Wed Jun 28 10:17:34 PDT 2006 T0352 Make started Wed Jun 28 10:19:28 PDT 2006 Running on lopez.cse.ucsc.edu Wed Jun 28 17:46:07 PDT 2006 Kevin Karplus BLAST found nothing in PDB (top hit 1jkmB E-value 3.0) HMMs found nothing either (top hit 1jpaA Evalue 2.3) Top alignments all disagree, with each other and with try1-opt2. Only consensus is that the protein is mostly helical. Thu Jul 13 21:29:19 PDT 2006 Kevin Karplus I agreed to take on the new-fold prediction, so I'd better start on it. The multiple alignments are all very similar, so I see nothing to choose one another. I should probably redo the make, since there are some new neural nets now. Make started Thu Jul 13 21:36:20 PDT 2006 Running on cheep.cse.ucsc.edu There are some highly conserved residues: W24, P29, P39, F40, W50, W53, P70, P77 I suspect that the conserved aromatics are clustered (somehow) in a hydrophobic core. This is a very aromatic protein: five F, 3 W, and 2 Y, not to mention 6 H (not counting the HIS tag). There are a lot of aliphatics also. I think I might aim for a bundle on the protein---but I'm not sure exactly how. I'll also want to look at the server models, so I'm starting a scoring run for them (using unconstrained.costfcn and try1.costfcn) on the farm cluster. Make started Thu Jul 13 22:02:50 PDT 2006 Running on cheep.cse.ucsc.edu Sat Jul 15 09:44:42 PDT 2006 Kevin Karplus I've been chiding the students about not starting far enough ahead on the ab-initio models, and here I am starting at T0352, with nothing done on it and only days to the deadline. I looked at the n_sep, o_sep, n_notor2, and o_notor_2 predictions, to see if I could get cripser ends to the helices (as HBond constraints). It looks like the first helix should extend from T3 to Q22, with an extra Hbond from L18.O to H23.N There seem to be 2 sep3 Hbonds: Q30.O Q33.N and P31.O F34.N. The next helix is more weakly predicted, from E46 to I56. Then there may be a helix from R58 to K66 with an extra Hbond from L62.O to Q67.N Helix A76-T85, with extra Hbond L83.O D86.N helix P88-A105 with extra Hbond H87.O R90.N Sat Jul 15 10:32:51 PDT 2006 Kevin Karplus I made up a try2 costfcn with these constraints (also upping the packing terms dry6.5, dry8, and phobic_fit) and removing the cysteine and beta terms. I also added a few (perhaps too strong) distance constraints to try to get W50, W53, Y78, Y79 to pack close together. I'm starting try2 from the alignments on cheep. I'll do similar run but with only fragments (no long alignments) as try3. Sat Jul 15 10:38:38 PDT 2006 Kevin Karplus try3 started on lopez. try2 does a bit better thatn try1 (according to try1, try2, and unconstrained costfcns and rosetta), but try3 is not as successful. In try2, M1-D86 looks plausible, as does D86-D106, but I don't like how the helix continues straight through D86. I should probably also chop off the HIS tag and just predict through M1-S109 Incidentally, I now think that W50 and W53 should be packing with F40 and F41, rather than Y78 and Y79 (though I could change my mind). Sat Jul 15 12:11:09 PDT 2006 Kevin Karplus M1-S109 make started on cheep. Sat Jul 15 12:18:24 PDT 2006 Kevin Karplus try4 started on lopez, starting from alignments with somewhat different distance constraints to bring aromatics together. Sat Jul 15 13:14:12 PDT 2006 Kevin Karplus try4-opt2 scores best with try4 and unconstrained costfcn, but rosetta doesn't like it much. try4 actually looks fairly plausible to me, though it isn't the bundle I was expecting, and F40 and M45 are on the outside. Sat Jul 15 17:30:04 PDT 2006 Kevin Karplus I am singularly unimpressed with M1-S109/try1-opt2 The individual helices are fine, but they don't pack against each other. Sun Jul 16 11:50:17 PDT 2006 Kevin Karplus I modified the constraints for M1-S109/try2 based on what I saw in the o_notor2, n_notor2, o_sep, and n_sep predictions. It won't help much, but it may improve the search a little: # from T0352.dssp-ehl2.constraints (modified by hand) HelixConstraint T3 Q22 1 HelixConstraint P47 I56 1 HelixConstraint P57 N65 1 HelixConstraint A76 A84 1 HelixConstraint P88 F104 1 # from rr.constraints Constraint L11.CB V54.CB -10. 6.0 10.0 0.2 Constraint V54.CB V75.CB -10. 6.0 10.0 0.2 # predicted Hbonds Hbond L18.O H23.N 1 Hbond Q30.O Q33.N 1 Hbond P31.O F34.N 1 Hbond L62.O Q67.N 1 Hbond H87.O R90.N 1 Sun Jul 16 11:51:28 PDT 2006 Kevin Karplus M1-S109/try2 started on cheep Sun Jul 16 12:31:57 PDT 2006 Kevin Karplus The M1-S109/ try2 costfcn likes try1 better than try2, but rosetta likes try2 better. Both look pretty terrible to me. Maybe I need to increase phobic_fit and other packing terms. M1-S109/try3 started on lopez. Sun Jul 16 14:06:14 PDT 2006 Kevin Karplus M1-S109/try3 is no better and may be worse than the previous ones. Mon Jul 17 16:06:14 PDT 2006 Kevin Karplus I took M1-S109/try1-opt2 and proteinshopped it into a 5-helix bundle (M1-S109/decoys/shop2.pdb.gz) I then ran it through gromacs and rosetta, which reduced the very bad clashes by introducing breaks. Mon Jul 17 16:20:17 PDT 2006 Kevin Karplus Started M1-S109/try4 to optimize shop2.gromacs0.repack-nonPC.pdb I tried making some other bundles (shop3 and shop4), but they looked so unconvincing to me that I'm not even going to try optimizing them. Mon Jul 17 16:46:12 PDT 2006 Kevin Karplus M1-S109/try4 is often adding alignments in the initial optimization runs, so I'll do M1-S109/try5 with the same starting point, but with InsertAlignment turned off. Mon Jul 17 17:16:55 PDT 2006 Kevin Karplus try4 fixed the gaps, but threw away the bundle. The result has bad burial and is quite foamy. It looks like try5 is headed in a similar direction. I'll have to add some constraints to hold the bundle together at the open ends. Mon Jul 17 17:37:33 PDT 2006 Kevin Karplus M1-S106/try6 started with some pretty stiff constraints on CB atoms near the open ends of the bundle. Let's see if that helps. Mon Jul 17 18:41:24 PDT 2006 Kevin Karplus try6-opt1 has not blown up, but the stripe L51, L55, M59, L63 clearly needs to be buried. This may require more ProteinShopping. Is there a ProteinShop that runs under OS X? Mon Jul 17 19:39:24 PDT 2006 Kevin Karplus I tried downloading ProteinShop, but when I use StuffitExpander to open the zip file, it lists a lot of files that it is supposedly putting into the same directory, but they never appear. Mon Jul 17 19:50:51 PDT 2006 Kevin Karplus Since I can't seem to get ProteinShop for my Mac, I'll try turning the helix with constraints (usually pretty futile). M1-S106/try7 is similar to try6 (starting from the same shop2 models), but with constraints in the costfcn to try to pack L51, L55, M59, and L63 on the inside. (started on cheep) Mon Jul 17 19:57:41 PDT 2006 Kevin Karplus M1-S106/try8 is similar to try7 (same costfcn), but starting from all models (which means mainly try6-opt2). (Started on lopez) Mon Jul 17 23:00:00 PDT 2006 Kevin Karplus M1/S106/try8-opt2 is beginning to have the shape I want, but some of the constraint are probably wrong. I tried putting in a different set of constraints for try9, and optimizing again from all models. I'll probably have to do another run after that with weak constraints to let the gaps close. Tue Jul 18 10:07:08 PDT 2006 Kevin Karplus M1-S109/try9-opt2 has bad breaks as expected, but try10-opt2 does a fair job of closing them. There are two big remaining problems: R25 is buried, but there is no reason for it to be---it is on a loop that can easily by swung out into solution. helix P47-D64 is unwound at P57. Hmmm, given that this is a proline, maybe that's not so bad. In fact, we may have too many mid-helix prolines in this model. Rosetta still prefers the loosly packed models with fewer breaks, but try10-opts.gromacs0.repack-nonPC is the best of the 5-helix bundles for it. There *are* still bad breaks: T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz has 14 breaks T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)L69 with cost 2.59991 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)M59 with cost 2.54371 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)P68 with cost 2.4265 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)R58 with cost 1.94041 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)Q67 with cost 1.64926 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)E46 with cost 1.52463 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)P88 with cost 1.142 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)M45 with cost 0.71072 T0352.try10-opt2.gromacs0.repack-nonPC.pdb.gz breaks before (T0352)H23 with cost 0.65222 Conformation[10] T0352.try10-opt2.pdb.gz has 14 breaks T0352.try10-opt2.pdb.gz breaks before (T0352)L69 with cost 2.76406 T0352.try10-opt2.pdb.gz breaks before (T0352)P68 with cost 2.48652 T0352.try10-opt2.pdb.gz breaks before (T0352)Q67 with cost 1.81321 T0352.try10-opt2.pdb.gz breaks before (T0352)R58 with cost 1.80222 T0352.try10-opt2.pdb.gz breaks before (T0352)E46 with cost 1.2833 T0352.try10-opt2.pdb.gz breaks before (T0352)P88 with cost 1.18552 T0352.try10-opt2.pdb.gz breaks before (T0352)K66 with cost 0.840041 Since I need to do a submission today, I should probably stick on the HIS tag and try polishing the whole chain. Tue Jul 18 12:35:01 PDT 2006 Kevin Karplus try5-opt2 has the basic structure of M1-S106/try10-opt2, with the HIS tag appropriately stuck on the end. I made a chimera-5-3 that took try5-opt2 and replaced H21-R25 with the version from try3-opt2, to unbury R25. This introduced a bad gap, which I hope can be closed without reburying R25. Tue Jul 18 12:39:12 PDT 2006 Kevin Karplus started try6 on cheep to try to close the gap in chimera-5-3. Tue Jul 18 13:31:10 PDT 2006 Kevin Karplus Looking at the score-all.try6.pretty scores, I see that try4 actually ends up with a better phobic_fit score than try5-opt2, so I probably want to decrease the weight of that term, and increase the dry weights instead. It doesn't look like try6-opt1 has done anything very sensible with the loop around R25, so I'm not holding out much hope of improvement there. Tue Jul 18 14:19:20 PDT 2006 Kevin Karplus try6-opt has a big beak after R25. I'll try another polishing run with breaks turned way up, but I don't expect much from it. I'll probably have to ProteinShop the model a bit tomorrow. (try7 started on cheep) Tue Jul 18 15:10:24 PDT 2006 Kevin Karplus try7-opt1 closed the gaps, but messed up the bundle---helices are turned to have the hydrophobic face outward! Tue Jul 18 15:50:17 PDT 2006 Kevin Karplus Preliminary submission done with comment We did not get good fold-recognition hits for T0352, and undertaker seemed incapable of finding a compact packing of the helices, so we tried using ProteinShop to create a 5-helix bundle. The bundle had bad clashes and unrealistic loops, so we used undertaker to try to clean up the bundle. We have not yet got a bundle that we really like, and may not be able to by the deadline. Model 1 is try5-opt2, probably the best of our 5-helix bundles. Model 2 is try7-opt2, a 5-helix bundle with fewer breaks, but with a lot of hydrophobics exposed. Model 3 is try4-opt2, the best-scoring of our models before trying to use ProteinShop. Model 4 is try2-opt2.gromacs0.repack-nonPC, the model that Rosetta likes best of the ones it has repacked sidechains for. Model 5 is try1-opt2, the fully automatic model. -------------------------------------------------- I'll want to try ProteinShopping try7-opt2 tomorrow and seeing if I can bring the hydrophobics back into the center. Wed Jul 19 11:27:40 PDT 2006 Kevin Karplus Should I try to ProteinShop try7-opt2 (to get the hydrophobic buried again) or try5-opt2 (to fix the broken helix and try to make gaps close)? Wed Jul 19 12:36:17 PDT 2006 Kevin Karplus I tried tweaking try5-opt2 (as try5-tweak1, try5-tweak2, and try5-tweak3) with ProteinShop. I picked up some of the distance constraints from the try5-tweak2 model to use in try8.costfcn (along with helix constraints and some Hbond constraints from the sep alphabets). Wed Jul 19 15:52:16 PDT 2006 Kevin Karplus try8-opt2 made a mess of the central helix, not only splitting it but folding it back on itself. R25 is still pointing into the interior. Wed Jul 19 16:01:50 PDT 2006 Kevin Karplus I'll make the constraints much stronger in try9, and I'll start from just the try5-tweak3 models. try9 started on cheep. Wed Jul 19 17:08:17 PDT 2006 Kevin Karplus try9-opt2 has good helices, but P47-L62 is rotated approx 1/4 turn so that the hydrophobic stripe is on the outside. Wed Jul 19 21:27:11 PDT 2006 Kevin Karplus try10 buries the residues predicted to be buried, but also buries a cluster of ASP residues. This may not be so terrible, if they are coordinating an ion (there is also a HIS in the neighborhood), but is most likely wrong. I still think try10 is our best so, but I'll give it a polishing run to try to pack things tighter. Wed Jul 19 22:39:10 PDT 2006 Kevin Karplus try11-opt2 is the best scoring with the try11 costfcn, which has no constraints. Rosetta still prefers try2. I will submit ReadConformPDB T0352.try11-opt2.pdb ReadConformPDB T0352.try5-opt2.pdb ReadConformPDB T0352.try4-opt2.pdb ReadConformPDB T0352.try2-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0352.try1-opt2.pdb Wed Jul 19 22:48:19 PDT 2006 Kevin Karplus Submitted with comment: We did not get good fold-recognition hits for T0352, and undertaker seemed incapable of finding a compact packing of the helices, so we tried using ProteinShop to create a 5-helix bundle. The bundle had bad clashes and unrealistic loops, so we used undertaker and ProteinShop alternately to try to clean up the bundle. Model 1, try11-opt2, was heavily tweaked using ProteinShop to get the 5-helix bundle, then optimized with undertaker (mixing in some gromacs and rosetta sidechain packing to get out of local minima). There are some buried charges, which is not too encouraging, though most of the hydrophobics are also buried. The buried ASP residues are also near a HIS, so there may be a metal-binding site, though we don't have time to try to create one. Model 2, try5-opt2, is an earlier attempt at a 5-helix bundle, but one in which the middle helix broke. It broke at a proline, so this is not really so bad. Model 3, try4-opt2, is the best-scoring of our models before trying to use ProteinShop. Model 4, try2-opt2.gromacs0.repack-nonPC, isthe model that Rosetta likes best of the ones it has repacked sidechains for. Model 5, try1-opt2, is the fully automatic model. ------------------------------------------------------------