Wed Jul 7 08:13:57 PDT 2004 T0230 Due 5 Aug 2004 Wed Jul 7 11:14:44 PDT 2004 Kevin Karplus sam-t2k gets no strong hits, but t04 has a strong hit to c.47.1.10 (1qk8A), which isn't even in the running for t2k. The initial model (model 1 in t2k.undertaker-align) created from d.52.7.1 (1pa4A) looks reasonable, until you see the large gap between A69 and I70. That gap is probably fixable, but not easily. When try1 finishes, I'll remake all-align.a2m.gz and run a similar run for try2. Wed Jul 7 14:36:37 PDT 2004 Kevin Karplus try1 looks ok, but not great. There are some small breaks and the C-terminal helix is floating off away from the rest. Sun Jul 18 20:20:16 PDT 2004 Kevin Karplus Geroge asked me to look at what he has done on T0230, but he left no notes, so I've no idea what he was trying to do and what things were successful. I repacked try3, try4, and try5. and score them all with an unconstrained cost function. Try5-opt2 is the best scoring of the current models, but it is still rather "foamy"---not densely packed. The secondary structure prediction seems to have 3 clear helices and 3 clear strands. The structure in try5 seems to agree with the secondary structure prediction and have a proper right-handed crossover for the strand2-strand3 connection, but the final helix is not well packed against the sheet. Also the parallel connection between strand 2 and strand 3 is not making Hbonds. We probably want to pair N44-M49 with N78-L83, hbonding V47. There is a 3-10 helix predicted for P91-R93 that is not being formed---I'm not sure how best to encourage it---maybe an Hbond P91.O-M94.N? There is also a fainter indication of another antiparallel strand between the first helx and the strongly predicted strand at L33-D40. I'll add that for try6 as well. Running as try6 from the initial alignments. Sun Jul 18 21:34:33 PDT 2004 Kevin Karplus try6 ran very quickly---too short a run to get a good result. The I39-V45-V79 stripe across the beta sheet did not form. Let's increase the weight of the sheet constraints and of constraints in general, and run again, for 4 times longer. Mon Jul 19 09:02:37 PDT 2004 Kevin Karplus In try7, strand V45-M51 seems to be flipped over. Perhaps the sheet constraint is wrong? I want SheetConstraint V45 T52 V79 D86 Hbond E80 which is compatible with what I requested, and the echoed constraints in try7-opt2.constraints seem compatible. Maybe for try8 I should loosen the breaks a bit more to allow the strands to align better. Mon Jul 19 20:30:36 PDT 2004 ggshack To fill in the earlier work: I did try2 to examine the results of adding some RR contact constraints to the existing constraints while relaxing two of the sheet constraints. It suffered fromm a bad hbond L46 that should have been an hbond V45. Try3 was developed during a group meeting as an example of how to make changes to '*.under' to speed up the process of getting new results. It still had the bad hbond. Try4 corrected the hbond - or did it? Back to the latest work I am doing a TRY8 using some RR constraints and see what undertaker does with it. Tue Jul 20 08:54:58 PDT 2004 Kevin Karplus I looked at try8, but the V45-M51 strand is flipped over. George did a try9 (again he left no notes on what he was trying to achieve with it). Let's name the strands S0 D27 V29 (a bit dubious) S1 L33 I39 S2 V45 M51 S3 N78 T84 and the helices H1 K9 K17 H2 L58 K72 h3 P96 K101 In try9 strands S1^vS2 seem to have formed well, but S3 is not firmly attached. Some of the "rr" constraints seem to cluster well. For try10, I'll put in many of the rr constraints (those with p>0.5) and switch to "bonus_constraints". I'll raise the weights of the sheet constraints also. I'll still try to add the hairpin for s0 ^v s1, but with lower weight, since I'm not so sure of that. We'll start from the existing models (try9 is the best so far). After this run, we'll probably want to submit this to VAST, and see if there is a fold we should have found---it seems like a simple enough structure that it must be fairly common. Tue Jul 20 10:00:25 PDT 2004 ggshack As I look at the try8 results I am struck by the fact that we are assuming an exposed parallel sheet. I don't accept that; a small protein needs help to stay together - such exposure of a parallel only weakens it. Theses strands need rearranging. Tue Jul 20 13:08:55 PDT 2004 Kevin Karplus There is nothing inherently wrong with exposing a parallel sheet. Although many parallel sheets are covered on both sides by helices or another sheet, it is not an invariable rule. We also have the possibility that this puppy dimerizes, so that even an exposed hydrophobic surface is not necessarily a deal breaker. Tue Jul 20 13:30:38 PDT 2004 Kevin Karplus I looked at try10, and there are few hydrophobics on the surface---most of it is charged or at least polar, so this protein does NOT look like a dimer or that we are exposing something that should be buried. try10-opt2 scores best with an unconstrained cost function, and looks fairly good. Rosetta prefers try2-opt2.repack-nonPC, probably because of clashes. I'll try doing a polishing run on the existing models, with no constraints, but with break and clash penalties turned up. Tue Jul 20 13:50:31 PDT 2004 Kevin Karplus I've submitted try10-opt2 to vast: job id VS59947 password T0230try10. Tue Jul 20 15:11:03 PDT 2004 Kevin Karplus try11-opt2 does not look bad, but the parallel hbonds are still not forming. Perhaps we need to change the pairing, backing up L83 to pair with V47, instead of M49. The would put the P87 where the strand leaves the sheet (a good choice), but lengthens a few of the rr-predicted constraints. There are several good hits from the vast search: 1lqlC=1lqlA, 1n2fA, 1hh2P, 1fugA=1rg9A, 1qwiA, 1egaB=1egaA Tue Jul 20 17:35:47 PDT 2004 Kevin Karplus try12 does not score well and does not come close to getting V45 lined up with V81--it lines it up with A78, getting a strand flip as well as being way off. Perhaps the use of bonus_constraints from alignments was a bad idea now that I took out the rr constraints---let's try again with constraints, not bonus_constraints. One interesting thing that try12 does is to wind S3 up into a helix, which happens to put the mutual-info cluster R93, L98, R99, W89 in close proximity. Tue Jul 20 20:32:22 PDT 2004 Kevin Karplus try13 does quite well with the shifted sheet constraint. Despite the short run and lack of polishing, it does almost as well as some of the highly polished runs with the previous sheet constraint. There is an interesting thing done with s0, causing a twist in s1. I kind of like the effect and will encorage it in the try14. I'll add some sidechain constraints based on the top rr constraints and try polishing it up. Of course, the constraints may mess everything up, since they are probably not completely consistent. With try14.costfcn, the best models are try13 and try11. We'll probably want to submit the best model from each line of reasoning about the sheet structure. Wed Jul 21 08:26:34 PDT 2004 Kevin Karplus I changed the scaling of Hbonds in undertaker, so cost functions need to have weights increased. I fixed unconstrained.costfcn, and the best unconstrained model is try11-opt2. try14-opt2 (using a different sheet constraint) has a much lower rank (after try10, try12, and try9). It has some good features though, so I'm tempted to submit try11 as our first model, try14 as our second, and Rosetta's favorite try12-opt2.repack-nonPC as the third. try1 can be 4th and the best fold-recognition alignment model 5th. I don't think we're going to improve on this much more, so it is time to move on. Fri Aug 6 13:57:21 PDT 2004 Kevin Karplus It is too late for this information to be useful, but on the FORCASP web site (Date: July-21st/04), Alexey Murzin said The 1H-, 13C- and 15N-NMR assignments of the target have been released by the BMRB (http://www.bmrb.wisc.edu/data_library/data/bmr5976.str.html). These data could be of some use, in particular, for filtering "ab initio" predictions. The CSI-derived secondary structure is in a good agreement with the consensus prediction by the servers, but correlates poorly with the 3D-Jury selected fold. We submitted try11-opt2 try14-opt2 try12-opt2.repack-nonPC try1-opt2 T0230-1pa4A-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 Thu Nov 18 23:35:27 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 37.8622 (try11-opt1, the opt1 version of our best model) best submit 37.8299 (also model1) model1 37.8299 auto 31.7607 align 27.9824 robetta best 41.1893 (robetta model2) robetta1 22.1917