Fri Jun 20 09:02:27 PDT 2008 T0465 Sun Jun 29 14:19:21 PDT 2008 Kevin Karplus The MQA run for T0465 produced lots of core dumps. undertaker: XYZbins.cc:157: XYZbins::XYZbins(const Pointlist&, const int*, int, double): Assertion `NumBinsX >0' failed. apparently while processing panther_server_TS1 The problem seems to be a lot of "nan" values in the locations of the atoms. Sun Jun 29 14:36:35 PDT 2008 Kevin Karplus I made a minor modification to undertaker so that when it reads in atoms it rejects any with non-finite coordinates. This should fix the problem, I think. Make started Mon Jun 30 12:49:09 PDT 2008 Running on peep.cse.ucsc.edu Mon Jun 30 14:47:07 PDT 2008 Kevin Karplus The CASP organizers changed the classification of this target: T0465 - switching to human/server target We are switching T0465 over to human/server target. In order to provide enough time for human-regime modeling, we are adding 10 extra days to the prediction window for this target so that it will expire for human prediction on July 21. This is one that the MQA assessments felt was hard, with predicted GDT of only 27%-33%. Somewhat surprisingly, the SAM-T08-server_TS1 model was predicted to be tops by the MQAC consensus method (not the MQAU method, which is biased in favor of SAM-T08 by the shared use of the alignment constraints). The HMM searches don't have any confident finds. Best e-values are around 50-70 for a.39.1.5. Mon Jun 30 17:12:01 PDT 2008 Kevin Karplus Started MQAU1 and MQAC1 metaserver runs. Mon Jun 30 19:55:45 PDT 2008 Kevin Karplus Best with try1 are MQAC1-opt3 < SAM-T08-server_TS1 MQAU1-opt3 < SAM-T08-server_TS1 try1-opt3.repack-nonPC < align(1wdcB) I need to start a MQAX1 metaserver run that excludes the SAM servers. Mon Jun 30 20:10:42 PDT 2008 Kevin Karplus So far, it looks like there are no particularly good templates for this target---the SAM-T08-server and the run here have little if anything in common. The small number of homologs in the t06 alignment seem to be mostly phage proteins. Mon Jun 30 21:37:52 PDT 2008 Kevin Karplus MQAX1 is optimizing BAKER-ROBETTA_TS5 Tue Jul 1 14:59:53 PDT 2008 Kevin Karplus The MQAX1 scores best with the try1 costfcn, and has two hairpins consistent with the secondary structure prediction. Burial is not very good though. The rr prediction is fairl compatible with MQAX1. I'll do a polishing run with MQAX sheets, and a run from alignments with the same costfcn. Tue Jul 1 15:57:22 PDT 2008 SAM-T08-MQAO hand QA T0465 Submitted Tue Jul 1 15:57:22 PDT 2008 SAM-T08-MQAU hand QA T0465 Submitted Tue Jul 1 15:57:22 PDT 2008 SAM-T08-MQAC hand QA T0465 Submitted Tue Jul 1 19:21:46 PDT 2008 Kevin Karplus The try2 polishing run doesn't change MQAX much. The try3 run looks like it is trying to form a 4-strand sheet. I wonder if I can make it do it. It is a hairpin-helix-hairpin arrangement. Sat Jul 12 13:15:53 PDT 2008 Kevin Karplus The try4 costfcn has constraints for the two hairpins that try3 seems to be trying to form, but not for a full sheet. I started a try4 run from alignments and an MQAX4 run from servers. Sat Jul 12 16:27:14 PDT 2008 Kevin Karplus MQAX4-opt3 scores better than try4-opt3, which actually scores rather poorly. I wonder if I can force an anti-parallel connection between the two hairpins of MQAX4. The helix between I49 and N61 (T51-E60) implies a parallel arrangement for N40-I49 and N61-V67, but either of the other two strands could be in between, and the phase is not obvious. There are some residue-residue contact pairs joining those strands, but they're not very consistent: F44 N61 +17 F46 Y64 +18 F46 N61 +15 K47 K65 +18 A48 Y64 +16 Even the even/odd prediction isn't consistent. Here are the 2 hairpins: SheetConstraint T63 K68 D77 R72 hbond Y64 1 SheetConstraint I30 K37 K47 N40 hbond S31 1 Perhaps I should look for predicted contacts between anything in 30-49 and 60-79: F44 N61 17 F46 Y64 18 F46 N61 15 K47 K65 18 A48 Y64 16 A48 K65 15 P43 Y64 21 I45 K65 20 I49 L76 27 I45 T63 18 A48 L76 28 I'm also not wild about the separation of 4 on the turn between the 60s and the 70s strand. I'd much rather see 3 or 5. I'm not really seeing a way to do strand-helix-strand with such a short helix and long strands. Mon Jul 14 13:55:40 PDT 2008 Firas Khatib I looked at other models to see if anyone was close to getting the strand-helix-strand for I49-V67 and the closest I found was the server model: FALCON_CONSENSUS_TS2. I think starting with this model it might be possible to rotate N61-V67 so that they form a parallel sheet with N40-I49. I will attempt to do this for try5. I added this sheet constraint to try5.costfcn: ConstraintSet parallel.sheets SheetConstraint T63 V67 V41 I45 Hbond P43 1 try5 is running on shaw Mon Jul 14 17:13:55 PDT 2008 Firas Khatib try5 is complete garbage because FALCON_CONSENSUS_TS2 only has CA and CB as sidechains, so try6 is exactly the same as try5 but with SCWRL run on FALCON_CONSENSUS_TS2 first. try6 is running on shaw Mon Jul 14 22:05:12 PDT 2008 Firas Khatib so try6 almost worked, it added some chain breaks (which is what I wanted it to do so that it would make the parallel strand. I will run it again with stronger weights on forming that parallel strand and worry about closing the chain breaks later try7 is running on shaw, I will launch a similar run with try8 allowing even more clashes just to see if that sheet can form successfully. try8 is running on shaw Tue Jul 15 13:50:12 PDT 2008 Firas Khatib try7 and try8 were able to form a parallel strand with only 1 break in the entire chain, but the sheet is quite ugly. I'm going to try to fix it. I think the main problem is that V41 and T63 are 11.38A apart, so if we can pull T63 down more (bringing the helix closer) it would look like a proper sheet. I will add in some distance constraints to V41/T63 and V42/Y64 for try9 try9 is running on shaw Tue Jul 15 16:58:55 PDT 2008 Firas Khatib darn it, those distance constraints broke the sheet in try3.sheets I'll ramp the try3.sheets weight up and add more distance constraints to the "flatten" ConstraintSet for try10 try10 is running on shaw Wed Jul 16 13:38:52 PDT 2008 Firas Khatib ok, try10 got the strands next to each other, but the other strands are messed up and nothing really looks like a sheet. I'm going to run try11 without the distance constraints, but with the sheet constraints up high and try12 with the same but the distance constraints still there (with a lower weight) if this doesn't work I might just combine different models together, the ones with the nice sheets and then this region with the parallel strand both try11 and try12 are running on shaw Wed Jul 16 15:29:23 PDT 2008 Firas Khatib well, removing the distance constraints reverted try11 back to the similar models of try6/try7/try8 although it does a bit better with the hand.sheets ConstraintSet I'm going to ramp up the sheet constraints with this model since my distance constraints did not do what I wanted them to. Looking at them, I should have used CBs and not CAs... that is probably why they messed everything up. try13 is running on shaw Wed Jul 16 15:51:15 PDT 2008 Firas Khatib try12 just ended and is garbage ok... at this point I have attempted 9 tries on this and pretty much have failed miserably. so while try13 is running I'm going to try something different. starting from T0465.MQAX4-opt3 I will try to bring the two strands closer to each other actually, BAKER-ROBETTA_TS1 has all 4 strands forming a sheet: S31-V35 & V41-I45 are anti-parallel V41-I45 & G71-K74 are parallel Y64-V67 & G71-K74 are anti-parallel that is a bit different than what we wanted: V41-I45 & T63-V67 being parallel but might be worth playing with although the burial on the helices isn't very good (maybe we can just copy the sheet) Wed Jul 16 16:34:37 PDT 2008 Firas Khatib I modified T0465.MQAX4-opt3 to T0465.MQAX4-modified.pdb by inserting breaks around the T51-E60 helix (breaks inserted from I49-T62. For try14 I will put in sheet constraints to try to bring the 2 sheets closer to one another (and hopefully with that helix out of the way it might work and swing everything before K47 closer to T63) try14 is running on shaw Wed Jul 16 18:16:52 PDT 2008 Firas Khatib yeah, that was wishful thinking indeed... nothing has budged in opt1 or opt2 so I doubt opt3 will look any different. I almost want to ProteinShop this one. Let me see how hard that will be to do. Wed Jul 16 18:22:22 PDT 2008 Firas Khatib ok... now I am mad at myself for not thinking about this earlier... this is the way to do it (and what I should have been doing this whole time instead of putting in incorrect distance constraints). I made a folder: decoys/proteinShop/ and in a couple minutes was able to make try14-opt2-modified.pdb I will run try15 on this and then come back after dinner (and try15 has run) and do a lot more proteinshopping! Wed Jul 16 18:35:11 PDT 2008 Firas Khatib ok, I have to take it back, T0465.try14-opt3 just finished and actually was able to form the sheet with many chain breaks and messing up the T63-N66/G71-K74 sheet. try15 and try16 are very similar except that that try15 uses T0465.try14-opt3 as input and try16 uses try14-opt2-modified.pdb both try15 and try16 are running on shaw Wed Jul 16 21:06:15 PDT 2008 Firas Khatib dang, try16 didn't work because the proteinshopped pdb file was "incomplete conformation T0465 can't currently be optimized by undertaker" I had to go back to CASP7 target T0283's README file to find my old script to fix this: renumberChain.pl try16 is running again on shaw with T0465.try14-opt2-modified.clean2.pdb as input Wed Jul 16 21:36:29 PDT 2008 Firas Khatib looking at try15-opt3, I don't understand what happened to this sheet: T63-N66/G71-K74. It was good in T0465.MQAX4-opt3 but got messed up in try14 so I increased the hand2.sheets constraint, but it is just as bad. Looking at score-all.try15.pretty they seem to be scoring well: try15-opt3 = -43.0 for hand2.sheets try14-opt3 = -38.3 for hand2.sheets try13-opt3 = -60.1 for hand2.sheets even though they look bad to me! I am going to use the sheets from MQAX4-opt3.sheets for try17 try17 is running on shaw and will be based on try15-opt3 Wed Jul 16 22:18:36 PDT 2008 Firas Khatib I proteinshopped a model with the parallel strand that I had tried to do this whole time (try5-try13): decoys/proteinShop/flat-and-parallel.pdb try18 will have a similar costfcn to try11, with flat-and-parallel.clean.pdb as input try18 is running on lopez Wed Jul 16 22:33:45 PDT 2008 Firas Khatib recapping so that I can remember what is what: try6/try11/try13 < FALCON_CONSENSUS_TS2 (with try11 being the best) try17 < try15 < MQAX4-opt3 < BAKER-ROBETTA_TS3 try16 < proteinshopped try14-opt2 (named: T0465.try14-opt2-modified.clean2.pdb) try18 < proteinshopped try6-opt3 (named: flat-and-parallel.clean.pdb) and I still think that BAKER-ROBETTA_TS1's sheets might be worth using since they are different than everything else Note: none of these models have been optimized for burial or chainbreaks yet. Thu Jul 17 11:47:13 PDT 2008 Kevin Karplus Firas, have you updated superimpose-best.under to put your favorite models first? Have you decided to do only meta-server models and not submit any of the SAM+undertaker models? Thu Jul 17 11:54:50 PDT 2008 Firas Khatib Yes, superimpose-best.under was updated before the last 3 runs I launched last night (try16-try18). I haven't worked on the SAM/undertaker models, but I think that would be a great idea. I was really just trying to get the strand-helix-strand thing to work this whole time (first with parallel strands and then when I got very frustrated, I tried anti-parallel ones) and the server models seemed like a decent start point. If you don't mind taking a look and picking a direction to focus the remaining work on, that would be great. Thu Jul 17 12:58:23 PDT 2008 Kevin Karplus I modified superimpose-best.under to extract sheet constraints robetta1 and falcon2. The falcon2 sheets SheetConstraint E141 E144 D139 D142 hbond E144 1 SheetConstraint S118 G120 V116 S118 hbond G120 1 SheetConstraint R72 K74 K68 N66 hbond K74 1 SheetConstraint N40 F44 D36 K32 hbond F44 1 include 2 impossible separation=2 hairpin. The SheetConstraints I would want from that model are SheetConstraint N40 F44 D36 K32 hbond F44 1 SheetConstraint R72 K74 K68 N66 hbond K74 1 but even those are unlikely separation=4 hairpins. The robetta1 sheets SheetConstraint G71 L76 V41 F46 hbond L76 1 SheetConstraint K65 V67 V73 G71 hbond K65 1 SheetConstraint N40 F46 D36 I30 hbond F46 1 are also separation=4 hairpins. try2-opt3.sheets has one of the hairpins more reasonably at separation=3 SheetConstraint (T0465)Y64 (T0465)K68 (T0465)E75 (T0465)G71 hbond (T0465)N66 1 SheetConstraint (T0465)V28 (T0465)D36 (T0465)A48 (T0465)N40 hbond (T0465)I30 1 but the other still has separation=4 try3-opt3 almost forms a 4-strand antiparallel sheet, with strands 1 and 4 antiparallel. One hairpin is a sep=3 hairpin SheetConstraint (T0465)V41 (T0465)I45 (T0465)D36 (T0465)K32 hbond (T0465)F44 1 that is, SheetConstraint N40 I45 K37 K32 hbond N40 The other is too sloppy for undertaker to detect, but could be SheetConstraint Y64 V67 V73 R70 hbond R70 We could join them into a 4-strand sheet with SheetConstraint R33 D36 K74 G71 hbond V73 Thu Jul 17 13:28:49 PDT 2008 Kevin Karplus I'll put these into try19.costfcn, and do a metaserver MQAX19 run and a try19 run from existing models. I'll also do a try20 run from alignments with the same costfcn. Thu Jul 17 13:31:47 PDT 2008 Kevin Karplus try19, try20, and MQAX19 started on the moai cluster. The rosetta runs may step on each other if the jobs end at the same time---I really ought to fix the make file to be safer with simultaneous makes. Firas, look at the way I set up try19 and see if there is another set of sheet constraints that you would like to try. It may be necessary to set the weight for the sheet constraints higher to force things to happen. Thu Jul 17 15:05:14 PDT 2008 Firas Khatib the ConstraintSet four_strand.sheets look fine by me, I hope it'll be able to form those sheets! I see that try16-18 were all useless... damn! Thu Jul 17 16:21:08 PDT 2008 Kevin Karplus It looks like try19 scores better than try20, but try20 does better on the 4-strand sheets. I think that MQAX19-opt3 (from BAKER-ROBETTA_TS4) will beat them both, when it finishes. Thu Jul 17 21:42:54 PDT 2008 Kevin Karplus I've put MQAX19 and try20 at the front of superimpose-best.under, but I don't necessarily believe them to be best. Firas, please look at the models and pick out the best 5 soon. Fri Jul 18 12:36:56 PDT 2008 Firas Khatib My 3 favorites looking only at burial: T0465.try11-opt3.pdb < FALCON_CONSENSUS_TS2-scwrl.pdb T0465.MQAC1-opt3.pdb < SAM-T08-server_TS1 T0465.MQAX4-opt3.pdb < BAKER-ROBETTA_TS3 then I would include: T0465.MQAX19-opt3.pdb < BAKER-ROBETTA_TS4 and for the last one I really don't know, I guess try20 or try3 just to be different, but they aren't very good models Fri Jul 18 14:56:20 PDT 2008 Kevin Karplus Since I don't have any good ideas on this target, I'll submit Firas's choices, but I'll toss in try19, rather than try20, as the 5th model. Sat Jul 19 09:07:52 PDT 2008 Kevin Karplus Nope: try19 is almost identical to MQAX4. Sat Jul 19 09:41:34 PDT 2008 Kevin Karplus try20-opt3.gromacs0.repack-nonPC seemed to be based on try19, not try20 (probably because rosetta optimizations can step on each other). I re-ran the repacking in series, rather than in parallel to be sure of getting the right models. The repacking of try20 scores badly, because of big breaks. ReadConformPDB T0465.try11-opt3.pdb # < FALCON_CONSENSUS_TS2-scwrl.pdb ReadConformPDB T0465.MQAC1-opt3.pdb # < SAM-T08-server_TS1 ReadConformPDB T0465.MQAX4-opt3.gromacs0.repack-nonPC.pdb # < BAKER-ROBETTA_TS3 # best rosetta energy ReadConformPDB T0465.MQAX19-opt3.pdb # < BAKER-ROBETTA_TS4 ReadConformPDB T0465.try20-opt3.pdb # < align(2z69A?)