Thu Jun 1 09:29:11 PDT 2006 T0311 Make started Thu Jun 1 09:29:34 PDT 2006 Running on orcas.cse.ucsc.edu Thu Jun 1 09:34:37 PDT 2006 Kevin Karplus No good hits with BLAST. Best is 1vbuA (41%id over 37 residues) E-value 0.1. We have non-overlapping blast hits though, so this may be assemblable from long fragments. Thu Jun 1 10:15:06 PDT 2006 Kevin Karplus This looks like a fairly easy fold-recognition target, with 1utxA (a.35.1.*) as possible template. Thu Jun 1 21:57:37 PDT 2006 Kevin Karplus residues D11-W74 are copiable from templates, which only leaves the two ends to be modeled. The N-terminal extension of the first helix looks ok, but the C-terminal helices after W74 look a bit dubious. That is, I believe that there are helices, but not with the orientations shown. This might be one for a little fixing up with Proteinshop. Fri Jun 2 11:44:24 PDT 2006 Firas Khatib Where do you think the helix should swing to? There is a very weak rr.constraint: 68 79 0 8 0.26194197458 which is L68 with A79, I could try swinging the helix that way. It does seem like the following residues might be a tad bit exposed: W67,L68,Q71 Maybe swinging the helix to cap that region off could work. We could just do a quick run adding in the L68 and A79 constraint, just to see where that places the C-terminal helices. I'll see what I can do with Proteinshop, since it's at the end, it shouldn't be too difficult. Fri Jun 2 13:11:36 PDT 2006 Firas Khatib I think fromTry1cTermRotAndCap.renum.pdb is decent, and I will run this as try2. Mon Jun 5 13:44:50 PDT 2006 Firas Khatib try2 scored a tiny bit better, but the C-terminal helix stick points up too much for my liking. I do like what it did with the N-terminus though, breaking the helix in two at the Proline 7. I'm going to try adding constraints to A79 to bury it down more and see what happens. I added these 2 contraints: Constraint A79.CB R8.CZ -10. 4.0 8.0 0.8 Constraint V92.CB L68.CB -10. 3.0 8.0 0.8 This will be try3. I will also submit this as try4 using try2 as input instead of the Protoshop model. Mon Jun 5 15:55:01 PDT 2006 Firas Khatib try3 is garbage because the constraint I gave was useless, the one I gave to try4 is correct: Constraint A79.CB R8.CZ -10. 1.0 4.0 0.8 Constraint V92.CB L68.CB -10. 3.0 8.0 0.8 and try4 did better than try2 by a tiny bit, but is essentially the same! Mon Jun 5 16:47:27 PDT 2006 Firas Khatib I tried Proteinshop inserting a kink in the helix at S75 I will run this as try5 using fromTry4breakatS75.renum.pdb as input I inserted these constraints: Constraint A79.CB R8.CZ -10. 1.0 4.0 0.9 Constraint V92.CB L68.CD1 -10. 1.0 4.0 0.9 Tue Jun 6 14:23:20 PDT 2006 Firas Khatib Well try5 scores best, 3 points better than try4-opt2, and I think it's pretty good. The termini really seem to pack better than try1 where we started. The only thing to note was that I did introduce a kink in helix 63-83 at S75. We might want to submit different models, some with the kink and some without it. Fri Jun 9 12:38:08 PDT 2006 Firas Khatib try5-opt2 also scores better in score-all.unconstrained, but not by much Mon Jun 19 14:31:32 PDT 2006 Firas Khatib I will try to get a model with a straight helix at S75, instead of the kink (starting with try3) turning up phobic_fit, and all the dry terms. Mon Jun 19 19:01:43 PDT 2006 Firas Khatib I am reading in all the try3 files (which had the straight helix) and yet using constraints to bring the terminus down. this will be try6 (on shaw) Tue Jun 20 15:16:54 PDT 2006 Firas Khatib try6 did well, especially compared to try3 that it started from! Tue Jun 20 15:47:16 PDT 2006 Firas Khatib I have added more constraints to bring the C-terminal end down more, and I took out the constraint that would cause the S75 helix to break: Constraint V92.CG2 L68.CD1 -10. 3.2 8.0 100 Constraint P50.CG S95.CB -10. 4.6 10.0 70 this is try7, taking all the try3s & try6s as input. If that doesn't work (it's running on whidbey) I will swing it around with Proteinshop. Tue Jun 20 16:32:34 PDT 2006 Firas Khatib well, it worked with the constraints for try7 BY INSERTING A GAP!!! Even though break was set to 50! So I will try inputting the Proteinshopped model: helixDownFromTry6.renum.pdb and see if undertaker can deal with it. I will submit try8 with just helixDownFromTry6.renum.pdb as input I will submit try9 with it as well as the previous contraints from try7 (both running on orcas) Tue Jun 20 17:43:38 PDT 2006 Firas Khatib I think this target is ready for the soft deadline: # put believed-to-be-best decoy first ReadConformPDB T0311.try8-opt2.pdb ReadConformPDB T0311.try5-opt2.pdb ReadConformPDB T0311.try9-opt2.pdb ReadConformPDB T0311.try6-opt2.pdb ReadConformPDB T0311.try4-opt2.pdb try5 could be model 1, it depends on the kinked helix. Tue Jun 20 19:20:12 PDT 2006 Kevin Karplus OK, I can accept that ordering for a monomeric T0311. We do have time to explore a multimeric design before final submission. We should also look at the server models (particularly ROBETTA and Pmodeller6). Tue Jun 20 19:31:32 PDT 2006 Kevin Karplus Preliminary submission done. Mon Jul 3 13:08:42 PDT 2006 Kevin Karplus We should make a dimer based on 1utx[AB] and optimize in dimer context. Tue Jul 4 11:53:27 PDT 2006 Firas Khatib I made: casp7/T0311/dimer/1utxA but with Kevin we decided that the dimer interface was not on the knuckle so we are looking at the other alignments to see if they are dimers in a better spot. Tue Jul 4 18:12:14 PDT 2006 Firas Khatib 2b5a (model2 in T0311.undertaker-align.pdb.gz) is a dimer and might work. 1y9q (model3) is not a dimer so that one is out. 2a6c (model4) is a dimer, but I don't think it will work since the interface is not on the knuckle either. 1b0nA (model5) is not a dimer, it seems like it just has a ligand as the second chain. so it looks like 2b5a is the only option at this point. I will try making a dimer using 2b5aA and see how it looks. Wed Jul 5 07:03:16 PDT 2006 Kevin Karplus What about the 1lmb[34] dimer (does our protein look like a DNA binder?) Or 1y7y[AB]? Or 2bnm[AB]? Wed Jul 5 07:45:24 PDT 2006 Kevin Karplus T0311 does *not* look like a DNA-binding protein----there are lots of negative charges, and DNA binders need positive charges to bind the phosphate backbone. All the conserved residues are hydrophobics inside the knuckle. The burial of the conserved W67 makes it unlikely that the P62-T84 helix swings out away from the knuckle. I'm beginning to think that we should just treat this as a monomer. I'll try doing a polishing run on the existing models. Wed Jul 5 07:55:58 PDT 2006 Kevin Karplus try10 started as simple polishing run on cheep. Wed Jul 5 08:35:30 PDT 2006 Kevin Karplus try10 completed and has a new best score. Also rosetta likes best decoys/T0311.try10-opt2.gromacs0.repack-nonPC.pdb.gz I'll try a polishing run from just the gromacs models. Wed Jul 5 08:44:17 PDT 2006 Firas Khatib I think you might be right about the monomer, Kevin. dimer-2b5aA.pdb doesn't look good at all, anyway. Wed Jul 5 10:46:33 PDT 2006 Kevin Karplus try11-opt2 does not seem to score quite as well as try10-opt2 with the try10.costfcn, but it is very close and clashes and breaks are smaller. Rosetta likes try11-opt2.gromacs0.repack-nonPC best so far. Upping the break and clash weights moves try11-opt2 to the top. I'll do a polishing run with high breaks and clashes starting from all models. Wed Jul 5 11:54:05 PDT 2006 Kevin Karplus try12-opt2 improves slightly on try11-opt2, without apparent tradeoffs. Rosetta's favorite is now decoys/T0311.try12-opt2.gromacs0.repack-nonPC.pdb.gz though undertaker does not like it much, seeing worse bond-angles for CA, worse peptide planes, worse sidechains, and only a small improvement in clashes and breaks (though all breaks *are* closed). There are only 88 Hbonds, instead of 100, and the pred_alpha values are worse. Wed Jul 5 12:04:59 PDT 2006 Kevin Karplus Unless someone has an idea for a different way to handle the non-knuckle helices, I'm going to submit ReadConformPDB T0311.try12-opt2.pdb ReadConformPDB T0311.try12-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0311.try5-opt2.pdb ReadConformPDB T0311.try9-opt2.pdb ReadConformPDB T0311.try6-opt2.pdb Wed Jul 5 12:09:45 PDT 2006 Kevin Karplus so submitted. Mon Oct 9 11:54:30 PDT 2006 Kevin Karplus Our best model seems to be try6-opt2, submitted as model5. It is among the best server models (but Pmodeller6_TS1 beats it). The knuckle fits quite well, but we made the C-terminus more ordered than the real one. Thu Mar 22 09:14:19 PDT 2007 Kevin Karplus Redoing the evaluation with fixed GDT and new real_cost function, our try10-opt1 is our best model, with try10-opt2 and try6-opt2 close behind. Zhang-Server_TS4 and FOLDpro_TS1-scwrl beat us. The SAM_T06_server is 3rd among the servers, but our model1 beats it.