Mon Aug 16 17:24:22 PDT 2004 T0281 DUE Sept 2 2004 Mon Aug 16 20:37:43 PDT 2004 Kevin Karplus This looks like a fold-recognition or new-fold target. There is some agreement between sam-t02 and sam-t04 on 1gefA (c.52.1.18) and more weakly on 1peeA. I picked out the top several hits from each method and put them in MANUAL_TOP_HITS. I did a "make extra_alignments" and "make all-align.*" to get the alignments up to date. I think that I'll leave this protein to Sol to work on. Fri Aug 20 10:20:41 PDT 2004 Sol Katzman The t2k and t04 str2 predictions are fairly good for these strands: define s1 20-24 define s2 29-33 define s3 38-42 But try1 starts at the N-terminus with a few residues of strand (not predicted at all by t2k or t04) that form a sheet with s3: define s0 1-3 Furthermore, although there is obvious potential for a DG hairpin s2 ^v s3 at: R29-H33 P34 D35 G36 R37-P42, s2 is not formed in try1. Instead, in try1 we apparently have s0 ^v s3 ^v s1. Looking at the sheet constraints from the alignments, we have: s1 ^v s3 align 1,7,8,10 s1 ^v s2 align 5,9 s2 ^v s3 align 6 s1 ^v s2 ^v s3 align 3,4 For try2, leave out s0 for the moment and see if we can get the more straightforward: s1 ^v s2 ^v s3 as predicted, using the align3 constraints. The rasmol script pointed to by 'strands' defines the above strands. Fri Aug 20 19:38:54 PDT 2004 Sol Katzman We did get the s1 ^v s2 ^v s3 sheet in try2. There is a fairly bad break at S45-G46. And some exposed hydrophobics. Strand s3 is IVVVP. It might be nice to have the helix K50-D59 bury at least one side of that. I will use a constraint to put I56 L57 on top of s3. That will be try3. Sat Aug 21 16:00:08 PDT 2004 Sol Katzman We lost the s2 ^v s3 hbonding in try3. But the K50-D59 helix is somewhat closer to s3. I will try a constraint to get the E8-R15 helix aligned with the K50-D59 helix. Mon Aug 23 10:32:14 PDT 2004 Sol Katzman Rats. Did it again. In try4.under I had the "try3" pointers, so overwrote some of the try3 results. Fortunately, it did not overwrite the gzipped models, just wrote the plain pdb files and the analysis files (breaks,helices,etc). I will delete the overwritten files, fix try4.under and rerun try4. Tue Aug 24 14:12:59 PDT 2004 Sol Katzman Try4 does not look too bad. s3 became somewhat unbonded from s2, but helix K50-D59 elongated all the way to L70. At our weekly group meeting it was suggested to try and place one helix on either side of the sheet. I will do this by eliminating the constraint attempts to pull the helices together, and instead try to put I56,A60 next to the s2 and s3 residues Y31,V40 Tue Aug 24 15:48:59 PDT 2004 Sol Katzman Try5 split the N-terminal helices into two again, but did get helices on opposite sides of the sheet. It seems like time to start polishing. So eliminate TryAllAlign, and read in all previous models. Increase the cost of clashes and breaks a little. Tue Aug 24 20:29:17 PDT 2004 Sol Katzman Created an unconstrained costfcn. Using it try6 scores best. Rosetta also likes try6 best among its repacked models. I am not sure we can do much more with this one. Some hydrophobics are exposed, but the way they are distributed on both sides of the helices and both sides of the strands, it is not really possible to bury everything. Fri Aug 27 16:04:25 PDT 2004 Martina Koeva I was looking at the try6-opt2 model and as Sol mentioned above, it is quite hard to keep both all hydrophobics buried. There is one in particular that sticks out right now, namely W2 is quite exposed. I am going to try and introduce a distance constraint to vury W2 closer to the strand, predicted to be fully buried. The only problem (that I can see) with burying W2 in such a manner is that there is a danger of pulling in some polar, or charged residues (R6). Fri Aug 27 17:15:59 PDT 2004 Martina Koeva Try7-opt1 is finished and my distance constraint must have brought W2 too close to V40. Even though the tryptophan is a little more buried now, there is a break on the strand near residues 39-41. Sun Aug 29 10:50:16 PDT 2004 Kevin Karplus try6-opt2 looks fairly good, but it would be nice to bury more of W2. I'll try turning the rotamer around so that W2.CE3 fits against V10.CG1. Moving P5.CG closer to K13.CD may also help. I'll start a try8 run to do that, also tightening up on clashes and breaks. I should probably also use VAST to see where we are getting our template from. I'll submit try6-opt2 to VAST. vast job: VS61177 Password: T0281try6 Sun Aug 29 11:15:27 PDT 2004 Kevin Karplus VAST finds several hits. The ones with the smallest P-values are AL len SCORE P-value RMSD %ID in top hits? 1r9cA 26 4.8 0.0055 1.1 15.4 no 1ecsA 34 4.7 0.0088 1.1 17.6 no 1e9tA 14 3.5 0.0092 2.2 7.1 no The longest ones are 1hh1A 40 6.3 0.0449 1.2 17.5 yes 1ecsA 34 4.7 0.0088 1.1 17.6 no 1vj0C 34 5.1 0.0405 1.6 11.8 no (would be 1vj0A) 1bf5A 33 4.3 0.0361 6.0 24.2 no 1xan 32 5.2 0.0333 3.5 18.8 no I'm a little worried that the highest residue-ID (1bf5A) has such a large RMSD. I'll pick up the 1hh1A, 1r9cA, 1ecsA, and 1bf5A alignments, adding 1r9cA, 1ecsA, and 1bf5A to MANUAL_TOP_HITS. Maybe I should look at the high ID hits also: 1bf5A 33 4.3 0.0361 6.0 24.2 no 1bylA 26 4.4 0.0270 1.2 19.2 no 1xan 32 5.2 0.0333 3.5 18.8 no 1ecsA 34 4.7 0.0088 1.1 17.6 no 1hh1A 40 6.3 0.0449 1.2 17.5 yes 1r9cA 26 4.8 0.0055 1.1 15.4 no Sun Aug 29 12:01:45 PDT 2004 Kevin Karplus try8-opt2 does tuck W2 away and achieves a new best unconstrained cost. Rosetta also likes try8-opt2.repack-nonPC best. I wonder if I should do another run from alignments, using the sheets and helices from try8, but without the burial contraints added in try3, try4, and try5. The new alignments in all-align should let us detect the burial without having to force it so hard (I hope). The result will probably not be as good as try8, which has had a lot of careful polishing, but I'm curious to see what comes up. I'd also like to look at what the 1b5fA alignment is suggesting. I made show-vast-align.under to create T0281.vast-align.pdb. The 1bf5A is pretty useless, as it is in such small pieces that the conversion to undertaker AlignedFragments leaves only 83 atoms. All the other VAST alignment get essentially the same parts of the model: the helix P7-L17 and two or three of the strands near it. Sun Aug 29 12:32:03 PDT 2004 Kevin Karplus The try9 costfcn was set up with the try8-opt2 helices and sheets, and try9 started from alignments (SCWRLing the alignments). It currently scores try8-opt2 best, not surprisingly. Sun Aug 29 12:41:21 PDT 2004 Kevin Karplus I just noticed that we had never picked up the robetta models for this target. They do not score at all well with the unconstrained cost function, forming much less sheet than we are. I think Sol did a good job of finding a consistent fold. Incidentally, the new templates 1r9cA and 1ecsA seem to be the favored ones in the TryAllAlign for try9. try9-opt2 did not even get up to the score of try3-opt1 on the try9 costfcn. But Rosetta likes try9-opt2.repack-nonPC best! Maybe I'd better try polishing try8---what Rosetta dislikes about try8 is the Eintra score for R29. Ah---there is a CB-N self-bump self-bump: 1.50309 Ang (T0281)R29.N and (T0281)R29.CB threshold= 2.24843 cost= 0.910745 I'll have to increase the weight of soft_clashes to fix this problem. Sun Aug 29 14:54:28 PDT 2004 Kevin Karplus Although try10 cleared up a number of clashes, it still has 3 bad clashes: other-bump: 1.96663 Ang (T0281)V40.CG1 and (T0281)I56.CG2 threshold= 3.08864 cost= 0.933359 self-bump: 1.50303 Ang (T0281)R29.N and (T0281)R29.CB threshold= 2.24843 cost= 0.910766 other-bump: 2.19523 Ang (T0281)V40.CG2 and (T0281)A60.CB threshold= 3.0227 cost= 0.853273 repacking removes 2 and introduces others: self-bump: 1.50303 Ang (T0281)R29.N and (T0281)R29.CB threshold= 2.24843 cost= 0.910766 other-bump: 2.36795 Ang (T0281)T32.OG1 and (T0281)V39.CG2 threshold= 2.96581 cost= 0.740958 other-bump: 2.58588 Ang (T0281)L14.CD1 and (T0281)T32.CG2 threshold= 3.06655 cost= 0.640459 other-bump: 2.6405 Ang (T0281)P49.CD and (T0281)T52.OG1 threshold= 3.10354 cost= 0.620708 The nasty self-bump in R29 came in on try5, so we might need to get rid of it by doing something like try9 does so that the helix doesn't try to cover it. In try9-opt2, R29 almost forms a salt bridge R29.NH2-E21.OE2. Sun Aug 29 15:50:44 PDT 2004 Kevin Karplus Adding this as an Hbond constraint doesn't seem to help---the constraint cost is lower for the other models than for try9-opt2, though the distance 2.861 is only slightly less than what undertaker considers ideal (2.94146), and some of the better-scoring ones have much larger distance (like 6.963 for try9-opt2.repack-nonPC). [Wait a sec---undertaker is reporting a desired distance of 2.78216, which seems better. Ahh, I see my misunderstanding---I was reading the distance for acceptors that have hydrogens, but the acid residues are assumed to have already given up their hydrogens. But this makes the poor score for the Hbond even MORE inexplicable.] The CDA angle of 137.9 degrees is not far from the "ideal" 111.253+-18.3, but the DAC angle of 167.0 degrees is a long way from the 119.71 +- 17.86 "ideal". The try11-opt2 version with a distance of 3.63, CDA1=156.2 degrees, DAC=86.2 degrees scores much better, but I don't really believe it should. Maybe there is a bug in the Hbond code. Optimizing with try11 does create a new best with the try11 costfcn, and a new favorite for Rosetta (try11-opt2.repack-nonPC), but with the unconstrained costfcn, try11-opt2 does not even get up to the level of try6-opt1. Sun Aug 29 16:35:06 PDT 2004 Kevin Karplus There definitely seems to be a bug in the hbond code---undertaker sees a ludicrous hbond in try11-opt2: Hbond (T0281)R29.NH2 (T0281)Y31.OH # d= 3.47692 cost= -0.60146 (The distance is fine, but the angles are terrible.) It does not see R29.NH2-E21.OE2 at a distance of 3.633, though the angles should be much better---that's correct, as the cutoff is currently set at 3.6 Ang (perhaps I should loosen that). Even in try9-opt2, we see some strange scoring here: Hbond (T0281)R29.NE (T0281)Y31.OH # d= 2.90433 cost= -0.497161 Hbond (T0281)R29.NH2 (T0281)Y31.OH # d= 3.4764 cost= -0.60169 Hbond (T0281)E21.OE2 (T0281)R29.NH2 # d= 2.86319 cost= -0.538197 The bad angle is the CDA1 angle---perhaps it is not being computed?? Sun Aug 29 17:12:03 PDT 2004 Kevin Karplus The CDA1 angle is just within the tolerance of the detector, and the other geometric parameters are near ideal, so the really bad angle here does not kill the hbond, though one could argue that it should. (In a slightly different setting, where we wanted to tolerate some errors in positioning, we might want to consider R29.NH2-Y31.OH a pretty good H-bond.) The Hbond scoring is not behaving quite as I thought it should, but I can't find any real bugs. I might want to revisit the code when I have more time---particularly revisiting the independence assumption for the different parameters. (In this case, when CDA1 is unusually small, we need for the donor and acceptor atoms to be much closer than usual.) Sun Aug 29 17:24:13 PDT 2004 Kevin Karplus For try12, I'll polish try11 some more For try13, I'll polish everything (i.e. I'll polish try10). Sun Aug 29 18:30:17 PDT 2004 Kevin Karplus We now have two leading contenders: try12-opt2 try13-opt2 try13 still has the bad rotamer for R29, and has other clashes and breaks. T0281.try13-opt2.pdb.gz breaks before (T0281)D35 with cost 1.14711 T0281.try13-opt2.pdb.gz breaks before (T0281)P34 with cost 0.681228 other-bump: 1.9665 Ang (T0281)V40.CG1 and (T0281)I56.CG2 threshold= 3.08864 cost= 0.933387 self-bump: 1.50256 Ang (T0281)R29.N and (T0281)R29.CB threshold= 2.24843 cost= 0.910934 other-bump: 2.28516 Ang (T0281)V40.CG2 and (T0281)A60.CB threshold= 3.0227 cost= 0.813306 bond-length: 1.07642 Ang (T0281)T32.C and (T0281)H33.N threshold= 1.24746 cost= 0.58722 try12 forms a little extra sheet, but has a bulge in one strand. It does not match the predicted alpha values as well as try13, nor are the residues as comfortable in Ramachandran space (worse bys cost). There are fewer clashes or breaks: T0281.try12-opt2.pdb.gz breaks before (T0281)D35 with cost 0.787118 other-bump: 1.9455 Ang (T0281)P4.CD and (T0281)V41.O threshold= 2.83689 cost= 0.895977 other-bump: 2.37053 Ang (T0281)I38.CG2 and (T0281)L70.CD2 threshold= 3.09357 cost= 0.797553 Both conformations have some hydrophobics on the surface, but I like the way they form a tight flat patch on try12, looking like a dimerization interface. On the other hand, the try13 model does a better job of matching the rr constraints, as weak as they are. Mon Aug 30 15:00:43 PDT 2004 Kevin Karplus Both try12 and try13 have bulges, just in different places. Perhaps I should try repacking try13, but WITHOUT the "native" rotamers, so that Rosetta is forced to pick a better rotamer, even though it clashes horribly. Then undertaker can try reoptimizing the result to fix the backbone. Mon Aug 30 15:07:49 PDT 2004 Kevin Karplus I tried that as T0281.try13-opt2.force-repack.pdb (see Makefile for instructions), but this didn't help much since the backbone has an improper bond angle for N-CA-C at R29. If we add n_ca_c to the unconstrained costfcn, try13 looks really terrible. Let's reoptimize try13 with this unconstrained costfcn. Mon Aug 30 15:38:04 PDT 2004 Kevin Karplus try14-opt1 seems to have fixed the problem with R29, though maybe there could be a few more Hbonds in the sheet. Perhaps we should do the same optimization with the other family of models, since they also have non-ideal N-CA-C bond angles (most likely introduced by HealGap). Mon Aug 30 16:17:13 PDT 2004 Kevin Karplus try14-opt2 scores much better and has the R29 residue fixed, but try15-opt2 also scores better, though Rosetta still prefers the fewer clashes of try12-opt2. Mon Aug 30 16:45:15 PDT 2004 Kevin Karplus I made a try15-try14-chimera sticking V20-G27 from try14 into try15, to try to fix the bulgy sheet in try15. The positioning is not great, so there are some bad clashes, but a round of optimization should fix that. try16-opt2 gets almost as good a score as try15-opt2, but still has some clashes and breaks: other-bump: 2.09213 Ang (T0281)I38.CG2 and (T0281)L70.CD2 threshold= 3.09357 cost= 0.904331 other-bump: 1.96048 Ang (T0281)P4.CD and (T0281)V41.O threshold= 2.83689 cost= 0.891076 neighbor-bump: 1.50686 Ang (T0281)A24.O and (T0281)K25.N threshold= 2.0822 cost= 0.856351 other-bump: 1.93195 Ang (T0281)R22.O and (T0281)L30.O threshold= 2.47194 cost= 0.772098 neighbor-bump: 1.79923 Ang (T0281)F19.CA and (T0281)V20.N threshold= 2.2725 cost= 0.753685 neighbor-bump: 1.97414 Ang (T0281)F19.C and (T0281)V20.CA threshold= 2.26657 cost= 0.563431 other-bump: 2.42916 Ang (T0281)V20.CB and (T0281)T32.O threshold= 2.77262 cost= 0.547729 T0281.try16-opt2.pdb.gz breaks before (T0281)K25 with cost 1.93942 T0281.try16-opt2.pdb.gz breaks before (T0281)D35 with cost 0.788675 It may be possible to help with closure by insisting on an A24.N-H28.O hbond---possibly even an A24.O-H28.N. Ah--I see that the try16 costfcn was trying to make things difficult, by being off by one. I had SheetConstraint (T0281)F19 (T0281)M23 (T0281)P34 (T0281)L30 hbond (T0281)V20 3.0 when I meant SheetConstraint (T0281)F19 (T0281)M23 (T0281)H33 (T0281)R29 hbond (T0281)R22 3.0 Let's fix that and try again Mon Aug 30 18:08:07 PDT 2004 Kevin Karplus Although try17-opt2 score better overall, it didn't really fix the specific problems: other-bump: 2.03364 Ang (T0281)I38.CG2 and (T0281)L70.CD2 threshold= 3.09357 cost= 0.919297 other-bump: 1.96495 Ang (T0281)P4.CD and (T0281)V41.O threshold= 2.83689 cost= 0.889577 neighbor-bump: 1.5076 Ang (T0281)A24.O and (T0281)K25.N threshold= 2.0822 cost= 0.855926 other-bump: 1.92226 Ang (T0281)R22.O and (T0281)L30.O threshold= 2.47194 cost= 0.77887 neighbor-bump: 1.79936 Ang (T0281)F19.CA and (T0281)V20.N threshold= 2.2725 cost= 0.753571 neighbor-bump: 1.97349 Ang (T0281)F19.C and (T0281)V20.CA threshold= 2.26657 cost= 0.564297 T0281.try17-opt2.pdb.gz breaks before (T0281)K25 with cost 1.93158 T0281.try17-opt2.pdb.gz breaks before (T0281)D35 with cost 0.772037 Perhaps I should try again, with the same cost function, but reoptimizing from the more moveable try15-try14-chimera. Mon Aug 30 20:16:37 PDT 2004 Kevin Karplus try18 definitely does better than try16 and try17, but still falls behind try15, try14, and try12 on the unconstrained costfcn. There is still a break before D35 and there are some fairly bad bumps, but not as bad as earlier runs. other-bump: 2.05613 Ang (T0281)P4.CD and (T0281)V41.O threshold= 2.83689 cost= 0.855042 neighbor-bump: 1.79916 Ang (T0281)F19.CA and (T0281)V20.N threshold= 2.2725 cost= 0.753737 other-bump: 2.56694 Ang (T0281)R29.CB and (T0281)V40.CG2 threshold= 3.0227 cost= 0.624919 neighbor-bump: 1.97454 Ang (T0281)F19.C and (T0281)V20.CA threshold= 2.26657 cost= 0.562894 T0281.try18-opt2.pdb.gz breaks before (T0281)D35 with cost 0.808437 T0281.try18-opt2.pdb.gz breaks before (T0281)P34 with cost 0.409984 The worst mess seems to be some uncoiling of the helix at K50-T52. Perhaps I should do another cut-and-paste to fix this? I'm not sure what with though, since try12-opt2 already has the problem. Maybe I should do a polishing run with try18 (without tries 12, 14, and 15) using the unconstrained costfcn. Tue Aug 31 07:51:51 PDT 2004 Kevin Karplus try19-opt2 is a new best for unconstrained.costfcn, and Rosetta now likes try19-opt2.repack-nonPC best, but it has the bulge back (probably from crossover). If I really want to polish try18, I may have to eliminate all the other models, to avoid this sort of leap-frogging. Tue Aug 31 10:16:59 PDT 2004 Kevin Karplus try19-opt2 still scores best, but try20-opt2 without the bulge is almost as good. Best models with unconstrained try19-opt2 bulgy 4-strand try20-opt2 non-bulgy 4-strand try15-opt2 bulgy 4-strand try14-opt2 3-strand try12-opt2 bulgy 4-strand I think I should do one more polishing run with try14-opt2, then submit 3-strand (try21-opt2) non-bulgy 4strand try20-opt2 (or try20-opt2.repack-nonPC or both) bulgy 4-strand try19-opt2 full auto try1-opt2 Tue Aug 31 10:51:40 PDT 2004 Kevin Karplus Looking at the RR constraints, there is one other possibility we haven't looked at---putting M1-P4 parallel to E21-A24 or R22-K25 Constraint P4.CB A24.CB -10. 7.0 14.0 0.448 bonus Constraint P4.CB E21.CB -10. 7.0 14.0 0.386 bonus Constraint M3.CB A24.CB -10. 7.0 14.0 0.372 bonus Constraint P4.CB G27.CA -10. 7.0 14.0 0.345 bonus Constraint P4.CB M23.CB -10. 7.0 14.0 0.323 bonus Constraint W2.CB M23.CB -10. 7.0 14.0 0.3 bonus 1 MWMPp 21 eRMAK Let's try the sheet constraints of try21-opt2, plus SheetConstraint M1 P4 R22 K25 hbond E21 This will be try22, which might take a while to run, because it is SCWRLing all the alignments. One problem with this topology is that it makes an S-twisted crossover, instead of the more normal Z-twisted one. Inserting the strand to split the first hairpin might also be possible, but I'm not sure I want to try to set that up---we'd lose a lot of H-bonds. Tue Aug 31 11:15:16 PDT 2004 Kevin Karplus Oops, E21 not in R22-K25, so try22 is running without the Hbond constraints. Let's try again with hbond M23 as try23. Despite the lack of Hbond specification for the added sheet in the try22 costfcn, try22 scores bettter than try23 on the try23 costfcn. Let's polish them both with the unconstrained costfcn, and see if the result is good enough to include in the set of models we submit. Tue Aug 31 12:44:03 PDT 2004 Kevin Karplus Not only is try24 scoring "good enough", but it has the new best unconstrained cost! Its repacked version doesn't quite beat try20-opt2.repack-nonPC in Rosetta's view. So the choices are now 3.5-strand Scross try24-opt2 3-strand try21-opt2 non-bulgy 4strand try20-opt2 (or try20-opt2.repack-nonPC or both) bulgy 4-strand try19-opt2 full auto try1-opt2 I'll submit these. Thu Nov 18 23:58:37 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 51.1893 (also our model1) best submit 51.1893 (model1) model1 51.1893 auto 34.6127 align 30.1066 robetta best 50.4041 (robetta model10) robetta1 26.3188