFEIg v v v v 100: fKIQATIKKKRWVYKLng Let's try the pattern above and see what happens. Sun Jun 20 19:07:12 PDT 2004 Kevin Karplus Try8 seems to have gotten a little of the parallel strand I was trying for but at the cost of a lot of breaks and ugly snippets. I'm not sure what to do to this mess. Mon Jun 21 07:52:53 PDT 2004 Kevin Karplus I rewrote the key-to-rasmol script, so that the conserved_t2k and conserved_t04 scripts produce set definitions. I can now highlight strongly conserved residues by doing "select conserved_t2k and conserved_t04" and fairly conserved residues with "select conserved_t2k or conserved_t04". Looking at the strongly conserve residues in try8-opt2, I think I want F44 and I134 to be neighbors in a beta sheet (rather than Y43 and I134). I think L170 on the helix needs to pack near them also. This should help D41, which needs to stay on the polar side of the sheet. I'm not sure where the conserved polars E10, E12, R110, N123, and Y43 go---whether they are clustered or just spread out on the outside of the structure. Mon Jun 21 14:40:31 PDT 2004 Kevin Karplus The strand-helix-strand for Y79-K107 is looking pretty bad--it is trying to make an S-twisted (lefthanded) conformation, which is quite rare. Maybe I have the strand order wrong and across the sheet should be V81-F67-A104 and F83-L65-I106. No--I don't like that---it looks like there is an antiparallel sheet with L51-V57, Y63-E69, F78-E84, and an antiparallel sheet I102-L114, V118-V125, F131-I137. It makes more sense to have a parallel connection between F78-E84 and the first part of I102-L114. I've put in a different set of constraints for try10, but my head is a bit fuzzy today, so I'm not sure I put in consistent constraints. Mon Jun 21 19:57:46 PDT 2004 Kevin Karplus Hey, try10 is starting to look like a real protein! I need to figure out where strand S9-I17 goes, and fix up F78-E84 to pair better with the previous strand Y64-D72. I'll give up on pairing it with the next strand, which I'll continue pairing Y113-L120. I'll try optimizing again, with minor changes to the cost function, removing constraints that were badly violated and adding constraints for extending the pairing Y113-L120. I'm not sure where to put the first strand yet, so I'll leave it alone for now. Tue Jun 22 07:22:48 PDT 2004 Kevin Karplus try11 looks like a beta sandwich. Problems: 1) strand A104-W111 was inconsistently specified, so is not well-placed. 2) strand G8-K15 is wound into a helix. 3) beginning of stran V34-Q39 is wound into a helix 3) P49-H58 is predicted to be helix, but is used as a strand. (this strand is requested in the cost function) Tue Jun 22 09:03:20 PDT 2004 Kevin Karplus The constraints I'm coming up with are very similar to those of the second alignment (1d8hA). If I start liking that, then I may want to change some of the initial alignment set to just d.63.1.1 (1d8hA, 1d8iA). Tue Jun 22 15:55:18 PDT 2004 Kevin Karplus Phil Bradley suggested that since many of the top homologs were Adenylate cyclase, that the "adenylyl cyclase" superfamily (a ferredoxin-like fold) would be good candidates. This is domain d.58.29, and do not score well in our test. The best in the template library is 1azsA with an E-value of 27.47 (rank 149). There are some hits to the same fold in the score files, and even one to that superfamily (1ab8A). I'll add it and the top few d.58 hits (1kr4A, 1ftrA, 1aw0) to the MANUAL_TOP_HITS list and to the alignment set that gets sampled. Tue Jun 22 17:51:02 PDT 2004 Kevin Karplus On try12, which does not yet have the ferredoxins, the top-scoring model without sheet constraints is for try3---a rather ugly model. I'm going to try another run without sheet constraints but including the ferredoxins, to see if anything decent comes up. Based on the few alignments I saw, I'm not too hopeful. I'll toss in the robetta models also, since they are scoring badly enough not to really outcompete the alignments (I think). Wed Jun 23 08:57:47 PDT 2004 Kevin Karplus On try13, I'm now getting a very different structure. It has a two-fold repeat: S1 H S2 S3 S4 S5 with sheet S5v S1^ S4^ S3v S2^. On the first repeat, strand S5 has not attached to the sheet and strands 2 and 4 are predicted to be helix by some of the neural nets. To attach S5 of the first sheet, I need an antiparallel constraint between S9-F15 and F67-K61, with E10 hbonding. Wed Jun 23 11:33:37 PDT 2004 Kevin Karplus The model that came up on try13 was coming from robetta, which is based on 1q0cA (d.32.1.3). We have 1cjxA, 1ewjA, 1ecsA, and 1f1xA in this family (our 3rd best fold family). The Robetta model takes a weird subdomain-swapped alignment though, so I don't really like it. Wed Jun 23 14:25:50 PDT 2004 Kevin Karplus I implemented PrintConformSheets in undertaker today, and have the following constraints: REMARK 44 model 1 is called T0197-1gm6A-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m:1gm6A T0197.t2k.undertaker-align.sheets1 SheetConstraint (T0197)R55 (T0197)N59 (T0197)K68 (T0197)Y63 hbond (T0197)R55 SheetConstraint (T0197)Y63 (T0197)I70 (T0197)F83 (T0197)N75 hbond (T0197)Y64 SheetConstraint (T0197)K101 (T0197)K107 (T0197)K108 (T0197)F100 hbond (T0197)K101 SheetConstraint (T0197)T105 (T0197)V112 (T0197)E126 (T0197)V118 hbond (T0197)I106 SheetConstraint (T0197)V112 (T0197)K114 (T0197)L120 (T0197)V118 hbond (T0197)Y113 SheetConstraint (T0197)V118 (T0197)E126 (T0197)S138 (T0197)G129 hbond (T0197)L120 REMARK 44 model 2 is called T0197-1d8hA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m:1d8hA T0197.t2k.undertaker-align.sheets2 SheetConstraint (T0197)D41 (T0197)F44 (T0197)E69 (T0197)L65 hbond (T0197)V42 SheetConstraint (T0197)K62 (T0197)Y64 (T0197)E80 (T0197)F78 hbond (T0197)Y63 SheetConstraint (T0197)T105 (T0197)L115 (T0197)G129 (T0197)V118 hbond (T0197)K107 SheetConstraint (T0197)G117 (T0197)V122 (T0197)S138 (T0197)V132 hbond (T0197)T119 REMARK 44 model 3 is called T0197-1cjxA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m:1cjxA T0197.t2k.undertaker-align.sheets3 SheetConstraint (T0197)A104 (T0197)K109 (T0197)V125 (T0197)T119 hbond (T0197)I106 REMARK 44 model 4 is called T0197-1ewjA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m:1ewjA T0197.t2k.undertaker-align.sheets4 SheetConstraint (T0197)E82 (T0197)I85 (T0197)L120 (T0197)V125 hbond (T0197)F83 SheetConstraint (T0197)F100 (T0197)Q103 (T0197)L115 (T0197)W111 hbond (T0197)K101 SheetConstraint (T0197)W111 (T0197)L115 (T0197)N123 (T0197)V118 hbond (T0197)W111 REMARK 44 model 5 is called T0197-1ew3A-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m:1ew3A T0197.t2k.undertaker-align.sheets5 SheetConstraint (T0197)K62 (T0197)L71 (T0197)F83 (T0197)N74 hbond (T0197)Y63 SheetConstraint (T0197)W111 (T0197)K114 (T0197)V122 (T0197)V118 hbond (T0197)V112 SheetConstraint (T0197)V118 (T0197)V122 (T0197)S140 (T0197)I134 hbond (T0197)L120 REMARK 44 model 6 is called T0197-1l9aA-t2k-global-adpstyle5.a2m:1l9aA T0197.t2k.undertaker-align.sheets6 SheetConstraint (T0197)N59 (T0197)K61 (T0197)E121 (T0197)T119 hbond (T0197)L60 SheetConstraint (T0197)E82 (T0197)E84 (T0197)L65 (T0197)F67 hbond (T0197)F83 SheetConstraint (T0197)I102 (T0197)I106 (T0197)V122 (T0197)V118 hbond (T0197)Q103 REMARK 44 model 7 is called T0197-1ku9A-t04-global-adpstyle1.a2m:1ku9A T0197.t2k.undertaker-align.sheets7 SheetConstraint (T0197)L71 (T0197)E73 (T0197)N123 (T0197)E121 hbond (T0197)D72 SheetConstraint (T0197)K108 (T0197)V112 (T0197)E126 (T0197)L120 hbond (T0197)R110 REMARK 44 model 8 is called T0197-1f1xA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m:1f1xA T0197.t2k.undertaker-align.sheets8 SheetConstraint (T0197)F15 (T0197)I17 (T0197)Y64 (T0197)F67 hbond (T0197)F15 SheetConstraint (T0197)R35 (T0197)E40 (T0197)G56 (T0197)K50 hbond (T0197)E37 SheetConstraint (T0197)L51 (T0197)R55 (T0197)T66 (T0197)K62 hbond (T0197)L52 SheetConstraint (T0197)E80 (T0197)F83 (T0197)V136 (T0197)V132 hbond (T0197)E80 SheetConstraint (T0197)F100 (T0197)Q103 (T0197)G127 (T0197)N123 hbond (T0197)K101 These constraints are terribly inconsistent: Many sheets include residue F83 for example, but it pairs with various partners: Y63, 123, 66, V132. I should probably build a strand constraint set for each of the major possible folds, and optimize separately for the folds. What a pain! ---------------------------------------------------------------------------- Thu Jun 24 16:23:10 PDT 2004 Sol Katzman I am starting to look at this target from the point of view of SCOP d.32.1.2 with representatives 1ewjA,1ecsA (essentially identical). The topology of these templates consists of two structural repeats that are 180deg relative to each other. The S1 and S5 strands are relatively close, but NOT H-bonded xxxxxxxxxxxxHHHHHxxxx x x xxxxxxxxxxxxxxx|xxxxs x x x x x x xxxxxx x x x xxxxxx xxxxxx H ^ S ^ ^ S S ^ S H H S2 S3 S4 S1 S5 S8 S7 S6 H H S S S S S S S S H H S S S S S S S S H H S S S S S S S S H H S S S S S S S S H HxxxxS VxxxxS S V V SxxxxV H x x x x N-term xxxxx|xxxxxxxxxxxxxxx x C-term By contrast, the topology of try13 contains the pair above with an extra anti-parallel strand (S9 below) followed by two additional alpha helices (not shown): xxxxxxxxxxxxxxxxxxxxx x x xxxxxxxxxxxxxxx|xxxxs x x x x x x xxxxxx x x x xxxxxx xxxxxx H ^ S ^ ^ ^ S S ^ S H H S2 S3 S4 S1 S9 S5 S8 S7 S6 H H S S S S S S S S S H H S S S S S S S S S H H S S S S S S S S S H H S S S S S S S S S H HxxxxS VxxxxS S S V V SxxxxV H x x x x x histag x xxxxx|xxxxxxxxxxxxxxx x x x N-term xxxxxxxxxxx Now for the details: -- Try13 has no obvious helix in the linker segment V57-N74, where 1ecs has a short helix L53-W59 -- Try13 does not have a very long strand S6 and it is thus not H-bonded along the length of strand S7 -- The His tag is not extended For tryA14: SheetConstraint to lengthen S6 and bond it to S7 HelixConstraint (weak) in the middle of the linker Increase weight on StrandConstraint for His tag PrintSheetConstraint ------------------------------------------------------------ Fri Jun 25 16:15:34 PDT 2004 Kevin Karplus I'll see if I can make progress on the b.60.1 structures, exemplified by 1gm6A. The b.60.1 structure is a barrel that is antiparallel: 1^ 2v 3^ 4v 5^ 6v 7^ 8v Strand 5 is short, and strand 9 comes after a helix and is antiparallel to 1, almost lined up with strand 2, which lines up with earlier part of strand 1. I don't see any way to get 8 antiparallel strands out of our target. There are 6 solid strands and 4 or 5 helices. Based on t2k.str2: S1: S9-K16 S2: Y63-F67 S3: E77-E84 S4: E101-L115 LONG! S5: V118-V122 S6: F131-I137 Based on t04.str2, I do have 8 strands: S1: E10-I17 S1A: V42-E45 S1B: L51-V557 S2: Y63-E79 S3: F78-E84 S4: I102-L115 S5: V118-V125 S6: F131-I137 The antiparallel relationship between S4 and S5 seems clear---a classic turn: SheetConstraint K108 L115 V125 V118 hbond L115 S6 is probably anti parallel to S5, perhaps SheetConstraint V118 E126 D130 S138 We probably also have S1A, S1B, S2, S3 as an antiparallel sheet: SheetConstraint V42 E45 G56 R53 SheetConstraint K62 K68 V57 L51 SheetConstraint Y63 E69 I85 Y79 If I assume that the two helices make parallel connections, I can make a barrel: S1^ S1A^ S1Bv S2^ S3v S4v S5^ S6v S1^ Something like SheetConstraint E10 F15 D41 V46 SheetConstraint E80 E84 Q103 K107 SheetConstraint F131 I137 E10 K16 I'll try these constraints as tryB14, and see what they lead to. I'll probably have to fuss with the strand alignment. Sat Jun 26 07:39:46 PDT 2004 Kevin Karplus tryB14 did not manage to produce a barrel--it was having trouble even forming sheets. I may have some phases wrong. Perhaps it would help if I started from something that almost got the barrel. The best-scoring models with the tryB14.costfcn are tryB14, try8, try11, try7, and try10. Of these, try11, try7, and try10 seem to be forming beta sheets. try7: SheetConstraint (T0197)Y63 (T0197)D72 (T0197)I85 (T0197)E76 hbond (T0197)Y64 SheetConstraint (T0197)W111 (T0197)L115 (T0197)V122 (T0197)V118 hbond (T0197)Y113 SheetConstraint (T0197)G117 (T0197)N123 (T0197)I137 (T0197)F131 hbond (T0197)T119 SheetConstraint (T0197)R124 (T0197)E126 (T0197)F131 (T0197)G129 hbond (T0197)R124 try10: SheetConstraint (T0197)Q39 (T0197)V42 (T0197)I137 (T0197)I134 hbond (T0197)E40 SheetConstraint (T0197)Y43 (T0197)V46 (T0197)I134 (T0197)F131 hbond (T0197)E45 SheetConstraint (T0197)L52 (T0197)G56 (T0197)E69 (T0197)L65 hbond (T0197)R53 SheetConstraint (T0197)I70 (T0197)E73 (T0197)Y79 (T0197)E76 hbond (T0197)L71 SheetConstraint (T0197)Y113 (T0197)L115 (T0197)L120 (T0197)V118 hbond (T0197)Y113 SheetConstraint (T0197)V118 (T0197)N123 (T0197)V136 (T0197)F131 hbond (T0197)T119 try11: SheetConstraint (T0197)Q39 (T0197)V42 (T0197)I137 (T0197)I134 hbond (T0197)E40 SheetConstraint (T0197)L52 (T0197)N59 (T0197)E69 (T0197)K62 hbond (T0197)L52 SheetConstraint (T0197)K62 (T0197)D72 (T0197)G86 (T0197)E76 hbond (T0197)Y64 SheetConstraint (T0197)K109 (T0197)W111 (T0197)V122 (T0197)R124 hbond (T0197)R110 SheetConstraint (T0197)Y113 (T0197)L115 (T0197)L120 (T0197)V118 hbond (T0197)Y113 SheetConstraint (T0197)G117 (T0197)V125 (T0197)I137 (T0197)G129 hbond (T0197)T119 Agreement? SheetConstraint (T0197)K62 (T0197)D72 (T0197)G86 (T0197)E76 hbond (T0197)Y64 SheetConstraint (T0197)W111 (T0197)L115 (T0197)V122 (T0197)V118 hbond (T0197)Y113 SheetConstraint (T0197)V118 (T0197)N123 (T0197)V136 (T0197)F131 hbond (T0197)T119 Let's take the agreement into tryB15.costfcn, along with other previously formed sheets that seem to be what we need. Sat Jun 26 15:30:14 PDT 2004 Kevin Karplus tryB15 almost forms the barrel I'm looking for. One problem with this barrel design: almost all the hydrophobics are one the *outside*. With only 4 helices, there aren't enough to cover the outside of the barrel. Maybe they cover half and the other half dimerizes? Perhaps the conserved polars and charged residues form salt bridges in the middle?? For this to work, we'd have to turn the "dry" weights down a lot. I'll try adding some plausible salt brdiges and turning down the dry weights for tryB16. I'll also move one strand over 2, to try to avoid a break and make the turn a bit more plausible. The tryB16.costfcn definition still favors tryB15 over the rest. Sat Jun 26 22:13:04 PDT 2004 Kevin Karplus tryB16 has almost formed the barrel that I'm trying to make. I should turn off wet6.5 and the other terms that try to drive the charged residues out of the center (way-back and near-backbone burial), and tweak the strand pairing to make it fit better. If the strand tweaking works to produce something more barrel-like on the next round, then I'll try grouping the charged residues in the middle better. Sun Jun 27 07:57:41 PDT 2004 Kevin Karplus tryB17 does not look very different from tryB16---the backbone has only moved a little bit. All that happened was to pack things a tiny bit tighter, resulting in better burial but more clashes. Maybe I should add some constraints to pack the helices against the sheets (L23, V42) (I148, L120) (L166, V122) (V94, F83), and strengthen the constraints for strand 1 on both sides. Then reoptimize without tryB16 or tryB17 in the mix, to get away from this local minimum. Sun Jun 27 16:55:29 PDT 2004 Kevin Karplus tryB18 does not look as good as tryB17 to the score function, but looks fairly barrel-like. It might be a good idea to add hbond constraints for R53 to D133 and E135---already there, just strengthen them Sun Jun 27 23:25:10 PDT 2004 Kevin Karplus tryB19 scores almost as well as tryB17, but the helix from K18 to Y36 is in completely the wrong place, with an enormous break after Y36. When I remove the constraints other than the strand pairing, tryB19 does worse than tryB15, tryB16, and tryB17. Perhaps I should reoptimize it with just the strand-pairing constraints, with perhaps one constraint on the misplaced helix. (L23 near F15, L26 near F44) I'll also increase the weight of hbond_geom, so that salt bridges (and other sidechain hbonds) are seen as important. Mon Jun 28 07:48:43 PDT 2004 Kevin Karplus tryB20 scores almost as well as tryB16, but is still a long way from tryB17. The segment L19 to Y36 was not moved, despite the high probability given to OptSegment. Mon Jun 28 16:15:52 PDT 2004 Kevin Karplus The d.31.1 prediction is looking more likely to me It is a half barrel, though in some proteins there is domain-swapping that makes it looks like two unrelated 1/4 barrels. The barrels DO have the active polar residues on the inside and the hydrophobics on the outside (making a dimerization interface). I could convert the barrel costfcns into half-barrel ones simple by omitting one sheet constraint. The question is, which one? The d.31.1 domains do have a conserved repeat structure, so the sheet constraint to omit would be one that is at the edge of the repeats. Mon Jun 28 17:16:14 PDT 2004 Kevin Karplus Despite bug fixes to the OptSegment operator and the new OptAllSegments operator, I'm still not getting the segment L19-Y36 to move. I may want to make further changes to those operators. ------------------------------------------------------------------------------------------- Mon Jun 28 17:35:05 PDT 2004 Sol Katzman For SCOP d.32.1.3 we have 3 templates, each with two instances of the domain: 1cjxA 1-153, 154-356 1f1xA 1-147, 148-322 1q0cA 1-147, 148-322 These are similar to the d.32.1.2 topology, but close the two sheets. 1f1xA(1-147) and 1q0cA(1-147) are like this, with the crossover strand S5 H-bonded to S1. This just closes up the d.32.1.2 topology: xxxxxxxxxxx x x xxxxxxxxxxxxxxx|xxxxs x x x x x x xxxxxx x x x xxxxxx xxxxxx H ^ S ^ ^ S S ^ S H H S2 S3 S4 S1 S5 S8 S7 S6 H H S S S S S S S S H H S S S S S S S S H H S S S S S S S S H H S S S S S S S S H HxxxxS VxxxxS S V V SxxxxV H x x x x x xxxxx|xxxxxxxxxxxxxxx x x N-term C-term 1f1xA(148-322) and 1q0cA(148-322) have an additional strand S9 that pairs with S2: C-term x xxxxxxxxxxx x x x xxxxxxxxxxxxxxxxxxxxx|xxxxs x x x x x x x x xxxxxx x x x xxxxxx xxxxxx H ^ ^ S ^ ^ S S ^ S H H S9 S2 S3 S4 S1 S5 S8 S7 S6 H H S S S S S S S S S H H S S S S S S S S S H H S S S S S S S S S H H S S S S S S S S S H H S S VxxxxS S V V SxxxxV H Hxxxx|xxxxxx x x x x x x xxxxx|xxxxxxxxxxxxxxx x x x xxxxxxxxxxxxxxxxxxxxx|xxxxxxxxxx x x N-term 1cjxA(1-153) is a little different, having a definite helix between S1 and S5, and also having residues that do not form strand S6: xxxHHHHHHxx x x xxxxxxxxxxxxxxx|xxxxs x x x x x x xxxxxx x x x xxxxxx xxxxxx H ^ S ^ ^ S S ^ x H H S2 S3 S4 S1 S5 S8 S7 x H H S S S S S S S x H H S S S S S S S x H H S S S S S S S x H H S S S S S S S x H HxxxxS VxxxxS S V V Sxxxxx H x x x x x xxxxx|xxxxxxxxxxxxxxx x x N-term C-term 1cjxA(154-356) reinstates the strand S6, but only after some extra gyrations and helices: xxxHHHHHHxx x x xxxxxxxxxxxxxxx|xxxxs x x x x x x xxxxxx x x x xxxxxx xxxxxxxxHHHHHx H ^ S ^ ^ S S ^ S x H S2 S3 S4 S1 S5 S8 S7 S6 HxxxH x H S S S S S S S S H H x H S S S S S S S S H H x H S S S S S S S S H H x H S S S S S S S S H Hxxxx HxxxxS VxxxxS S V V Sxxxxx H x x x x x xxxxx|xxxxxxxxxxxxxxx x x N-term C-term As Kevin pointed out, the t04.str2 prediction contains 8 strands. It also seems to have helices as needed to match the topology of 1f1xA/1q0cA(1-147) (better than try13 does): S1: E10-I17 H : L19-T28 S2: V42-E45 S3: L51-V57 S4: Y63-E69 S5: F78-E84 H : F88-R97 S6: I102-L115 S7: V118-V125 S8: F131-I137 H : P141-M154 H : P164-E173 For the S2-S3-S4 anti parallel sheet: SheetConstraint E40 E45 G56 L51 SheetConstraint L51 V57 E69 Y63 For the S6-S7-S8 anti parallel sheet: SheetConstraint K108 L115 V125 V118 SheetConstraint T119 V125 I137 F131 For the S1-S4 parallel sheet: SheetConstraint V11 I17 Y63 E69 For the S5-S8 parallel sheet: SheetConstraint F78 E84 F131 I137 For the S1-S5 anti parallel sheet SheetConstraint V11 I17 F78 E84 For the helices: HelixConstraint L19 T28 HelixConstraint F88 R97 The best scoring models to these constraints are: try7,9,13,8. For now, just start with the constraints as tryA15. ------------------------------------------------------------------------------------------- Tue Jun 29 11:30:35 PDT 2004 Sol Katzman tryA14 did not show much change from try13. The his tag did get extended, but the strand S6 did not get lengthened . tryA15 got hit by a power outage so we only have tryA15-opt1,tryA15-opt1-scwrl to look at. About the only things that worked were the anti parallel sheets S2-S3 and S6-S7. Look at the sheets from the previous tries that scored best against the SCOP d.32.1.3 constraints, namely try7,try9,try13,try8 try7: SheetConstraint (T0197)Y63 (T0197)D72 (T0197)I85 (T0197)E76 hbond (T0197)Y64 SheetConstraint (T0197)W111 (T0197)L115 (T0197)V122 (T0197)V118 hbond (T0197)Y113 SheetConstraint (T0197)G117 (T0197)N123 (T0197)I137 (T0197)F131 hbond (T0197)T119 SheetConstraint (T0197)R124 (T0197)E126 (T0197)F131 (T0197)G129 hbond (T0197)R124 try9: SheetConstraint (T0197)K68 (T0197)E73 (T0197)V81 (T0197)E76 hbond (T0197)L71 SheetConstraint (T0197)K109 (T0197)V112 (T0197)N123 (T0197)L120 hbond (T0197)R110 SheetConstraint (T0197)Y113 (T0197)L115 (T0197)L120 (T0197)V118 hbond (T0197)Y113 SheetConstraint (T0197)V118 (T0197)R124 (T0197)V136 (T0197)D130 hbond (T0197)T119 try13: SheetConstraint (T0197)G8 (T0197)E12 (T0197)L51 (T0197)R55 hbond (T0197)S9 SheetConstraint (T0197)P31 (T0197)R35 (T0197)V46 (T0197)V42 hbond (T0197)E32 SheetConstraint (T0197)D41 (T0197)E45 (T0197)R55 (T0197)L51 hbond (T0197)V42 SheetConstraint (T0197)Y79 (T0197)F83 (T0197)I137 (T0197)D133 hbond (T0197)E80 SheetConstraint (T0197)E80 (T0197)G86 (T0197)E121 (T0197)G127 hbond (T0197)V81 SheetConstraint (T0197)E82 (T0197)E84 (T0197)D133 (T0197)F131 hbond (T0197)E84 SheetConstraint (T0197)K101 (T0197)Q103 (T0197)K114 (T0197)V112 hbond (T0197)K101 SheetConstraint (T0197)R110 (T0197)K114 (T0197)V125 (T0197)E121 hbond (T0197)W111 try8: SheetConstraint (T0197)Q39 (T0197)D41 (T0197)I137 (T0197)E135 hbond (T0197)E40 SheetConstraint (T0197)I70 (T0197)L71 (T0197)N75 (T0197)N74 hbond (T0197)L71 SheetConstraint (T0197)G117 (T0197)E121 (T0197)I137 (T0197)D133 hbond (T0197)T119 SheetConstraint (T0197)D133 (T0197)I134 (T0197)F44 (T0197)Y43 hbond (T0197)I134 // Attempt to get SCOP d.32.1.2 topology, specifically 1f1xA(1-147) and 1q0cA(1-147) SheetConstraint E40 E45 G56 L51 SheetConstraint L51 V57 E69 Y63 SheetConstraint K108 L115 V125 V118 SheetConstraint T119 V125 I137 F131 SheetConstraint V11 I17 Y63 E69 SheetConstraint F78 E84 F131 I137 SheetConstraint V11 I17 F78 E84 HelixConstraint L19 T28 HelixConstraint F88 R97 Consistent with SCOP proposed constraints: 13: SheetConstraint (T0197)D41 (T0197)E45 (T0197)R55 (T0197)L51 hbond (T0197)V42 7: SheetConstraint (T0197)W111 (T0197)L115 (T0197)V122 (T0197)V118 hbond (T0197)Y113 9: SheetConstraint (T0197)K109 (T0197)V112 (T0197)N123 (T0197)L120 hbond (T0197)R110 9: SheetConstraint (T0197)Y113 (T0197)L115 (T0197)L120 (T0197)V118 hbond (T0197)Y113 13: SheetConstraint (T0197)R110 (T0197)K114 (T0197)V125 (T0197)E121 hbond (T0197)W111 13: SheetConstraint (T0197)V118 (T0197)R124 (T0197)V136 (T0197)D130 hbond (T0197)T119 7: SheetConstraint (T0197)G117 (T0197)N123 (T0197)I137 (T0197)F131 hbond (T0197)T119 7: SheetConstraint (T0197)R124 (T0197)E126 (T0197)F131 (T0197)G129 hbond (T0197)R124 9: SheetConstraint (T0197)V118 (T0197)R124 (T0197)V136 (T0197)D130 hbond (T0197)T119 13: SheetConstraint (T0197)Y79 (T0197)F83 (T0197)I137 (T0197)D133 hbond (T0197)E80 So for tryA16, read in try7,try9,try13,try8, and use some of the try13 sheetconstraints as well as the putative scop constraints: SheetConstraint (T0197)D41 (T0197)E45 (T0197)R55 (T0197)L51 hbond (T0197)V42 4.0 SheetConstraint L51 V57 E69 Y63 2.0 SheetConstraint (T0197)R110 (T0197)K114 (T0197)V125 (T0197)E121 hbond (T0197)W111 4.0 SheetConstraint (T0197)V118 (T0197)R124 (T0197)V136 (T0197)D130 hbond (T0197)T119 4.0 SheetConstraint V11 I17 Y63 E69 2.0 SheetConstraint (T0197)Y79 (T0197)F83 (T0197)I137 (T0197)D133 hbond (T0197)E80 4.0 SheetConstraint V11 I17 F78 E84 2.0 ------------------------------------------------------------------------------------------- Tue Jun 29 15:48:56 PDT 2004 Kevin Karplus Even my best attempt at making a barrel is having trouble closing it. I'm going to give up on the barrel for now and release the N-terminal strand from the C-terminal one. This will be the tryC series, and may look more like the d.63.1.1 family. I'll start with the constraints I'm using on the tryB series (without the barrel closing), but start over from the alignments. Tue Jun 29 21:19:47 PDT 2004 Kevin Karplus BZZT. The d.63.1.1 family IS a closed barrel, but the strand ordering is a bit wierd: There are 10 strands, in an 8-strand barrel: S1 S5 S4 S3(short) S2 S6 S7 S8 S1a antiparallel to S4 on S3 side S1b antiparallel to end of S1, on S8 side ALL strand-strand interactions are antiparallel. We don't seem to have enough strand residues for all this, so need to omit S1a and S1b. I'll try something vaguely like this as tryC2. Wed Jun 30 18:24:27 PDT 2004 Kevin Karplus The tryC1 and tryC2 jobs that were running on abyss both died at about 1:34 or 1:34 this afternoon, and I can't login to abyss now, so I suspect that it has been powered down or removed from the network. Oh well, they were wasting time anyway, since I had a bug in the tryC1.costfcn file (I'd constrained strand 1 inconsistently, having a remnant of a sheet constraint I no longer believed in. Thu Jul 1 02:44:22 PDT 2004 Kevin Karplus Foo! SCWRL got wedged again, during the "scwrl_each" call on the first iteration. I've got to find a way of automatically killing the scwrl process when it takes too long! This will probably mean having to fork off the scwrl process, rather than just doing a simple system() call. Thu Jul 1 04:58:06 PDT 2004 Kevin Karplus tryC3-opt1 is beginning to get the right barrel form, but the overtightness of the constraints on the clustering atoms is making it hard for the barrel to form properly. The atoms shouldn't be making saltbridges, but clustering around a sulphate ion. For tryC4, I'll make the looser constraints. I'll also try shifting the S3-S4 constraints by 2, since they seem to want to line up that way. I should probably start a tryC4 run with looser constraints there. Even better might be to try to create a hand alignment to 1d8iA that lines up the strands and conserved residues, and make the constraints consistent with that. I'm too tired to hand-edit alignments tonight, so I'll just start tryC4 from the tryC3-opt1 model. Thu Jul 1 16:21:05 PDT 2004 Kevin Karplus Although tryC4 is still not done, I'm starting a tryC5 run, trying to get the hbonding pattern for the sheet that tryC4 seems to be trying to to get. I'm pretty tired this afternoon though, so I may very well have slipped up on the sheet constraints. I'll run this as a shorter run on a faster machine, so there won't be as much delay before I can look at it. Right now I think that our first model should be for d.63.1.1, and our second model for d.32.1.3. After that we can submit some of the alignments. Perhaps the t2k alignment for 1gm6A, and t04 alignments for 1d8hA and 1fa8A? Thu Jul 1 17:28:05 PDT 2004 Kevin Karplus I submitted tryC5-opt1 to VAST at http://www.ncbi.nlm.nih.gov/Structure/vast/vastsearch.cgi/submit with password T0197tryC5 getting search id VS59399 status can be checked at http://www.ncbi.nlm.nih.gov/Structure/vast/vastresults.cgi?result=Here Results can be viewed with Cn3D: /projects/compbio/bin/i686/Cn3D There doesn't seem to be much here---we may need a better structure before VAST can align it? In tryC5-opt1, the strand containing R124 and E126 is flipped so that the charged residues don't point in to the ion, but E121 does correctly point in. Fixing this requires a bulge, with both N123 and R124 pointing in. We want to hbond T119-E135, E121-D133, but I get confused about the hbonding for F131 (maybe to N123.N and R124.O?). SheetConstraint V118 V122 I137 V132 hbond T119 Hbond N123.N F131.O Hbond R124.O F131.N The barrel is also having a hard time closing. Maybe it needs Hbonds E10-E82, E12-Y79+E80, K14-E77? hard to say---there are so many charged residues and I don't want to bulge everything. SheetConstraint S9 V11 F83 V81 hbond E10 Hbond E12.N E80.O Hbond E12.O Y79.N From www-data@came.sbg.ac.at Thu Jul 1 18:50:08 2004 To: karplus@soe.ucsc.edu Subject: ProSup: job finished (T0197.tryC5-opt1.pdb vs. 1d8i) Date: Fri, 2 Jul 2004 03:50:01 +0200 (CEST) From: www-data@came.sbg.ac.at (www-data) 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 Your job has been finished and is available at http://lore.came.sbg.ac.at:8080/CAME/CAME_EXTERN/PROSUP/TMP/108873298885565751 . This message has been created automatically by http://lore.came.sbg.ac.at:8080/CAME/CAME_EXTERN/PROSUP/ . If your receive this message without having asked for it, please contact prosup-admin@came.sbg.ac.at . Regards, The Prosup Team ------------------------------------------------------------ Thu Jul 1 21:12:35 PDT 2004 Kevin Karplus Since I did not get an alignment from VAST, I submitted the model to prosup, telling it to align to 1d8iA. Prosup returned 5 alignments (saved in 1d8iA/1d8iA-tryC5-opt1.prosup.txt). I edited the first (and longest) of these into an a2m file (1d8iA/1d8iA-T0197-prosup1.a2m) and made an alignment-based model from it with show-align-prosup.under. I looked at the conserved residues and the charged residues of this alignment, and decided that two of the strands needed to be moved over 1 to flip them over. I did that editing in 1d8iA/1d8iA-T0197-hand1.a2m, and added that alignment to show-align-prosup.under. I also put the resulting conformation into superimpose-best.under, to get out the sheet constraints and because it is probably the best partial conformation we have so far. I then put the sheet constraints from the hand1 model into tryC6.costfcn, plus some distance constraints copied from that model (measured with rasmol). Thu Jul 1 21:49:10 PDT 2004 Kevin Karplus I just realized that the initial strand constraint to make the HIS tag stick out is flawed---the constraint should not extend past H7. For the next run I should fix that constraint, and also replace all the "included" constraints with a consensus constraint on where the strands and helices are. I should probably also strengthen the SheetConstraints--I left them at the default weight of 1.0, which may be too weak. Looking at tryC6-opt1: Further problems---I128 and G127 have to make a turn--they can't participate in the strand pairing with T105 and I106. I'll peel back the sheet constraints a residue or two where the strand are having a hard time joining, and increase the break costs. --------------------------------------------------------------------- Thu Jul 1 22:09:40 PDT 2004 Sol Katzman The result of tryA16 did not show any strands despite the sheet constraints. For tryA17: -- correct an error where I had a sheet constraint for S1-S5 being parallel (should be antiparallel) -- use an output sheet constraint from try10 for S3-S4 -- use an output sheet constraint from tryA16 for S1-S4 The results of tryA17-opt1 still show no strands, even though all but one of the constraints came from actual models. Also, S1-S5 still look parallel. For tryA18, I will make the S1-S5 constraint have higher weight. Also, I will fix the HIS-tag constraint as per Kevin's comment. And I will NOT read in tryA16 which has the backwards S1-S5 arrangement. For tryA19, same costfcn as tryA18, but I will not read in any of the old try conformations. ----------------------------------------------------------------------- Fri Jul 2 08:12:35 PDT 2004 Sol Katzman TryA18 has a few strands, and at least all the putative sheets have the proper orientation of parallel or antiparallel. Darn. Due to faulty query/replace in tryA19.under, the files tryA18-opt*.pdb are actually the tryA19 outputs. On the other hand, it looks like tryA18 died without producing any output (in generation 41 of the first iteration of opt1) so nothing was actually overwritten by tryA19 but we have no true tryA18 output. There is insufficient time before the casp deadline today, but for the record, I will create tryA20 as a duplicate of the original tryA18. -------------------------------------------------- Fri Jul 2 12:10:08 PDT 2004 Kevin Karplus I submitted one of each type of model: 1 d.63.1 T0197.tryC9-opt1.pdb 2 d.23.1 T0197.tryA14-opt2.pdb 3 b.60.1 T0197.tryB15-opt2.pdb 4 b.60.1 from T0197-1gm6A-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m 5 d.63.1 from 1d8iA-T0197-hand1.a2m When superimposed, these different predictions from different templates were remarkably similar. Thu Nov 18 22:02:48 PST 2004 Martina Koeva Based on the evaluations: best sam-t04 32.4447 best submit 31.1386 (model 5) model1 31.0487 auto 13.5561 align 19.5129 (align1 scored a bit worse than align2, which was at 21.7008) robetta best 14.3871 (robetta model2) robetta1 14.2708 Fri Nov 19 13:13:55 PST 2004 Kevin Karplus I reran the evaluation and got #targ best bestsub model1 auto align bestrob robetta1 T0197 32.4447 31.1529 30.3452 13.5561 19.5129 14.3871 14.2708 The best model was T0197.tryC5-opt1.pdb (same backbone and lower all-atom rmsd than the scwrled version). This is a barrel from the d.63.1 family, our second-best hit in the t2k library. Our model1 and model5 were from this family. There are some minor misalignments, and the helices are twisted a bit from their correct positions, but overall, I think we did quite well on this one.