Thu Jul 10 09:22:55 PDT 2008 T0495 Make started Thu Jul 10 09:23:58 PDT 2008 Running on peep.cse.ucsc.edu Thu Jul 10 12:42:05 PDT 2008 Kevin Karplus This looks like fold recognition for superfamily c.52.1.* Thu Jul 10 13:50:55 PDT 2008 Kevin Karplus The fold recognition looks good, but there are a lot of breaks to close. I'll try 2 approaches: polishing try1 (as try2) and using the same costfcn to make a model from alignments. Thu Jul 10 20:58:12 PDT 2008 Kevin Karplus Starting from alignments (try3-opt3.gromacs0) scored better than polishing existing (try2-opt3.gromacs0). I should try again from alignments, with a costfcn based on try3. Fri Jul 11 00:05:44 PDT 2008 Kevin Karplus try4-opt3.gromacs0 doesn't score even as well as try2. The main decision/work to be done on this model is figuring what happens with K57-G97. There is a helix in that loop, but what else is going on in unclear. Mon Jul 14 09:20:53 PDT 2008 Kevin Karplus The MQAC quality assessment likes the SAM-T08-server models, the Zhang-Server models, RAPTOR, Pcons_dot_net, ... (that is, the usual suspects). Predicted GDT is 45-47%. The MQAU quality assessment likes Zhang-Server, SAM-T08-server, and RAPTOR also, but predicts 60-66% GDT. I should probably create an MQAY1 run that excludes the SAM models. Mon Jul 14 09:26:47 PDT 2008 Kevin Karplus I started both an MQAY1 run and an MQAY4 run (using the try4 costfcn). Wed Jul 16 14:31:20 PDT 2008 Kevin Karplus costfcn favorite models try1 MQAY1-opt3 try2-opt3 try1-opt3 MQAU1-opt3 try2-opt3.gromacs0 MQAC1-opt3 MQAY4-opt3 try2 MQAY4-opt3 try3-opt3.gromacs0 MQAY1-opt3.gromacs0 try2-opt3.gromacs0 try4-opt3.gromacs0 try1-opt3.gromacs0 try3 MQAY4-opt3 try3-opt3.gromacs0 MQAY1-opt3.gromacs0 try2-opt3.gromacs0 try4-opt3.gromacs0 try1-opt3.gromacs0 try4 try3-opt3.gromacs0 MQAY4-opt3 MQAY1-opt3.gromacs0.repack-nonPC try2-opt3 try4-opt3.gromacs0 try1-opt3.gromacs0 rosetta MQAY4 MQAY1 try3 try2 try4 try1 MQAC1 Wed Jul 16 14:49:42 PDT 2008 Kevin Karplus For F26-L45, MQAY4-opt3 has a much better hairpin than the other models. For later parts of the sheet, models try3-opt3, try2-opt3, try4-opt3, and try1-opt3 are better. Let me make a chimera-try3-MQAY4 that takes F26-L45 from MQAY4-opt3.gromacs0.repack-nonPC and the rest from try3-opt3.gromacs0 One possible problem---the phase of the strands is different, and MQAY4 has an unusual separation=4 Hbond for the hairpin. Wed Jul 16 15:03:59 PDT 2008 Kevin Karplus try5 will attempt to polish chimera-try3-MQAY4 Wed Jul 16 19:15:25 PDT 2008 Kevin Karplus MQAY4-opt3 still scores better than try5-opt3 with the try5 costfcn, but not by a lot. Despite that, I prefer try5-opt3, and will try polishing it further. Wed Jul 16 22:56:37 PDT 2008 Kevin Karplus try6 does not look bad, and has most of the structure of the c.52.1.18 superfamily (1gefA and 1ob8A). Dimerizing it will help hide the hydrophobic patch. There is a pair of antiparallel strands that we are not currently getting, after the third strand of the sheet and before the last strand of the sheet. 1gefA doesn't have the 6th strand of the sheet, but does have the pair of strands in back. If this is correct, then we need about 5 residues antiparallel: quite likely areound R78-A82 and A118-R121. 2eo0 (the best-scoring match) has 3 strands in the minor sheet: one right after the 3rd strand of the main sheet, then a hairpin after the 5th strand. Maybe I should give try6 to VAST, to see what it is closest to, and to try to make up my mind whether I should strive for creating the 2-3-strand minor sheet. Wed Jul 16 23:26:54 PDT 2008 Kevin Karplus Your VAST Search job was submitted at 07/17/2008 02:26:34(EDT). Request ID: 1130271914809858643 The 1y88A hit (c.52.1.30, instead of c.52.1.18) has a pair of short parallel strands after strands 3 and 4, so the antiparallel pair in a minor sheet is not a crucial feature of the c.52.1 superfamily. The 1y88 dimer does not look like it will help with burying the currently exposed hydrophobics, though. I'll do a dimer optimization with the try6 costfcn as dimer/try7 I won't add dimer constraints, but I'll up the pred-nb11-back weight. Wed Jul 16 23:37:41 PDT 2008 Kevin Karplus VAST thinks try6-opt3 is most similar to 1ob8B=1ob8A 1ddkA 1hh1A 2eo0A 1qw2A 2fhxA 1xm8A Thu Jul 17 07:25:03 PDT 2008 Kevin Karplus dimer/try7-opt3 does an adequate job of burying the hydrophobics, but not great. I'm still concerned about S58-L92, which does not seem to have the predicted secondary structure, nor get properly buried. Perhaps I should look at the sheets of some of the MQAC preferred models. For example, Zhang-Server_TS2 and Zhang-Server_TS3 put R78-V83 into the main sheet as strand 4, which looks feasible. circle_TS3 follows the alignment to 1gefA pretty closely. Perhaps I should try extracting sheet constraints from those 3 models and use them as costfcns for a meta-server run? (Note: one of the templates I used, 1hh1A, has *2* hairpins in the minor sheet) Thu Jul 17 07:57:20 PDT 2008 Kevin Karplus The circle1.sheets file does not have as many sheet constraints as I expected. In particular, there is no antiparallel constraint for F63-D66 with M111-I108. Rasmol reports hbonds M111.N-F63.O, and both for A65-V109. Undertaker sees M111.N-F63.O and A65.N-V109.O, but not A65.O-V109.N, probably because the backbone twists enough there to make the plane of the C-N-CA point too far from the O. I'll have to do some hand-extraction of sheet constraints if I want to use the circle_TS1 or Zhang-Server_TS2 constraints. Thu Jul 17 09:25:19 PDT 2008 Kevin Karplus I fixed undertaker to do a better job of extracting short sheets. Now circle1 reports SheetConstraint F80 V83 L99 S96 hbond V83 1 SheetConstraint V53 K57 Y79 V83 hbond K57 1 SheetConstraint D41 G46 E55 E50 hbond V42 1 SheetConstraint F26 E30 G46 V42 hbond V27 1 zhang2 reports SheetConstraint F80 V83 I100 G97 hbond A82 1 SheetConstraint I51 C56 L45 V40 hbond C56 1 SheetConstraint F26 S28 G46 G44 hbond V27 1 SheetConstraint E30 V33 V42 R39 hbond E31 1 zhang3 reports SheetConstraint F80 A82 I54 C56 hbond A82 1 SheetConstraint D41 L45 E55 I51 hbond V42 1 SheetConstraint F26 F32 G46 V40 hbond V27 1 I'm still not seeing the sheet SheetConstraint F63 D66 M111 I108 hbond M111 in circle1. Thu Jul 17 10:20:51 PDT 2008 Kevin Karplus Another little bug fix and I get circle1.sheets to be SheetConstraint (T0495)I108 (T0495)M111 (T0495)D66 (T0495)F63 hbond (T0495)M111 1 SheetConstraint (T0495)V53 (T0495)K57 (T0495)Y79 (T0495)V83 hbond (T0495)K57 1 SheetConstraint (T0495)D41 (T0495)G46 (T0495)E55 (T0495)E50 hbond (T0495)V42 1 SheetConstraint (T0495)F26 (T0495)E30 (T0495)G46 (T0495)V42 hbond (T0495)V27 1 What happened to SheetConstraint F80 V83 L99 S96 hbond V83 1 ? I have V83.N-S96.O, W91.N-L98.O, but not the Hbonds in the other direction. Perhaps I need to be sloppier yet in accepting missing Hbonds, looking for a pair of Hbonds with possible one skipped in-between. Thu Jul 17 10:39:11 PDT 2008 Kevin Karplus Yet a further loosening of the Hbond pair test in FindStrandPair now gives me circle1.sheets: SheetConstraint (T0495)I108 (T0495)M111 (T0495)D66 (T0495)F63 hbond (T0495)M111 1 SheetConstraint (T0495)F80 (T0495)V83 (T0495)L99 (T0495)S96 hbond (T0495)W81 1 SheetConstraint (T0495)V53 (T0495)K57 (T0495)Y79 (T0495)V83 hbond (T0495)K57 1 SheetConstraint (T0495)D41 (T0495)G46 (T0495)E55 (T0495)E50 hbond (T0495)V42 1 SheetConstraint (T0495)F26 (T0495)E30 (T0495)G46 (T0495)V42 hbond (T0495)V27 1 zhang2.sheets: SheetConstraint (T0495)F80 (T0495)V83 (T0495)I100 (T0495)G97 hbond (T0495)F80 1 SheetConstraint (T0495)I51 (T0495)C56 (T0495)L45 (T0495)V40 hbond (T0495)C56 1 SheetConstraint (T0495)F26 (T0495)S28 (T0495)G46 (T0495)G44 hbond (T0495)V27 1 SheetConstraint (T0495)E30 (T0495)V33 (T0495)V42 (T0495)R39 hbond (T0495)E31 1 zhang3.sheets: SheetConstraint (T0495)F80 (T0495)A82 (T0495)I54 (T0495)C56 hbond (T0495)A82 1 SheetConstraint (T0495)D41 (T0495)L45 (T0495)E55 (T0495)I51 hbond (T0495)V42 1 SheetConstraint (T0495)F26 (T0495)F32 (T0495)G46 (T0495)V40 hbond (T0495)V27 1 try7-opt3.sheets: SheetConstraint (T0495)I108 (T0495)R110 (T0495)R121 (T0495)P119 hbond (T0495)V109 1 SheetConstraint (T0495)E107 (T0495)A112 (T0495)D102 (T0495)G97 hbond (T0495)A112 1 SheetConstraint (T0495)I51 (T0495)K57 (T0495)E95 (T0495)A101 hbond (T0495)K57 1 SheetConstraint (T0495)V42 (T0495)L45 (T0495)I54 (T0495)I51 hbond (T0495)V42 1 SheetConstraint (T0495)F26 (T0495)V33 (T0495)G44 (T0495)G37 hbond (T0495)V27 1 I'll put together a set of sheet constraints to try to get the good parts of the server models: Based on circle1 with mini sheet: SheetConstraint F26 F32 G46 V40 hbond V27 1 SheetConstraint V40 G46 C56 E50 hbond V42 SheetConstraint V53 K57 Y79 V83 hbond K57 1 SheetConstraint F80 V83 L99 S96 hbond W81 1 SheetConstraint I108 M111 D66 F63 hbond M111 1 This will be try8 based on zhang2, plus try7 to make 7-strand sheet (unlikely) SheetConstraint F26 F32 G46 V40 hbond V27 1 SheetConstraint V40 G46 C56 E50 hbond V42 SheetConstraint V53 K57 Y79 V83 hbond K57 1 SheetConstraint F80 V83 I100 G97 hbond F80 1 SheetConstraint E107 A112 D102 G97 hbond A112 1 SheetConstraint I108 R110 R121 P119 hbond V109 1 This will be try9 I'll do both runs from alignments and metaserver runs for try8 and try9. Thu Jul 17 11:13:03 PDT 2008 Kevin Karplus try8, try9, MQAX8, MQAX9 started on moai cluster. Thu Jul 17 13:52:28 PDT 2008 Kevin Karplus I had typos in the MQAX8 and MQAX9 runs, so I'm having to rerun them. Thu Jul 17 14:06:49 PDT 2008 Kevin Karplus Both try8 and try9 have similar structures, only really differing in the placement of the final helix. Both have bad breaks that would need to be closed, and neither forms a minor sheet. MQAX9 is going to polish BAKER-ROBETTA_TS5 MQAX8 is going to polish SAM-T08-server_TS1 Thu Jul 17 16:57:24 PDT 2008 SAM-T08-MQAO hand QA T0495 Submitted Thu Jul 17 16:57:24 PDT 2008 SAM-T08-MQAU hand QA T0495 Submitted Thu Jul 17 16:57:24 PDT 2008 SAM-T08-MQAC hand QA T0495 Submitted Thu Jul 17 19:48:15 PDT 2008 Kevin Karplus Both try8 and try9 like MQAX9-opt3 best, and MQAX8-opt3 next. Both are based on BAKER-ROBETTA models, and both have an undesirable break in the sheet, between the 4th and 5th strands. I rather like try8-opt3, based on 1okfA, but it has some bad breaks. I'll try closing them in try10. After that, I should try dimerizing try10 based on 2okfAB Fri Jul 18 10:48:23 PDT 2008 Kevin Karplus try10 closed the easy gaps, but still left bad breaks from D77 to W81. I'll do the dimer optimization with dimer-try10-2okf as dimer/try11, but I don't think that the dimer will do what I want, as the C-terminal helices are probably misplaced. This may even be a helix-swap dimer, though just flipping the helix around to be antiparallel to the D66-W75 helix would make for better packing. Unfortunately, there is no C-terminal helix in 2ofkA, so figuring out the packing may be tough. Fri Jul 18 11:11:27 PDT 2008 Kevin Karplus I don't know what to do about the terminal helix, but I can try to improve the gap in try10 by increasing break costs. I'll try this as try12. Fri Jul 18 12:37:28 PDT 2008 Kevin Karplus try12 looks fairly reasonable, but I think I want to also try swinging the C-terminal helix around so that Y79 is near L125 and W81 is near F129 Fri Jul 18 12:50:23 PDT 2008 Kevin Karplus I'm trying to move the helix in try13. This looks like it would be another good application for FoldIt. Fri Jul 18 13:14:54 PDT 2008 Kevin Karplus try13 seems to be optimizing from try10-opt1, rather than from try12. That means I'll probably have to close breaks again when try13 is finished. Fri Jul 18 19:30:54 PDT 2008 Kevin Karplus But try13 only moved the helix a little way, not swinging it around as I had hoped, so there isn't much reason to keep try13 around. Sun Jul 27 18:43:24 PDT 2008 Kevin Karplus I set up MANUAL_TOP_HITS in Makefile with the top 13 hits from best-scores and with the top 7 VAST hits for try6. I'll use these in a run from alignments (with what cost fcn?) and then start trying to whittle down the excessive number of models I have to consider. Sun Jul 27 19:07:37 PDT 2008 Kevin Karplus For try15, I think I'll go back to a costfcn with little or no specific sheet constraints (maybe align_sheets), and with the MANUAL_TOP_HITS templates as the only alignments. Sun Jul 27 20:33:12 PDT 2008 Kevin Karplus OOPS, I forgot to take the read-pdb.under out of try15, so it is just polishing MQAX9-opt3, not generating a new start. I'll make try16 the same as try15, but without that mistake. Sun Jul 27 23:02:16 PDT 2008 Kevin Karplus try16 seems to be based mainly on 2okfA, but not quite as successfully as try12 is. I like the try12 model---it makes a fairly pretty sheet. Mon Jul 28 12:23:21 PDT 2008 Kevin Karplus Top hits in best-scores, and what they look like: 2eo0A 5-strand sheet followed by 3-strand sheet: 1 ^v 2 ^v 3 || 5 ^v 6 4 ^v 8 ^v 7 1gefA 5-strand followed by 2-strand sheet 1 ^v 2 ^v 3 || 5 ^v 6 4 ^v 7 1ob8A 6-strand followed by 2-strand sheet 1 ^v 2 ^v 3 || 5 ^v 6 ^v 8 4 ^v 7 1y88A 5-strand followed by 2-strand sheet 1 ^v 2 ^v 3 || 5 || 7 4 || 6 1hh1A 5-strand sheet followed by 3-strand sheet: 1 ^v 2 ^v 3 || 6 ^v 7 4 ^v 5 ^v 8 2inbA 6-strand sheet 1 ^v 2 ^v 3 || 4 || 5 ^v 6 1xmxA not clear which domain I should look at. 2okfA 6-strand sheet 1 ^v 2 ^v 3 || 4 || 5 ^v 6 both 2okfA and 2inbA have a hairpin and missing density between strands 1 and 2. I've been treating it as if it were a single long hairpin, which is unlikely given the missing density. 2pjuA 2 4-strand sheets 2 || 1 || 3 || 4 6 || 5 || 7 || 8 1yfbA sandwich/barrel formed by dimer Mon Jul 28 12:45:55 PDT 2008 Kevin Karplus What I currently have model 1 = T0495.try12-opt3.pdb 1 ^v 2 ^v 3 || 4 || 5 ^v 6 model 2 = T0495.MQAY1-opt3.gromacs0.repack-nonPC.pdb 1 ^v 2 ^v 3 || 4 5 ^v 6 ^v 7 model 3 = T0495.try15-opt3.pdb 1 ^v 2 ^v 3 || 4 5 ^v 6 ^v 7 model 4 = T0495.MQAX8-opt3.pdb 1 ^v 2 ^v 3 || 4 5 ^v 6 ^v 7 model 5 = T0495.MQAY4-opt3.gromacs0.repack-nonPC.pdb 1 ^v 2 ^v 3 (best Rosetta score, but very little sheet) model 6 = T0495.try7-opt3.unpack.pdb 1 ^v 2 ^v 3 || 4 ^v 5 ^v 6 model 7 = T0495.try6-opt3.pdb 1 ^v 2 ^v 3 || 4 ^v 5 ^v 6 model 8 = T0495.try2-opt3.gromacs0.pdb 1 ^v 2 ^v 3 || 4 ^v 5 (^v 6 ?) model 9 = T0495.try4-opt3.gromacs0.pdb 1 ^v 2 ^v 3 || 4 ^v 5 (^v 6 ?) model 10 = T0495.try16-opt3.gromacs0.pdb 1 ^v 2 ^v 3 || 4 model 11 = T0495.try9-opt3.pdb 1 ^v 2 ^v 3 || 4 || 5 ^v 6 So which of these seems most likely, assuming we are in the c.52.1 superfamily? It looks like the 1 ^v 2 ^v 3 || (4 or 5) is standard for this superfamily, followed by more strands on that sheet, so I prefer try12-opt3 (similar to try9-opt3, except for placement of C-terimnal helix) try7-opt3.unpack (almost identical to try6-opt3 ) try2-opt3 (very similar to try4-opt3, and not that far from try7) though the breaks are pretty bad in models 9-11. I should probably include one of the 2-sheet models (MQAY1 or try15). I don't seem to have any model that have the interspersed sheet that seems quite common in c.52.1. Should I try to get it and from what templates? Tue Jul 29 11:41:59 PDT 2008 Kevin Karplus I don't see any easy way to get that interspersed minor sheet, much as I would like to. I'll just submit what I have. T0495.try12-opt3.pdb # To: karplus@soe.ucsc.edu X-Scan-Signature: 148abf0c5acfe33f338c6ed44fdf9350 X-ASG-Orig-Subj: 495 - WoW !!! Subject: 495 - WoW !!! X-Barracuda-Connect: swlx167.swmed.edu[199.165.152.167] X-Barracuda-Start-Time: 1228893114 X-Barracuda-Virus-Scanned: by Barracuda Spam Firewall at soe.ucsc.edu X-Barracuda-Spam-Score: 0.46 X-Barracuda-Spam-Status: No, SCORE=0.46 using global scores of TAG_LEVEL=3.5 QUARANTINE_LEVEL=5.0 KILL_LEVEL=9.0 tests=PLING_PLING X-Barracuda-Spam-Report: Code version 3.2, rules version 3.2.1.12377 Rule breakdown below pts rule name description ---- ---------------------- -------------------------------------------------- 0.46 PLING_PLING Subject has lots of exclamation marks Kevin: You made some amazing predictions. See one of them: http://prodata.swmed.edu/CASP8/evaluation/T0495h.htm It is way better than all others, what did you do for this one (for the helix position "in front"). We totally ran out if time and submitted crappy ones, as the model with the correct topology was still being worked on by "humans", who are quite slow. Thanks, n From karplus@soe.ucsc.edu Wed Dec 10 08:47:24 2008 Date: Wed, 10 Dec 2008 08:47:23 -0800 From: Kevin Karplus To: grishin@chop.swmed.edu CC: karplus@soe.ucsc.edu In-reply-to: (message from Nick Grishin on Wed, 10 Dec 2008 01:11:41 -0600 (CST)) Subject: Re: 495 - WoW !!! Thanks for the praise on T0495. You can find all my notes for T0495 at http://www.soe.ucsc.edu/~karplus/casp8/T0495/README I spent a lot of time worrying about sheet topology, but not much on the helix.