Mon Jul 10 09:33:03 PDT 2006 T0367 Make started Mon Jul 10 09:34:26 PDT 2006 Running on lopez.cse.ucsc.edu Mon Jul 10 10:42:49 PDT 2006 Kevin Karplus BLAST does not get good hits in PDB (best is 1bcfA 39% over 51 residues, E-value 1.6), but the HMMs seem to be agreeing on SCOP family a.24.16.3, with pretty strong E-values (still waiting for the template-library search to finish). Fri Jul 21 15:22:25 PDT 2006 Firas Khatib I like the 3-helix bundle in try1, but I would like the burial of T59 & I63 to be better. I want to rerun try1 using str2's constraints, maybe it will help the helices in regions: 74-87 and 55-70 try2 is running on squawk using T0367.t06.str2.constraints and the rr constraints Mon Jul 24 20:25:41 PDT 2006 Kevin Karplus Pretty strong HMM hits to 1wolA, 1ufbA, 1o3uA but not to other templates. We don't want to muck with the model *too* much. I think that I63 is fine where it is, but T59 could be a bit more buried. The protein is slightly foamy, but otherwise ok. I'd just up the dry terms and phobic fit a little and reoptimize. There is no point to using the rr constraints on a fold-recognition hit, as they are more likely to move things apart than pack them better. Tue Jul 25 00:59:14 PDT 2006 Firas Khatib I just launched try3 which uses try2 as input with the dry terms and phobic fit increased from: SetCost wet6.5 15 near_backbone 5 way_back 5 dry5 15 dry6.5 20 dry8 15 dry12 5 \ phobic_fit 2 \ to: SetCost wet6.5 15 near_backbone 5 way_back 5 dry5 20 dry6.5 25 dry8 20 dry12 6 \ phobic_fit 4 \ try3 is running on shaw, and so is try4 which is the same, but using try1 as input try5 has the same costfcn, but uses the alignments instead of ReadConformPDB it is running on lopez Tue Jul 25 13:34:26 PDT 2006 Firas Khatib they all look pretty decent except try5 Sat Jul 29 12:15:29 PDT 2006 Firas Khatib all the models seem pretty consistent, I think only polishing runs are necessary at this point, unless we want to try a run based off of 1na3A (model 4 in T0367.undertaker-align.pdb), but as Kevin said: the strong HMMs hits were to the other templates so it is probably not worth messing it up! Sat Jul 29 12:56:56 PDT 2006 Firas Khatib The best scoring servers are SAM_T06 and Robetta4 and Robetta3, but they all score worse than all our trys. I will see how different the 2 robetta models are to ours. Their main difference with our models seems to be from F84-I94. Robetta3 uses 1wolA as a parent and Robetta4 uses 1o3uA Sat Jul 29 15:44:35 PDT 2006 Martin Madera Looked at 1wolA, 1ufbA, 1o3uA. They all look quite similar, but there are differences. The three helices that form the bundle are the same length in 1wolA & 1ufbA, but in 1o3uA the third helix is much longer... this could be because in the first two structures the third helix is bent, but in the third structure it's straight. The most variable region is the connection between the second and third helices (which includes the bend). In 1o3uA the most variable part of that region isn't shown, presumably because it's too disordered, but even the two short helices and the loops around them clearly vary somewhat between the three structures. Try1-try4 look similar to me, and look quite different from the templates. I don't like the short helix it's trying to create around S53-G58, the loop is straight in all three templates. I think try5 is doing the right thing there (trying to extend the short helix). And the extra short helix L86-I94 undertaker created just before the third long helix is wrong; it's probably based on the bend in 1wolA & 1ufbA but looks quite different. So my impression is that undertaker is deviating too far from the templates. I'll try to come up with some constraints. Overall, try5 seems to be even further from the templates than try1-try4. I can see how it follows the str2 predictions, but I think they're wrong. Sat Jul 29 16:51:05 PDT 2006 Martin Madera Some constraints. First, to destabilize the short helix at 54-58. The current distances are: Distance LYS54A.CA-GLY58A.CA: 4.644 ... try1 Distance LYS54A.CA-GLY58A.CA: 4.598 ... try2 Distance LYS54A.CA-GLY58A.CA: 4.640 ... try3 Distance GLY58A.CA-LYS54A.CA: 4.594 ... try4 Distance LYS54A.CA-GLY58A.CA: 6.371 ... try5 So I'll try: # to destabilize a short helix at 54-58, which Martin thinks is wrong Constraint K54.CA G58.CA 5.5 6.0 20 1 Next, I'd like to see the short helix extended. From tryX-opt2.helices: HelixConstraint (T0367)L62 (T0367)R67 ... try1-try4 HelixConstraint (T0367)T55 (T0367)L70 ... try5 I think try5 is excessive, but I'd like to see an extra turn compared to try1-try4, so: # extend the first short helix in the variable region (usually starts at L62) HelixConstraint T59 R67 1 Now the difficult part, which is the bend at the start of the third helix. I'll base this on 1ufb and and 1wol, which are the top two templates, and try4-opt2. The first question is, what's the right alignment? The good news is that all three structures have a conserved Y27 and the loop up to residue 22. The residue packing against the Y27 is: T0367 K100 1wol E100 1ufb R106 Now constraints. First 1wol: Distance ASP102A.CA-TYR98A.CA: 6.458 Distance GLU100A.CA-TYR98A.CA: 7.068 Distance LYS22A.CA-TYR98A.CA: 10.858 Distance TYR27A.CA-TYR98A.CA: 5.837 Distance TYR95A.CA-TYR98A.CA: 5.596 Distance TYR95A.CA-TYR27A.CA: 7.220 Distance TYR95A.CA-ASP102A.CA: 11.302 The corresponding distances for 1ufb are: Distance GLU108A.CA-TYR104A.CA: 6.607 Distance ARG106A.CA-TYR104A.CA: 7.086 Distance LEU22A.CA-TYR104A.CA: 11.131 Distance TYR27A.CA-TYR104A.CA: 5.589 Distance ALA101A.CA-TYR104A.CA: 6.157 Distance ALA101A.CA-TYR27A.CA: 7.306 Distance ALA101A.CA-GLU108A.CA: 12.351 The numbering for T0367 and 1wol should be exactly the same, but obviously the amino-acids will be different. So: # try to do a bend like in 1wol and 1ufb Constraint L102.CA V98.CA 6.0 6.5 7.0 1 Constraint K100.CA V98.CA 6.6 7.1 7.6 1 Constraint F22.CA V98.CA 10.0 11.0 12.0 1 Constraint Y27.CA V98.CA 5.0 5.7 6.4 1 Constraint Y95.CA V98.CA 5.3 5.9 6.5 1 Constraint Y95.CA Y27.CA 6.7 7.2 7.7 1 Constraint Y95.CA L102.CA 10.8 11.8 12.8 1 Just out of interest, the distances in try4 are: Constraint L102.CA V98.CA 6.0 6.5 7.0 1 ... 6.082 Constraint K100.CA V98.CA 6.6 7.1 7.6 1 ... 5.654 Constraint F22.CA V98.CA 10.0 11.0 12.0 1 ... 14.363 Constraint Y27.CA V98.CA 5.0 5.7 6.4 1 ... 11.492 Constraint Y95.CA V98.CA 5.3 5.9 6.5 1 ... 6.217 Constraint Y95.CA Y27.CA 6.7 7.2 7.7 1 ... 10.107 Constraint Y95.CA L102.CA 10.8 11.8 12.8 1 ... 9.917 So we'll see. Try6 and try7 running on lopez. These are from alignments, with the above constraints but otherwise using the try1 cost function. Sat Jul 29 20:02:06 PDT 2006 Martin Madera Try6 and try7 have finished. I like try7, but it scores worse than try1. I don't like try6 as much, but it scores better than try1. Need to check how well they satisfy the constraints, think about it a bit more, and start some polishing runs. Sun Jul 30 15:16:31 PDT 2006 Martin Madera Checking the constraints. The short helix got unwound in both cases, so let's check the bend. In try6-opt2: Constraint L102.CA V98.CA 6.0 6.5 7.0 1 ... 6.415 Constraint K100.CA V98.CA 6.6 7.1 7.6 1 ... 7.103 Constraint F22.CA V98.CA 10.0 11.0 12.0 1 ... 11.075 Constraint Y27.CA V98.CA 5.0 5.7 6.4 1 ... 5.674 Constraint Y95.CA V98.CA 5.3 5.9 6.5 1 ... 5.960 Constraint Y95.CA Y27.CA 6.7 7.2 7.7 1 ... 7.197 Constraint Y95.CA L102.CA 10.8 11.8 12.8 1 ... 12.059 ... perfect. Try7-opt2: Constraint L102.CA V98.CA 6.0 6.5 7.0 1 ... 6.612 Constraint K100.CA V98.CA 6.6 7.1 7.6 1 ... 7.086 Constraint F22.CA V98.CA 10.0 11.0 12.0 1 ... 11.189 Constraint Y27.CA V98.CA 5.0 5.7 6.4 1 ... 6.065 Constraint Y95.CA V98.CA 5.3 5.9 6.5 1 ... 6.155 Constraint Y95.CA Y27.CA 6.7 7.2 7.7 1 ... 7.201 Constraint Y95.CA L102.CA 10.8 11.8 12.8 1 ... 12.471 ... OK but slightly worse than try6. I've briefly looked at the conserved residues in 1ufb and 1wol, but I don't get the active site. I can see His56 (more on that later) and I can see the arginine, but I don't get the connection -- the relative positions are very different in 1ufb and 1wol. Unless I'm looking at the wrong arginine in 1wol -- but I can't find any other residue that would fit. Incidentally 1wol is strange, look at it in spacefill!!! However, I've noticed that His56 is conveniently conserved in both 1ufb and 1wol and in T0367. I think I'll pick the distance constraints from it: 1ufb: Distance TRP54A.CA-HIS56A.CA: 6.935 Distance TRP54A.CA-SER57A.CA: 8.446 Distance TRP54A.CA-ILE58A.CA: 9.770 Distance GLY55A.CA-SER57A.CA: 5.930 Distance GLY55A.CA-ILE58A.CA: 8.337 Distance HIS56A.CA-ILE58A.CA: 6.184 1wol: Distance ARG54A.CA-HIS56A.CA: 6.828 Distance ARG54A.CA-SER57A.CA: 8.418 Distance ARG54A.CA-ILE58A.CA: 9.603 Distance GLY55A.CA-SER57A.CA: 5.696 Distance GLY55A.CA-ILE58A.CA: 8.109 Distance HIS56A.CA-ILE58A.CA: 5.990 which means the following for T0367: # the turn around His56, taken from 1wol and 1ufb Constraint K54.CA H56.CA 6.67 6.87 7.07 1 Constraint K54.CA R57.CA 8.23 8.43 8.63 1 Constraint K54.CA G58.CA 9.48 9.68 9.88 1 Constraint T55.CA R57.CA 5.41 5.81 6.21 1 Constraint T55.CA G58.CA 8.01 8.21 8.41 1 Constraint H56.CA G58.CA 5.89 6.09 6.29 1 Just out of interest, the distances in try6 are: Constraint K54.CA H56.CA 6.67 6.87 7.07 1 ... 7.068 Constraint K54.CA R57.CA 8.23 8.43 8.63 1 ... 8.477 Constraint K54.CA G58.CA 9.48 9.68 9.88 1 ... 6.397 !!! Constraint T55.CA R57.CA 5.41 5.81 6.21 1 ... 5.460 Constraint T55.CA G58.CA 8.01 8.21 8.41 1 ... 5.144 !!! Constraint H56.CA G58.CA 5.89 6.09 6.29 1 ... 5.591 !!! ... basically G58 is in the wrong place, much as I had thought. For try7, Constraint K54.CA H56.CA 6.67 6.87 7.07 1 ... 6.932 Constraint K54.CA R57.CA 8.23 8.43 8.63 1 ... 8.223 Constraint K54.CA G58.CA 9.48 9.68 9.88 1 ... 10.016 ! Constraint T55.CA R57.CA 5.41 5.81 6.21 1 ... 5.857 Constraint T55.CA G58.CA 8.01 8.21 8.41 1 ... 8.685 ! Constraint H56.CA G58.CA 5.89 6.09 6.29 1 ... 6.514 ! again the problem is with G58, but this time it looks much better. This agrees with my intuitive feeling that try7 looks better than try6. For the bend, I think I need to extend the constraints. First 1ufb: Distance PRO99A.CA-ALA101A.CA: 5.572 Distance PRO99A.CA-TYR104A.CA: 10.558 Distance PRO99A.CA-GLU108A.CA: 17.071 Distance PRO99A.CA-LEU22A.CA: 18.322 Distance PRO99A.CA-TYR27A.CA: 12.862 and now 1wol: Distance ALA93A.CA-TYR95A.CA: 5.731 Distance ALA93A.CA-TYR98A.CA: 10.178 Distance ALA93A.CA-ASP102A.CA: 16.192 Distance ALA93A.CA-LYS22A.CA: 17.896 Distance ALA93A.CA-TYR27A.CA: 12.944 which gives: Constraint 93.CA 95.CA 5.35 5.65 5.95 1 Constraint 93.CA 98.CA 9.86 10.36 10.86 1 Constraint 93.CA 102.CA 15.30 16.65 18.00 1 Constraint 93.CA 22.CA 17.30 18.10 18.90 1 Constraint 93.CA 27.CA 12.60 12.90 13.20 1 Out of interest, for try7: Constraint 93.CA 95.CA 5.35 5.65 5.95 1 ... 6.513 !! Constraint 93.CA 98.CA 9.86 10.36 10.86 1 ... 11.779 !!! Constraint 93.CA 102.CA 15.30 16.65 18.00 1 ... 17.864 Constraint 93.CA 22.CA 17.30 18.10 18.90 1 ... 18.009 Constraint 93.CA 27.CA 12.60 12.90 13.20 1 ... 13.068 Hmmm, the last three constraints are pretty useless because they're so vague. I think I'll get rid of them. For try6: Constraint 93.CA 95.CA 5.35 5.65 5.95 1 ... 7.530 !!! Constraint 93.CA 98.CA 9.86 10.36 10.86 1 ... 12.281 !!! I am going to start three more runs from alignments using the modified constraints; these will be try8, try9 and try10. Running on pyro, camano and abyss. Sun Jul 30 19:09:46 PDT 2006 Martin Madera Try10 made the second helix in the variable region really long. I think it's almost certainly wrong, but may score quite well. Try9 looks OK, but it has a break in the chain. I think it's because undertaker made 89-92 into a helix, so it didn't have enough rope and/or flexibility to keep the constraints on the bend. Try8 looks similar to try9, again that short helix at 89-91 and break. Time to start some polishing runs. I'll use break 100 and soft_clashes 40, and SetCost wet6.5 15 near_backbone 5 way_back 5 dry5 30 dry6.5 30 dry8 20 dry12 5 \ phobic_fit 3 \ The runs are as follows: try11: try3, try4 ... camano try12: try6, try7 ... lopez try13: try8, try9 ... lopez Sun Jul 30 20:20:16 PDT 2006 Martin Madera I'll start one more run on try8 & try9, keeping the same cost function as for try8 and try9 but adding an extra constraint to destabilize the short helix at 89-91 with: # destabilize the short helix at 89-91 Constraint E89.CA D91.CA 5.8 6.0 20 3 The distance in try8 is 5.363, in try9 it's 5.356. Running as try14 on whidbey. Sun Jul 30 21:39:43 PDT 2006 Martin Madera Try11-try13 have finished, still waiting for try14. ----------------------------------------------------------------------- I'm about to put together best-models and write a summary for the models. ----------------------------------------------------------------------- Try14 just finished, it didn't manage to unwind the helix. Oh well. Sun Jul 30 21:59:12 PDT 2006 Martin Madera Method: We concentrated on modeling the variable region 50-98 based on 1wol and 1ufb. We're aware that 1o3uA is also a potential template and that it looks quite different from 1wol and 1ufb, but we have not had time to build a model based on it. The models we're submitting are: - Model 1: try12-opt2, our best-scoring model. This is a re-optimized version of try6 and try7. In try6 and try7 we tried to correct certain features of our automated runs (e.g. try2, submitted as Model 5 below) that we thought were wrong: a short helix around the conserved His56 and the region immediately preceding the final helix. - Model 2: try13-opt2, a well-scoring model closest to the two templates. It is a re-optimized version of try8 and try9, which attempted to keep the regions around His56 and just before the final helix even closer to the templates than try6/try7 mentioned above. - Model 3: try4-opt2. Our second-best model scorewise. This was an attempt to improve the packing of our automated runs (try1 and try2). - Model 4: try6-opt2.gromacs0.repack-nonPC.pdb. Rosetta favourite. - Model 5: try2-opt2. An automated run using our favourite secondary structure neural network, str2 (instead of consensus predictions). Sun Jul 30 22:24:32 PDT 2006 Martin Madera Done with this target. Sun Jul 30 23:02:13 PDT 2006 Kevin Karplus I edited Makefile to set up the MANUAL_TOP_HITS (needed for the PARENT record in the submission). I did "make T0367.method" then added Martin's comments to the end of the method file, did "make casp_models" and "make email". The models are submitted as described by Martin. Fri Aug 11 03:22:56 PDT 2006 Kevin Karplus T0367 was selected as a refinement target. tr367's score is ok, but can be improved slightly by running it through gromacs and repacking the sidechains. It has loop differences at E89-V98 and R51-T59, both regions where we had a high degree of variability. Fri Aug 11 06:45:07 PDT 2006 Kevin Karplus Our top hit (1wolA) is listed by PQS as a dimer, so is second hit 1ufbA. Should we be doing optimization in a dimer context? We didn't for the models we submitted, but we probably should have. We have a bit more time now, so we might as well do the refinement in a dimer context.