Thu Aug 26 14:02:57 PDT 2004 T0270 DUE 30 Aug 2004 Thu Aug 26 14:03:38 PDT 2004 Kevin Karplus I moved the old T0270 directory to T0270-pre1t0t, and am starting over in this directory with a completely new prediction (new aligments, new search, ...) now that 1t0tV is in the template library. The re-doing of the search is based on Alexey Murzin's comment on FORCASP: "The 1T0T coordinates have been released today. Happy comparative modeling of T0270." Thu Aug 26 17:42:31 PDT 2004 Kevin Karplus try1-opt2 looks fairly good, but I worry a bit about the alignments of some of the strands---some of them may be off by 1. I'll try using the rr constraints to help orient the strands on try2 (starting again from alignments). The constraints involving Y226 seem unlikely, and I'll remove them. I'll change the F108, V67 one to use CZ and CG2. Thu Aug 26 19:37:03 PDT 2004 Kevin Karplus try2 does not seem to be doing better than try1, even with the try2 costfcn. It does do slightly (0.09 out of 193) better on the unconstrained costfcn. I made a 5mer of try1 in 5mer-1t0tV-try1.pdb It looks pretty good, with only a few clashes between the oligomers: select *:A and within(2.0, *:B or *:E) 44 atoms selected! The 5mer of try2 does slightly better: select *:A and within(2.0, *:B or *:E) 40 atoms selected! There is no obvious change that we can make that will improve these (other than optimizing a double or triple chain, which would be very slow. What I'll do is to start a polishing run from the existing models, using a copy of the unconstrained costfcn. Fri Aug 27 09:30:21 PDT 2004 Kevin Karplus The differences between try1, try3, and the template are the first 10 residues M1-H10 residues K119-P123 the last 3 residues L247-L249 Of these, only P8 is in the multimeric interface, though T9 is close. The initial and middle sections are almost adjacent, but the C-terminal end is 40 Angstroms away---about as far as it can get in this protein. Not much is moving in this protein, but I should probably do one more polishing run with clashes and breaks turned up high. Fri Aug 27 11:55:45 PDT 2004 Kevin Karplus The try4 run reduced breaks and clashes without moving much. Rosetta now likes try4-opt2.repack-nonPC best. I'm a little worried about the chain at D122---it seems to have a sharp kink (picked up from try3). Maybe I should do another optimization run that does not include try3 and try4, to see if I can get a different closure for the loop there. Fri Aug 27 14:07:54 PDT 2004 Kevin Karplus try5 independently generated the same conformation as try4 around D122, so it seems that the fragment library I'm using and undertaker like that conformation. Oops---no, it came from try2, not try3. Let's do yet another optimization run, starting only from try1. Fri Aug 27 14:57:21 PDT 2004 Kevin Karplus try6-opt1 still has a separation of 5.922 Ang between P120 and L121 CA atoms (should be 3.8). If this is not fixed in the second round of optimization on try6, I might want to add some gap-closing constraints to push it harder. Fri Aug 27 16:08:02 PDT 2004 Kevin Karplus try6 does not score as well as try4 or try5, but beats try1 through try3. The P120.CA-L121.CA distance is still too large at 5.871. It may be fixable by changing the psi angle of K119 and the phi angle of D122. If I could figure out how to use deep-view, maybe I could fix it. Undertaker sees some H-bonds (D122.N-P120.O, L121.N-K119.O), which are keeping it from closing the gap. I suppose I could add a gap-closing constraint for P120-L121 to close the gap, and do a quick run. Fri Aug 27 16:52:25 PDT 2004 Kevin Karplus try7 is trying to close the gap, but I fear I have the constraints set too high, so that it will do enormous damage in trying to close it, and miss better solutions that come close to closing the gap but don't do so much damage. Fri Aug 27 17:01:36 PDT 2004 Kevin Karplus Well, try7-opt1 did not make a complete hash of things. It damaged the helix at P126-K129, but does not seem to have created breaks elsewhere. It may be possible to do a crossover operation and patch the try7 loop into try6. What try7-opt1 does is to add a tight kink at Y127, the way try2-5 did at D122. Perhaps I should transplant E118-E124 from try7-opt1 to try5-opt2. Maybe I'll wait until try7-opt2 is available. The resulting chimera will need reoptimization to close the resulting (small) breaks. Fri Aug 27 17:34:30 PDT 2004 Kevin Karplus I made the transplant from try7-opt2 to try5 -opt2, calling the result decoys/try5-try7-chimera.pdb. The result scores better than try6 and nearly as well as try4 and try5 (with the try6 costfcn). I'll try doing a short polishing run. Fri Aug 27 17:41:08 PDT 2004 Kevin Karplus It looks like Sol has added a patched try6 from deepview also. I'll toss that into the optimization also. Fri Aug 27 17:45:15 PDT 2004 Sol Katzman Yes, I used deepview to tweak K119-phi/psi and K122-phi/psi, keeping them in the legal ramachandran regions. That fixed the break at P120-L121. Fri Aug 27 18:02:24 PDT 2004 Kevin Karplus try8 is just optimizing the try5-try7-chimera, since it scored a bit better than T0270.dv1.try6-opt2, so I'm going to start a try9 run just to polish Sol's model. It has a different N-terminus, so may represent a slightly different good solution. Fri Aug 27 18:57:30 PDT 2004 Kevin Karplus On looking at try8-opt2 and try9-opt2, I like the N-terminus of try9-opt2 better, since it doesn't block the binding pocket. I made a chimera (crossing over after T11), but it has some bad clashes. Repacking with Rosetta reduced the clashes, but caused the overall score to get slightly worse. I'm going to do a try10 polishing run (same cost fcn) starting from just the try9-try8-chimera and its repacked version. Fri Aug 27 19:40:07 PDT 2004 Kevin Karplus Now, I have to decide what to submit try10-opt2.repack-nonPC good N-terminus and loop, best Rosetta try8-opt2 bad N-terminus, good loop, best score try9-opt2 good N-terminus, less good loop try1-opt2 full-auto T0270-1t0tV-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template Foo---I just noticed that try9 does NOT have a closed loop---it has the same loop as try1, which it got from try6. I'll have to redo that optimization without the other models around to steal the show. Fri Aug 27 20:25:25 PDT 2004 Kevin Karplus I just noticed that T0270 is not due until Monday. I will have time to do somewhat longer optimizations, probably starting with try10 models, to try to get better packing. Sat Aug 28 07:43:21 PDT 2004 Kevin Karplus try12-opt2, optimized from try10 models, is the best scorer with the try12 costfcn and the older try11 costfcn, finally beating out try8-opt2. Rosetta likes best the try12-opt2.repack-nonPC model, also. The sidechains of the repacked model have better intrinsic values, but have slightly worse clashes and the wet and dry burials are a little worse. Perhaps should drop try8 or try11 from the "best models" set and submit try12-opt2 and try12-opt2.repack-nonPC. The differences between try12 and try11 are pretty small visually, and all in favor of try12, while try8 has a distinctly different N terminus. I don't think there is much point in trying further optimization now---we're about as close as we're going to get. I'll submit try12-opt2 best score try12-opt2.repack-nonPC best Rosettta energy try8-opt2 alternative N-terminus try1-opt2 full auto T0270-1t0tV-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template Mon Oct 4 12:39:52 PDT 2004 Kevin Karplus We are fairly close to the correct solution (1vdhA). Our best model in terms of smooth GDT is try2-opt1 (84.7449%) our best submitted is model3 84.0132% our model1 gets 83.9354% our full-auto gets 83.0010% These are slight improvements over the initial alignment (model5) 82.7212% There is no point in comparing with robetta, since the 1t0tV template was not in PDB when robetta was run. The best all-atom rmsd are try2-opt2.repack-nonPC 3.3973 try3-opt2.repack-nonPC 3.4065 try3-opt2 3.4510 try3-opt1 3.4523 try2-opt1 3.4538 try2-opt2 3.4668 ... model3 3.5819 Rosetta repacking seems to have helped, but scwrl did not. The best rmsd_ca are try2-opt1 2.9473 try3-opt1 2.9512 try2-opt2 2.9523 try3-opt2 2.9656 ... model3 2.9679