Thu Jul 8 06:40:31 PDT 2004 T0232 Due 9 Aug 2004 Thu Jul 8 11:08:52 PDT 2004 Kevin Karplus This looks like a comparative model for 1jlvA and other a.45.1.1 + c.47.1.5 templates. Thu Jul 8 18:35:56 PDT 2004 Kevin Karplus Because the t2k alignment does not identify any key residues, and the t04 one does, I prefer the T04 alignment for this target. The T2k alignment has only 3 sequences, all essentially the same, while the t04 alignment has 2533 sequences, retaining 169 of them when thinned to 40% id. The key residues of the t04 alignment cluster well in the try1-opt1 model, but that model seems to have been pried apart a bit, exposing some of the buried interior. We may need to increase the dry weights for the try2 run. The mutual information constraints from t04-thin40 also seem to be quite good, though we don't really need them. Thu Jul 8 22:39:55 PDT 2004 Kevin Karplus try1-opt2 looks pretty good, but needs tighter packing. I wonder if C29-C31 is a disulfide. The two Cys are quite close and could bond, but there is also H28 in the neighborhood, so it COULD be a metal-binding site. It seems to be a mainly hydrophobic neighborhood, though, so I'm tempted to make it a disulfide. Y11-H36-C47-E50 and C223-E226 look more likely as metal-binding or ligand interfaces of some sort. Fri Jul 9 10:03:32 PDT 2004 Kevin Karplus try2 looks a bit better than try1. There is a big cleft between the two domains in the template, and try2 preserves that better than try1. There is some question about where the C-terminal helix goes, since it is not in the template. There may be some hinging motion possible around L88 and A203, making different domain interactions possible. I hope that not too much weight is gven to the exact angle of the hinge! I'm a bit worried about the burial of conserved residue D166. For try3, I'll try optimizing with no constraints (from the alignments), then for try4 try optimizing from all existing models. Without constraints, try2 looks better to the cost function than try1 did. Fri Jul 9 16:15:38 PDT 2004 Kevin Karplus The beta hbond terms must be a bit too high--try3 introduced a very bad break between I71 and S72, presumable to get the D64.N-I71.O Hbond. There is also a bad break before I194, but that is in try1 also. Fri Jul 9 20:34:06 PDT 2004 Kevin Karplus try4 has fixed the bad break somewhat. There are still large breaks at F17-R18 and I71-S72. There are also some exposed hydrophobics that could be tucked away a bit better, like F180 and I227. I'm not sure how much gain there will be in polishing try4 though. Rosetta still prefers the repacked version of try2. Fri Jul 16 14:47:16 PDT 2004 Sol Katzman My first task for try 5 will be to get the N-terminal domain sheet complete. Define the following strands. Based on the 1fw1 template and the SCOP domain, this should be in order S2-S1-S3-S4 with S3 antiparallel to the other three. There are sheet constraints for the required bonding in most of the previous runs, but the longest are in decoys/align4.sheets. What is missing from try4 is most of the S4 strand. This includes the break that Kevin mentioned above at I71. So I have added the following Strand and Sheet constraints, and increased the weight for the hbond_geom for beta and beta_pair. S1 StrandConstraint T7 W13 5 S2 StrandConstraint W33 V38 5 S3 StrandConstraint P60 R65 5 S4 StrandConstraint N68 S72 5 # from decoys/align4.sheets SheetConstraint F8 W13 W33 V38 hbond L10 6 SheetConstraint T7 Y12 R65 P60 hbond D9 6 SheetConstraint V61 R65 S72 N68 hbond L62 6 Sat Jul 17 12:12:23 PDT 2004 Sol Katzman The S4 strand is still not attached to the rest of the sheet. Kevin informs us that a bug in undertaker may have been causing improper constraints to be generated from SheetConstraint commands. So try6 is just a copy of try5. Unfortunately the seed was random so no way to be sure if any change is purely due to the undertaker bug fix. Mon Jul 19 10:54:43 PDT 2004 Sol Katzman The S4 strand is still not attached to the rest of the sheet. Another issue is that the C-terminal of the protein, which is beyond the end of the 1fw1A (and other) homologs, is predicted to be buried, but is currently a dangling helix. The t04 prediction is not that strong for a helix. The rasmol script 'ctail' defines this region as 'ctail' Thu Jul 22 16:32:47 PDT 2004 Sol Katzman For try7, updated the hbond_geom weights in cost function due to new undertaker handling. Also, read in both the t04 and t2k fragments (and write out new TemplateAtoms file.) Since the t04 alignments are much preferred, this may help. Increased the S4 SheetConstraint weight and the general weight of constraints. Looked at the rr.280 constraints, but none of them are for residues numbered higher than 210, and the troublesome helix is from E216 onward, there does not seem to be much point in using them. Fri Jul 23 11:21:14 PDT 2004 Sol Katzman The S4 strand, while still not bonded to the rest of the sheet, is now positioned closer than it was in try5. Not sure what to do about the N-terminal helix S214-K233. The CB-burial-14-7 prediction is actually very weak, and the Near script does not show much need to bury this helix either. So I think I will not worry about this. Rosetta likes try7 better than the earlier tries (try1,try2,try3,try4) and finds the clash Q127-P131 as well as a few other, less severe ones. For a polishing run, I will turn up soft_clashes in try8. Fri Jul 23 21:18:56 PDT 2004 Sol Katzman Still have the Q127-P131 clash in try8. I will increase the cost weight for soft_clash even more, slightly reduce the cost weight for breaks. Also, in InitMethodProbs I will increase ReduceClash and ClashingRotamer. Mon Jul 26 09:53:36 PDT 2004 Sol Katzman I copied Kevin's unconstrained.costfcn from T0228. Using it, the best models are try9 and try8, followed by try5,try6,try7. Created try10 costfcn by reducing the weight of 'constraint' down to 5. Also I added the latest rr.constraints, but made all of them use the 'bonus' option. Under this costfcn, try9,try8 are still the best, and try9.repack-nonPC is even better than the remainder. We will see if try10 reduces the clashes. I also reduced super_num_gen in opt2 (300->150), and super_iter in opt2 (4->3). Mon Jul 26 15:46:28 PDT 2004 Sol Katzman In the weekly group meeting, we discussed the fact that part of the problem with the clashes is that in some cases there was not enough training data to set the thresholds for cost. Kevin updated some of the values for these. Another point is that the Proline clashes will not be fixed by repack-nonPC, because it is non-Proline, non-Cysteine. Kevin also made changes to the costfcn that we can use for try11, increasing Dry and Break weights, while decreasing the Constraint weight. He considered that the result of try10 was not dense enough. Tue Jul 27 10:33:57 PDT 2004 Sol Katzman The try11 output is still fairly "foamy" although better than try10. For try12, in InitMethodProbs I will revert ReduceClash and ClashingRotamer to the original values they had before try9. And reduce soft_clashes (50->40), and slightly increase the wet and dry cost weights. Also I remembered to comment out PrintTemplateAtoms which has been on since try 7, when the t04 fragments were added. The try12 output seems just as "foamy" as try11 but I do not know how to do any better. Using the unconstrained.costfcn, the best models are: try12, try11, try10, try9, try8 Rosetta likes try11,try12 followed by try9,try10. Sat Jul 31 04:52:45 PDT 2004 Kevin Karplus The inidividual domains of ty12 look fairly good, but the two domains don't seem to be tightly packed against each other. Maybe we could add some knob-and-hole constraints to try to pushd them together. It would take a while to figure out which pairs to try---maybe F17 with Y222? M74 with L109? ... The constraints should be something like (-10, 4, 7) for atoms that could reasonably be expected to be in contact. Wed Aug 4 11:34:30 PDT 2004 Sol Katzman I guess Kevin's concern is the solvent-filled gap between the end of the long first helix in the C-terminal domain, namely V92-N115, and the beginning of the two parallel helices in the N-terminal domain, namely R18-H28 and M74-L85. But in fact there is a similar gap in the 1fw1A template, so I am not so sure this needs to be closed up too tightly. It is true however, that in the 1fw1A template there is a kink in its first C-terminal helix at N115, which allows it to take two helical turns that are parallel to and "in front" of the two N-terminal helices. In T0232 this would probably correspond to a kink at L109 or T113. Triangulating from 1fw1A: Distance ASN115A.ND2-GLN71A.NE2: 9.796 Distance ASN115A.OD1-SER14A.CB: 7.457 1fw1A T0232 ----- ----- N115 T113 Q71 M74 S14 R18 Constraint T113.CG2 M74.CE 6.796 9.796 12.796 Constraint T113.CG2 R18.NH2 4.457 7.457 10.457 I suspect that this will just bend the entire helix V92-N115 but I suppose it is worth doing this for try13, with constraints weight back up to 8, and break weight down to 60. In parallel, I will make try14 by just decreasing break weight (100 -> 80) and increasing soft_clashes (40 -> 50). Thu Aug 5 11:11:26 PDT 2004 Sol Katzman It looks like try13 and try14 finished at about the same time and collided in making the repack-nonPC versions because the non-uniquely named intermediate files XXXX.pdb, XXXXA.fasta were left in the T0232 directory. I will make the repack models separately now. With either the try13 or try14 costfcn, both try14 gets the best score, with try13 close behind. Rosetta likes the repacked try14 best too. The models look very similar. We did not get any kink in the helix in try13. Fri Aug 6 12:20:23 PDT 2004 Kevin Karplus try14-opt2 looks pretty good to me (though a little foamy), and scores best with the unconstrained costfcn also. It has many small breaks, but no really horrendous ones. I'm still worried about the burial of D166, since there is no compensating salt bridge---not even an Hbond. The try10 costfcn likes try10 best, but the try12 costfn likes try14-opt2 best. The differences between try1, try14, and the models directly from alignment are fairly small. I think we have reached the point of diminishing returns, so I'll just submit this now: try14-opt2 try14-opt2.repack-nonPC try1 T0232-1pd21-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 T0232-1gulA-t04-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 Thu Nov 18 23:35:41 PST 2004 Martina Koeva Based on the smooth gdt scores: best sam-t04 54.3960 (try7-opt1) best submit 54.1174 (also model1) model1 54.1174 auto 52.5196 align 40.8229 robetta best 46.4228 (robetta model3) robetta1 45.4003