Thu May 22 09:39:46 PDT 2008 T0413 Make started Thu May 22 09:40:46 PDT 2008 Running on cheep.cse.ucsc.edu Thu May 22 09:54:58 PDT 2008 Kevin Karplus BLAST finds no close homologs. The crystallographers report that this is "likely dimer". Thu May 22 12:44:21 PDT 2008 Kevin Karplus The iterated searches take a while---t06 found almost 25,000 matches! Thu May 22 16:37:12 PDT 2008 Kevin Karplus The c.69.1 superfamily matches well with the HMMs, but there are at least 35 families, with matches from all of them. Each multiple alignment favors a different template, from a different family. t06: 1l7aA 319 5.1095e-27 c.69.1.25 77772 2hu5A 574 1.6989e-26 b.69.7.2,c.69.1.33 136763,136764 2z3zA 652 1.7317e-25 t04: 2hu5A 574 1.1065e-27 b.69.7.2,c.69.1.33 136763,136764 2d5lA 666 9.6062e-27 2z3zA 652 2.5536e-25 t2k: 2jbwA 360 2.6069e-14 2d5lA 666 1.1029e-12 1jjfA 256 2.6736e-12 c.69.1.2 66765 (The pdb_blast top hit, 1hlgA, is a c.69.1.6 family member.) Try1 seems to favor 2hu5A, even though it results in a lot of clashes which undertaker is having some trouble removing. (The run still has a ways to go, and clash reduction gets more important in the later stages of optimization.) Fri May 23 05:55:58 PDT 2008 Kevin Karplus try1-opt3 has some bad breaks that need to be fixed, but otherwise looks pretty good. The templates are not quite full length, but I'm not going to fuss about the short N-terminal and C-terminal ends. Fri May 23 09:22:56 PDT 2008 Kevin Karplus try2-opt3 still has bad breaks, particlularly before W261,Q262, and W74. I should do another run (perhaps polishing the existing models) with breaks set higher, but clashes left pretty low. try3-opt3 has reduced clashes and breaks, but at some loss of the align_constraints. I also noticed that strand H167-A171 does not seem to be properly Hbonded. This is probably because try1.sheets and try2.sheets disagreed about its alignment, so try3 had conflicting constraints. I suspect that a stripe across the sheets is P55,Y143,A168,P207, which is more compatible with try1 than try2. There is also some question about how to align N25-D31 Fri May 23 14:11:34 PDT 2008 Kevin Karplus I tried picking up sheets from the first alignment for try4, which starts over from the alignments, rather than trying to polish existing models. Fri May 23 21:06:04 PDT 2008 Kevin Karplus try4-opt3 has moderately bad breaks and clashes, but good hbonds and align_constraints. I'll do a polishing run with try4.sheets and try4.helices. Sat May 24 05:06:47 PDT 2008 Kevin Karplus try5-opt3.gromacs scores the best with the try5 costfcn, but still has some breaks. I replaced the clash definition in costfcn-init.under: # ReadClashTable exp2-pdb-dunbrack-40pc-3157-2symm.clash # SetClashDefinition exp2-pdb-dunbrack-40pc-3157-2symm ReadClashTable exp2-pdb-cullpdb_pc80_res1.2_R0.2_d070810_chains408-2symm.clash SetClashDefinition exp2-pdb-cullpdb_pc80_res1.2_R0.2_d070810_chains408-2symm which made the gromacs models do better. Another polishing run with breaks turned up even higher may be in order. Sat May 24 10:44:03 PDT 2008 Kevin Karplus try6-opt3.gromacs0 scores best with the try6 costfcn, and has reduced clashes and breaks, though one helix has been disrupted a bit between A122 and A125. This is probably because we had no helix constraint there. There is a bit of foaminess, but no worse than previous models. Sat May 24 20:22:13 PDT 2008 Kevin Karplus try7-opt3 reduces the clashes and gets better helices and hbonds, but doesn't fix A122-A125. It might be worth copying V116-R126 from an earlier model (such as try3-opt3) and reoptimizing the resulting chimera to fix the breaks. I don't really believe in the helix that includes P222-P228. The three prolines there almost certainly represent selection for a non-helix structure. All our models have the helix, though, so I don't see that there is much we can do about it yet. Sun May 25 08:52:45 PDT 2008 Kevin Karplus try8-opt3 fixed the helix, but has some bad breaks: T0413.try8-opt3.pdb.gz breaks before (T0413)D37 with cost 1.76483 T0413.try8-opt3.pdb.gz breaks before (T0413)W261 with cost 1.74479 T0413.try8-opt3.pdb.gz breaks before (T0413)G260 with cost 1.5267 T0413.try8-opt3.pdb.gz breaks before (T0413)A36 with cost 0.949603 T0413.try8-opt3.pdb.gz breaks before (T0413)V116 with cost 0.832578 T0413.try8-opt3.pdb.gz breaks before (T0413)P52 with cost 0.623285 Rosetta thinks that try8-opt3.gromacs0.repack-nonPC has worse clashes than try7-opt3.gromacs0.repack-nonPC. the try8-opt3.gromacs0 model has smaller clashes but worse breaks than try8-opt3. Perhaps I should turn breaks up more and start from just the gromacs-optimized models. Sun May 25 09:09:08 PDT 2008 Kevin Karplus started as try9 on peep. try9-opt3 reduces breaks slightly and clashes substantially compared to try8-opt3. Rosetta likes try9-opt3.gromacs0.repack-nonPC best, though there is not much differene between try9 and try7 on this measure. I think I'll let this one sit until we have metaserver models to run. Mon May 26 17:48:02 PDT 2008 Kevin Karplus MQAC likes Zhang-Server models best, the FEIG. MQAU also likes Zhang-Server best, then SAM-T08-server, then FEIG. The bulk of the models are very similar, but I like try9-opt3 better than the Zhang-Server of FEIG models. It looks like the edge of the sheet should still be buried, so I wonder if this is a dimer. Probably, since the top hit, 2hu5A, is part of a dimer. It will take forever to optimze as a dimer, but I probably should do it anyway, starting with try9-opt3 and 2hu5A. Mon May 26 18:08:31 PDT 2008 Kevin Karplus The dimer looks pretty good, but we need to fix up the sheet in the middle, with a sheet constraint for anti-parallel strands Q262-V266 between the monomers (or even G260-P267) Mon May 26 18:26:11 PDT 2008 Kevin Karplus I started the optimization as dimer/try10 on the moai cluster. Tue May 27 20:33:04 PDT 2008 Kevin Karplus dimer/try10 is a total failure! The monomers were rotated and moved apart! The "dimer" constraint weight needs to be about as high as soft_clashes to keep the monomers in place. Let's try again with the dimer weight around 100. Tue May 27 20:39:57 PDT 2008 Kevin Karplus I started another optimization with para-trickle-make (dimer/try11), with the dimer constraint set much higher and soft_clashes set lower. Wed May 28 09:28:47 PDT 2008 Kevin Karplus dimer/try11 improved the dimerization slightly, but don't form good H-bonds across the interface---the parts are too close. There are also some backbone problems around P258-G260. I should turn breaks and clashes up and reoptimize. Wed May 28 09:43:59 PDT 2008 Kevin Karplus started dimer/try12 on cluster. Wed May 28 11:16:39 PDT 2008 SAM-T08-MQAO hand QA T0413 Submitted Wed May 28 11:16:39 PDT 2008 SAM-T08-MQAU hand QA T0413 Submitted Wed May 28 11:16:39 PDT 2008 SAM-T08-MQAC hand QA T0413 Submitted Sun Jun 1 11:48:51 PDT 2008 Kevin Karplus I started two metaserver runs with a cost-function that includes sheet and helix constraints from dimer/try11, though dimer/try12 looks like my best model so far. Sun Jun 1 12:15:03 PDT 2008 Kevin Karplus MQAC seems to be based mainly on FEIG_TS1 MQAU seems to be based mainly on SAM-T08-server_TS1 Sun Jun 1 12:22:58 PDT 2008 Kevin Karplus One thing that just occured to me---the C-terminal helices for the dimer may be domain swapped. That would allow them to pack better agains the other helices. Getting that packing to happen without any tools for manipulating the proteins may be difficult. Sun Jun 1 14:38:24 PDT 2008 Kevin Karplus I like the way that MQAU1 tucks G234-G260 out of the way of the dimer, but I'm not so fond of other parts of it. Sun Jun 1 16:11:34 PDT 2008 Kevin Karplus I made a chimeric model consisting mainly of try9-opt3, but with G234-G260 copied from MQAU1. I'll optimize this with the try12.costfcn (used for optimizing MQAU1 and MQAC1) as try12, then see about making a dimer out of it, perhaps based on the existing dimer/try12 Sun Jun 1 16:34:02 PDT 2008 Kevin Karplus Hmm, it looks from try12-opt1 that the chimera will not get fixed into something dimerizable, as closing the gaps for the helix has made it wander out over the dimer interface. Perhaps I have to dimerize the chimera and optimize it in the dimer. Sun Jun 1 16:57:24 PDT 2008 Kevin Karplus I made a dimer from chimera-try9-MQAU1.pdb.gz using dimer/try12-opt3 as a template (see make-dimer-chimera-try12.under). I'll optimize this (to try to close the bad breaks) as dimer/try13 Sun Jun 1 17:59:47 PDT 2008 Kevin Karplus The try12-opt3 run is indeed a bust, with the helices swinging out over the dimer interface. It looks like dimer/try13 is going to have trouble also, since the distance from L257 to W261 is too large to span with only 3 residues, and can't be shrunk without moving the helix ending at R232 out of the way first. There may be a rotation that works, but the chance of undertaker finding it is small. I may have to do another run with breaks turned up much higher than clashes, to get out of the local minimum. Since dimer/try13 is going to take a while, I might as well start that one at the same time on another machine. Sun Jun 1 21:57:05 PDT 2008 Kevin Karplus dimer/try14-opt3 still doesn't close the bad breaks before P258 and W261 (and another, smaller one before D37). It looks like I could help things along a little with a few extra constraints: perhaps Hbonds R250.NH1 Y245+304.O, R250.NH2 E246+304.O Maybe Q254.NE2 P180+304.O Moving G234,P235-L257 out so that it has to repack may also be useful. Sun Jun 1 22:16:42 PDT 2008 Kevin Karplus I made dimer/decoys/expoded-try14.pdb by moving out G234 a little and P235-L257 a lot. For dimer/try15, I added ConstraintSet overlap Hbond R250.NH1 Y549.O Hbond R554.NH1 Y245.O Hbond R250.NH2 E550.O Hbond R554.NH2 E246.O Hbond Q254.NE2 P484.O Hbond Q554.NE2 P180.O and upped the weight of breaks to 1200. Sun Jun 1 22:38:29 PDT 2008 Kevin Karplus I cancelled the dimer/try15 run and restarted it with a typo fix---the exploded file was not getting read, so it was trying to start from a random backbone! Sun Jun 1 23:35:01 PDT 2008 Kevin Karplus It looks like try15 found a way to remove the breaks, but only by sticking Q233-R250 out into empty space. Sigh. I wish I had tools to manipulate proteins more directly! Mon Jun 2 09:04:28 PDT 2008 Kevin Karplus I tried adding some more constraints to hold the helix in place and upped the weights of the overlap cost for dimer/try16 ConstraintSet hold_helix Hbond Y244.OH S226.OG Hbond Y548.OH S530.OG Constraint R242.CD V193.CB 4 5.5 7 Constraint R546.CD V497.CB 4 5.5 7 (Note: the actual run of dimer/try16 was missing the last constraint, because of a typo in the file that was not fixed in time. Because the optimization is being done with symmetrization, it should not matter much.) There seems to be something wrong with the "overlap" set, as all the dimer models are getting the same value (perhaps because no Hbond is being formed). But the value is -0.6930, which seems too good for an unformed H-bond. Could there be a bug in the Hbond scoring? Mon Jun 2 09:21:51 PDT 2008 Kevin Karplus Yes, there is a documented bug in the HbondDistProb function that says that the fits are too lenient on the long side for non-backbone Hbonds. I think that the problem may be that I forgot to do the "ideal gas" correction to the density functions. I should look into that today and update the Hbond cost functions. Meanwhile, if I want the constraints to mean anything, I should use distance constraints from hold_helix and overlap. Mon Jun 2 09:28:29 PDT 2008 Kevin Karplus I killed dimer/try16, since the extra constraints didn't mean what I wanted them to. Mon Jun 2 10:24:56 PDT 2008 Kevin Karplus For dimer/try17, I added constraints to parallel the Hbonds of dimer/try16 and am trying again. Mon Jun 2 11:21:54 PDT 2008 Kevin Karplus dimer/try17-opt1 crosses over the two helices we're trying to pack out of the way and makes terrible clashes and still has breaks. Perhaps they'll be resolved later in the optimization, but I'm not very hopeful. Tue Jun 3 19:24:44 PDT 2008 Kevin Karplus As expected, the dimer/try17-opt3 has a terrible clash between the monomers around Q249. Sat Jun 7 09:29:08 PDT 2008 Kevin Karplus So what do have right now? The chimera I made has not been optimizable. Perhaps I need to redo the chimera, copying R232-L263 from MQAU1 and the rest from try9-opt3. Sat Jun 7 09:42:40 PDT 2008 Kevin Karplus This is now chimera2-try9-MQUA1, and I'll make a dimer out of it: dimer/decoys/dimer-chimera2-try12.pdb I'll try optimizing that dimer with the dimer/try13 costfcn as dimer/try18 Sat Jun 7 12:48:53 PDT 2008 Kevin Karplus dimer/try18 looks pretty good. I'll try polishing it up by doing a run from all dimer models as dimer/try19. Then I'll have to pick some of my favorite monomers. The dimer/try18 model does still have 2 terrible breaks: before P258 and before R232. Both have been reduced from the chimera2 model, but not really enough. I doubt that dimer/try19 will fix these much, so we may have to do some deliberate chopping around the breaks, so that we have little segments that can be rejoined better. Sat Jun 7 20:10:25 PDT 2008 Kevin Karplus dimer/try19 has one terrible break, before P258, and one sort of bad one before D37. The distance between L257.CA nd P258.CA is about 3 Ang too big. The Hbonds between the monomers are also stretched way too far. I need to increase the weight of the constraints between them. Sat Jun 7 20:26:52 PDT 2008 Kevin Karplus I'm repeating dimer/try19 as dimer/try20, with the dimerization constraint and with break cost turned way up. I hope it can close things up without making a mess. Sun Jun 8 08:45:38 PDT 2008 Kevin Karplus The dimer constraint in dimer/try20 seem ok now, and breaks are way down. I could extend the dimer constraint to W261-P267 (rather than just Q262-V266) but that seems to have formed without being forced. I think that some flexing of the sheet may happen to pack one side or the other of the sheet more tightly. I put together a dimer/try21 costfcn, with less stringent dimer and break weights, to see what models are currently favored: dimer/try20-opt3 dimer/try19-opt3 dimer/try14-opt3.unpack.gromacs0 dimer/try18-opt3 dimer/try13-opt3 Rosetta likes best dimer/try12-opt3.unpack.gromacs0.repack-nonPC Most of the dimer models have very similar monomers. Currently I favor submitting dimer/try20-opt3 < dimer-chimera2-try12 chimera2 mainly try9-opt3, but R232-L263 from MQAU1 dimer/try12-opt3.unpack.gromacs0.repack-nonPC