Tue Jun 22 09:43:15 PDT 2004 T0210 DUE 14 July Tue Jun 22 12:51:38 PDT 2004 Kevin Karplus Looks like a fold-recognition hit on b.82.1.*. Tue Jun 22 18:07:59 PDT 2004 Kevin Karplus The try1 model looks pretty good---all the hits in the undertaker alignment seem to be for a similar beta-sandwich (b.82.1.*). The few hits for b.6.1.1 are partial-length matches. Tue Jun 22 23:59:39 PDT 2004 Kevin Karplus The t04 predictions of secondary structure seem worse than the t2k ones. I'm going to switch to t2k being the preferred alignment method. Hmm--I'm not so sure that is right. t04-str2 and t2k-dssp look like the good ones. For try3, I'll use just the t04-str and t2k-dssp constraints, plus any .mi constraints that look like they can be satisfied in one of the existing models. Wed Jun 23 07:07:16 PDT 2004 Kevin Karplus I don't particularly like try3. I really don't like the constraints for this protein, since there is so much ambiguity about the secondary structure. For try4, I'll take out the strandconstraint and helixconstraint commands, turn down the constraints based on mutual information, and use the t04 predictions of alpha angle rather than the t2k ones. Mon Jul 5 12:19:27 PDT 2004 Kevin Karplus try4 is looking pretty good, but not really much better than previous runs. I have to wonder about multimerization for this protein, as there is a large exposed hydrophobic patch. The template 1h1iB seems to have two copies of the domain joined by a linker. We should probably make a dimer by aligning with both domains. The dimer would allow us to see if we've accidentally tucked a helix into the dimerization interface, as I believe we have done in try4. We can pick up the sheet constraints from try4-opt2.sheets, since they are compatible with the other sheets we have had, but I think we'll need to add some constraints to tuck the helices out of the way. One way to get the dimerization alignment might be to submit the try1-opt2 model to VAST and see the different alignments it comes up with: VAST job VS59496 password T0210try1 Mon Jul 5 13:27:16 PDT 2004 Kevin Karplus I've saved some of the VAST alignments to 1j58A/1j58A-T0210-try1-vast.a2m 1fxzA/1fxzA-T0210-try1-vast.a2m They required some hand editing to get into a2m format, since the VAST output doesn't distinguish between . and - gap characters and puts both aligned and unaligned characters in the same vertical position. There is probably an algorithm for converting to a2m format, but I've just been doing it by hand. The 1j58A hit was our previous second highest hit (best in three-track t04 HMM), and 1fxzA didn't appear at all. I also saved the best-fit alignment from VAST: 1lr5A/1lr5A-T0210-try1-vast.a2m This was the best hit with the template library, but had not scored well enough to get to the best-hits list. I'll make extra_alignments to pick up the 1fxzA and 1lr5A alignments for our initial set. Mon Jul 5 14:07:01 PDT 2004 Kevin Karplus Both the dimerizations of try1-opt2 look pretty good. The helices on the back face could do a somewhat better job of covering the buried residues on the sheets. Mon Jul 5 17:35:02 PDT 2004 Kevin Karplus The try5-opt2 models do a bit better job of covering the back, but there is a small clash at the dimerization interface, involving 27-36 and 94-97. It might be fixed by demanding SheetConstraint M62 H66 V101 P97 Hbond W63 Sliding the first strand 2 later might also help Sheetconstraint Y22 Y26 L45 S41 Hbond Y26 Mon Jul 5 21:17:16 PDT 2004 Kevin Karplus try6 moved the N-terminal helix into the dimerization interface (probably trying to bury it). We'll need to pin the helix back out of the way---perhaps look at where it ended up in try1 and try5. Tue Jul 6 13:18 2004 Bret Barnes I'm adding constraints to put the N-terminal helix against the hydrophobic back side sheet. If that places the helix in the right spot (out of the interface) then I will create a dimer and use Rosetta to repack the dimer. Hopefully the repacked dimer will fix the clashes in the interface. If not then I might have to add some constraints to the interface and run another try, and then create another dimer and repack. Tue Jul 6 20:36 2004 Bret Barnes Try7 didn't do much with the helix, so in try8 I added two more constraints to force the helix out of the dimer interface and raised the weights of the constraints. Fri Jul 09 15:25 2004 Bret Barnes Try8 did not move the helix much. This time I am going to choose two new places to put the helix. Try10 will try to place it on the side and try11 will try to put on the back sheet just like we tried before. However this time I am raising these constraints to try to force the helix to move. Try10 is being ran on woof, and try 11 on meow. Hmm, for some reason I'm getting error with finding SCWRL temp files. I'll have to look into this... Sun Jul 11 19:36:14 PDT 2004 Kevin Karplus Scored all the current models with an unconstrained cost function. Currently, the best are try3, try4, try10, try11, try5, try6. try10 has a helix interfering with the dimerization interface. try11 has the helix tucked out of the way, but rather distorted in the process---it also has rather poor overall packing, looking a bit foamy. Dimer interface problems occur for models try2-opt2, try3-opt2, try4-opt2, try5-opt2, try6-opt2, try7-opt2, try8-opt2, and try10-opt2. (There does not seem to be a try9.) The best-looking ones seem to be try1-opt2, try11-opt2, and the first of the undertaker-align alignments: T0210-1h1iB-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5.a2m I'm going to do a "polishing run", starting with the existing models, but moving F7.CZ close to F54.CZ (say 7 Ang or closer). I'll turn off the other constraints, but turn the dry weights up a bit. George's rr constraints aren't of much use here---we have a good model for all the residues he generated constraints for. I'll also repack many of the "feasible" models before reoptimizing. Mon Jul 12 07:08:07 PDT 2004 Kevin Karplus Foo, it wasn't F7 that needed better packing, it was V151 and L154. F7.CZ-F54.CZ was already only 4.2 Ang in try11-opt2. If I had wanted to improve the ring stacking, I should have constrained F7.CZ to F54.CD2, which is at 6.4 down to 5 or less. Actually, now I'm not so sure that V151 and L154 do need to be packed in more---there is a cluster of charges: H53, H147, E133, D131, R23, R27 lining the "hole" in the protein. These probably bind a ligand, and so should not be buried. I might want to reduce the distance from H147.ND1 to D131.OD2 from 6.9 to 5. and to move H55.NE2 closer to H53.ND1 (say from 7.9 down to 6). These are tiny sidechain twiddles though and may not have a high enough probability of being correct to be worth trying. Try12 has a bad break before Q98, inherited from try11. I should increase the break penalty and the probability of using MoveGap, to move the break away from the proline into the more flexible loop, where it should be easier to close. The pairing Q146-F132 is pulling the final helix apart, so I'd like to go back to the packing of try1 with V150 and L154 near I106 and F7 near F54. We can keep H16 and H147 near H53. I've put these constraints into try13.costfcn, and tweaked the weight of the constraints until try1-opt2 just beats try12-opt2. Now I'll do a polishing run with the try13 score function. Mon Jul 12 09:47:48 PDT 2004 Kevin Karplus try13 looks pretty good as a dimer. I'll tweak the cost function a bit, turning up the "dry" weights, tweaking the constraints and turning the constraints down, to polish up the existing models. That may be it for this target. Mon Jul 12 13:40:40 PDT 2004 Kevin Karplus Except for the first helix, try14-opt2 looks pretty good. It scores the best not only with try14.costfcn, but also with unconstrained.costfcn. Rosetta prefers T0210.try11-opt2.repack-nonPC.pdb. I'll add more constraints for V51.CB-F10.CZ and I106.CD1-M1.CE, as well as strengthening the F7-F54 constraint. Mon Jul 12 15:53:26 PDT 2004 Kevin Karplus The try15.under file was not properly edited. Luckily nothing serious got stomped on, and I was able to rename the outputs correctly. The try15-opt2 is the best scorer, but Rosetta still prefers try11-opt2.repack-nonPC. Although we could probably get the helices to pack just a bit tighter, I think we've reached the point of diminishing returns on this target and am just going to submit.