Fri Jul 18 13:44:06 PDT 2008 TR454 Fri Jul 18 13:44:35 PDT 2008 Kevin Karplus CASP organizers say REFINEMENT TARGET TR454 (the best submitted model according to GDT_TS). Starting model: GDT_TS=64; RMSD=2.41. Multiple areas of possible improvement. Residues 1-4 and 197-203 are missing from the experimental structure and cut out from the refinement target. Make started Fri Jul 18 13:47:18 PDT 2008 Running on moai11.kilokluster.ucsc.edu Fri Jul 18 14:26:27 PDT 2008 Kevin Karplus MQAC quality assessment predicted METATASSER_TS1 with GDT of 72.3% MQAU predicted MULTICOM-REFINE_TS4 with GDT of 74.8% I did dimer optimization on the T0454 target, so may need to on this refinement target also. 1pb6 is the obvious dimerization template. Fri Jul 18 20:44:05 PDT 2008 Kevin Karplus TR454 is very close to pro-sp3-TASSER_TS4 (GDT=94.9%) SAM-T08-server_TS1 is a long way off (GDT=61.3%), very similar in score to T0454.try2-opt3.gromacs0. The main differences between models from different alignments here is off-by-a-turn alignments of the helices and small sideways shifts of the helices. There are no sheets. Sun Jul 20 14:16:25 PDT 2008 Kevin Karplus try2-opt3 scores best with the try2 costfcn, try2-opt3.gromacs0.repack-nonPC scores best in rosetta energy (by a lot). Thu Jul 31 12:17:26 PDT 2008 Kevin Karplus Starting try3 to try to polish try2-opt3.gromacs0.repack-nonPC further. Thu Jul 31 12:21:48 PDT 2008 Kevin Karplus I'll make a dimer from try3 when it is done, using 1pb6A (BLAST's favorite) and 2uxuA (SAM's favorite) as templates. Thu Jul 31 13:40:56 PDT 2008 Kevin Karplus Looking at try3, I can see that it has moved very little from the original. I think I'd like to paste in some pieces from try1-opt3, particularly T110-R138, where try1-opt2 has a continuous helix, but TR454 has an interrupted helix. Maybe also T168-P176. If I make such a chimera, it will need a lot of polishing to close the breaks cleanly. I'll start an optimization for the dimers of try3 first (dimer-TR454.try3-1pb6 has clashes, and dimer-TR454.try3-2uxu is a little loose). Thu Jul 31 13:55:53 PDT 2008 Kevin Karplus dimer/try4 optimization started. Thu Jul 31 14:40:39 PDT 2008 Kevin Karplus try5 started to optimize chimera-try3-try1, which is mainly try3-opt3, but which has T110-R138 and T168-P176 from try1-opt3. Thu Jul 31 16:30:34 PDT 2008 Kevin Karplus try5-opt3 still has bad breaks and clashes, scoring substantially worse than either try2 or try3. Thu Jul 31 17:43:33 PDT 2008 Kevin Karplus try5 seems to have messed up from G170 on, so I'll try copying that from try4-opt3 (dimer/try4-opt3 chain A, that is). Thu Jul 31 17:50:33 PDT 2008 Kevin Karplus Starting an optimization of chimera-try5-try4 as try6. Thu Jul 31 17:59:53 PDT 2008 Kevin Karplus Trying to pack dimer/try4 tighter in dimer/try7. Thu Jul 31 20:21:36 PDT 2008 Kevin Karplus dimer/try7 managed to improve on dimer/try4 (even Rosetta thinks so). try6 is better than try5, but still not as good as try2 or try3. I think I'll try dimerizing it anyway, and seeing what I get when I optimize it in a dimer context. Thu Jul 31 20:35:18 PDT 2008 Kevin Karplus dimer/try8 started to try to optimize the try6-opt3 monomers in a dimeric context. I made the dimers off of 1pb6 and 2uxu, but I probably should have superimposed on the try7 dimer. Fri Aug 1 03:42:00 PDT 2008 Kevin Karplus I don't like the short hairpin between helices in try6 and try8: SheetConstraint (TR454)L171 (TR454)S173 (TR454)T168 (TR454)L166 hbond (TR454)S173 1 For try9, I'll put pack in the other bit of sheet from try3/4/7: SheetConstraint (TR454)I386 (TR454)A388 (TR454)R334 (TR454)L332 hbond (TR454)A388 1 SheetConstraint (TR454)I194 (TR454)A196 (TR454)R142 (TR454)L140 hbond (TR454)A196 1 but optimize from try8. Fri Aug 1 07:44:03 PDT 2008 Kevin Karplus try9 did not close the gap in try8 by much. I'm going to go back to polishing try7 for a while, trying to get a tighter dimer. I'm hoping that try10 will also fix the problem with the CA-CB bond length that has developed in M195 (starting with try3-opt1). Eliminating the sheetconstraint may help here. Fri Aug 1 11:29:57 PDT 2008 Kevin Karplus dimer/try10-opt3 scores best with undertaker, and rosetta likes try10-opt3.unpack.gromacs0.repack-nonPC Fri Aug 1 11:57:52 PDT 2008 Kevin Karplus for dimer/try11, I'll try re-optimizing the gromacs-optimized dimers, but with Hbonds from try10-opt3 as mild constraints. I'll also turn down the helix constraints. For dimer/try12, I'll try reoptimizing without any constraints, starting from the same models as try11. Fri Aug 1 13:15:43 PDT 2008 Kevin Karplus It looks like both try11 and try12 started by squishing the monomers of try10-opt3.unpack.gromacs0 closer together, increasing clashes, then gradually removed the clashes. Tiny breaks tended to come and go. Fri Aug 1 14:29:56 PDT 2008 Kevin Karplus rosetta still likes try10-opt3.unpack.gromacs0.repack-nonPC best. try11 and try12 costfcns like try12-opt3 best, though the clashes and breaks are a bit bigger than in try11 or try10 For try13, I'll turn up clashes and breaks and reoptimize. Fri Aug 1 17:05:53 PDT 2008 Kevin Karplus For try14, I'll make another (monomeric) attempt to close the gaps in the alternative helix model. I made a chimera of try9 and try12 to fix the loop region that has the bad break--basically taking try9-opt3 and replacing A167 to the end with try12-opt3, thus replacing the loop and the final helix. I don't know whether try14 will be able to close the gap, but I hope it comes close enough that I can try dimerizing. Fri Aug 1 17:26:28 PDT 2008 Kevin Karplus Yes, the gap is almost closed very quickly in try14, and clashes are not increased, so I think that try14 will be dimerizable. Fri Aug 1 19:08:29 PDT 2008 Kevin Karplus try14-opt3 has a small break before E169 which gromacs makes worse. I think it is small enough that I'll try optimizing try14 in a dimeric context as try15. Fri Aug 1 21:45:28 PDT 2008 Kevin Karplus try15 does not reduce the break, but scores ok (not as good even as try7, but better than try8 or try9). In addition to the break, try15 is somewhat foamier than try13. I may be ready to submit on this one, as I'm running out of ideas of things to try with undertaker. Sat Aug 2 08:50:56 PDT 2008 Kevin Karplus I made an undertaker script (how-different.under) and a Makefile target (how-different.rdb, how-different.pretty) to see how far the refined models have moved from TR454.pdb. My current favorite models: model GDT TR454.try13-opt3.unpack 97.4 TR454.try11-opt3.unpack 96.9 TR454.try10-opt3.unpack 97.1 TR454.try10-opt3.unpack.gromacs0.repack-nonPC 97.0 TR454.try15-opt3.unpack 80.6 It is clear that none of these have moved the model enough, as the reported GDT of the initial model from the true one was 64%. Of course, it is somewhat safer to move a tiny distance from the provided models, as there are a lot more wrong directions than right ones for changing the model. Still, I think I've been too timid here. At the very least I should try to squeeze try15 harder, to get better packing in the center. Maybe I should also try squeezing try10-try13 harder, with the clash penalty turned down a bit. Sat Aug 2 09:09:14 PDT 2008 Kevin Karplus try16 attempts to close the break and squeeze try15 try17 attempts to squeeze the dimers hard, probably concentrating on try13. Sat Aug 2 09:11:29 PDT 2008 Kevin Karplus The most different models that I have from TR454 are the try1 and try5 models. try1-opt3 has GDT 61.5% and try5-opt3 has GDT 74.1%. try5 has the very bad break between Q178 and D179 which was finally fixed in try14 and try15. I think that I'll also try polishing try1 (closing gaps, mainly) to get another starting point for dimeric optimization. Sat Aug 2 09:24:59 PDT 2008 Kevin Karplus try18 started to do monomeric polishing of try1. Perhaps I should also choose another dimer as a template for creating dimers, in addition to 1pb6 and 2uxu. The obvious ones are the next ones in the PDB blast outputs (2fx0A and 2jj7A) or in the SAM HMM scoring (2f07A and 3bhqA). Sat Aug 2 09:30:51 PDT 2008 Kevin Karplus OK, I've created the dimer-a2m files for those 4 templates, and I'll make a make-dimer-try18 script to use them. Sat Aug 2 13:20:26 PDT 2008 Kevin Karplus try18-opt3 has one moderate break (before D76). I've made 6 dimers from it, and will optimize them in dimer/try19, using the same costfcn as dimer/try17. That one probably does not have clashes set high enough, so I should polish try17 and try19 each again with a higher clash weight. Even try17 isn't really packing the monomers against each other tight enough. Sat Aug 2 16:35:42 PDT 2008 Kevin Karplus Submitted with comment: For this refinement model, I redid the prediction, then did optimizations from the provided model. I also tried making chimeras of the independently generated models and models derived from the provided one. To get better sidechains and packing, I optimized the models in a dimeric context (some optimization was done in a monomeric context, to save time). The resulting dimeric models are still clearly not packed tightly enough, but I'm running out of time and energy to fiddle more with this target. The top-scoring models are still too similar to the provided model to have really gotten to the correct solution. MODEL 1 TR454.try17-opt3.unpack.pdb chain A # < try13-opt3 < try12-opt3 < try10-opt3.unpack.gromacs0 < try7-opt3 < try4-opt3 < dimer-TR454.try3-2uxu < try2-opt3.gromacs0.repack-nonPC < TR454 2 TR454.try11-opt3.unpack.pdb chain A # < try10-opt3.unpack.gromacs0 < try7-opt3 < try4-opt3 < dimer-TR454.try3-2uxu < try2-opt3.gromacs0.repack-nonPC < TR454 3 TR454.try10-opt3.unpack.gromacs0.repack-nonPC.pdb chain A # < try4-opt3 < dimer-TR454.try3-2uxu < try2-opt3.gromacs0.repack-nonPC < TR454 # best rosetta energy 4 TR454.try16-opt3.unpack.pdb chain A # < try15-opt3 < dimer-TR454.try14-2uxu < chimera-try9-try12 chimera-try9-try12: A5-L166 TR454.try9-opt3.unpack.pdb A167-A196 from TR454.try12-opt3.unpack.pdb try9-opt3 < try8-opt3 < dimer-TR454.try6-2uxu < chimera-try5-try4 try12-opt3 < try10-opt3.unpack.gromacs0 < try7-opt3 < try4-opt3 < dimer-TR454.try3-2uxu < try2-opt3.gromacs0.repack-nonPC < TR454 chimera-try5-try4: mostly try5-opt3 G170-A196 from TR454.try4-opt3.unpack.pdb try5-opt3 < chimera-try3-try1 try4-opt3 < dimer-TR454.try3-2uxu < try2-opt3.gromacs0.repack-nonPC < TR454 chimera-try3-try1: mostly from TR454.try3-opt3.pdb T110-R138,T168-P176 from try1-opt3.pdb try3-opt3 < try2-opt3.gromacs0.repack-nonPC < TR454 try1-opt3 < align(1qvtA+1vi0A) # try16-opt3 was an attempt (mostly unsuccessful, but very time-consuming) to move a little bit away from the provided model by using bits of the automatically-generated prediction. 5 TR454.try19-opt3.unpack.pdb chain A # < dimer-TR454.try18-2uxu < try18-opt3 < try1-opt3.repack-nonPC < align(1qvtA+1vi0A) # This re-prediction (without input from the provided model) is too far from the provided model to be correct.