Wed Jul 21 13:22:24 PDT 2004 T0246 DUE 25 Aug Wed Jul 21 21:22:27 PDT 2004 Kevin Karplus An obvious homology to 1cnzA, which has MANY templates Thu Jul 22 14:21:23 PDT 2004 Kevin Karplus The try1-opt2 model has a strand and helix that isn't stuck into the sheet and with no obvious place to put it. The strand is K152-I158, and it probably has to be put in between two of the existing strands, most likely on one side or the other of A102-K108. The RR constraints I104,S261,L245 imply that the strand with I104 is flipped over the wrong way. It may also be desirable to put K152-I158 antiparallel between two strands that are parallel in try1: D129-R134 and V234-T237. We should set up a "strands" file to name the strands with the consensus on the predictions of strands from the neural nets, then do an optimization with strand constraints, helix constraints, and bonus constraints for the rr predictions. We could add sheet constraints from try1, excluding the three places where we might want to put the currently extra strand. Sun Aug 8 14:41:51 PDT 2004 Kevin Karplus For try2, I won't add any sheet constraints, but I'll rely more on t04 than t2k for predictions. I'll add the rr constraints just as bonuses. Wed Aug 18 16:01:32 PDT 2004 Kevin Karplus The big question is whether G153-Y159 is in serted as a single strand in the sheet, or whether it gets inserted as a hairpin. This would change the preferred locations for inserting it. Inserting it as a single strand between D129-R134 and V234-T237 still makes sense to me, and the S137-K152 region could be a helix covering a currently exposed part of the sheet. We currently have v131 next to V234 and R134 next to T237, but could easily tolerate phase shifts. We have hbonds available on V131 and I235. If we do a phase shift, it will have to be an even one. Let's try adding constraints for these strands: 129> DLVTVRE dYIMTDFGke <151 232> fdVILTt SheetConstraint D129 R134 D160 D155 hbond D129 SheetConstraint F154 Y159 T238 D233 hbond Y159 Thu Aug 19 14:20:30 PDT 2004 Kevin Karplus Well, try3 got the strand NEAR the sheet, but failed to insert it INTO the sheet. Perhap I need to add some constraints to pry apart the two strands that are still there. Constraint D129.CB D233.CB 8.8 10.2 12 1.0 Constraint V131.CB V234.CB 8.8 10.2 10.6 1.0 Constraint T132.CB I235.CB 8.8 10.2 10.6 1.0 Constraint V133.CB L236.CB 8.8 10.2 10.6 1.0 Constraint R134.CB T237.CB 8.8 10.2 10.6 1.0 Sat Aug 21 13:05:01 PDT 2004 Kevin Karplus try4-opt2 similarly failed to insert the strand. Maybe I need to increase the weights for the strand insertion. Sat Aug 21 20:51:51 PDT 2004 Kevin Karplus try5-opt2 has almost inserted the strand. It may need a little adjustment of the sheet constraints to orient the strand better and form the hbonds. try5 now scores almost as well as tries 1,2,4 on the unconstrained cost function---it has the worst breaks and the worst beta_pair scores of the lot, but still comes within 0.5 points of the best. Wait a sec---there are horrible breaks: T0246.try5-opt2.pdb.gz breaks before (T0246)R161 with cost 37.5744 T0246.try5-opt2.pdb.gz breaks before (T0246)I244 with cost 14.334 T0246.try5-opt2.pdb.gz breaks before (T0246)D149 with cost 6.44256 T0246.try5-opt2.pdb.gz breaks before (T0246)D155 with cost 2.39047 T0246.try5-opt2.pdb.gz breaks before (T0246)V128 with cost 2.19508 The strand from F154 to D160 is backwards! No wonder I've been having trouble forcing it into the sheet! I'll fix the constraints for try6 and run again. Since I already have a model that pried the strands apart, I'll leave out those constraints, and just have the two sheet constraints for the inserted strands (try6.costfcn still scores try5-opt2 best, so it will be working on flipping the strand in that model). Sun Aug 22 07:56:17 PDT 2004 Kevin Karplus try6 gets the strand in more-or-less the right place, but has left the linker all at one end, so that there is a horrendous break between K152 and G153. We can try fixing this by adding an intermediate constraint (say V140 and V133), as well as by adding gap-closing constraints as in ../README. I'll try this in try7, using only try6 as a starting point (to avoid picking up something that lacks the gaps but is otherwise awful). Sun Aug 22 10:39:45 PDT 2004 Kevin Karplus I may need to switch to shorter runs on this target---they are taking a long time. Sun Aug 22 17:31:48 PDT 2004 Kevin Karplus Foo, the breaks just moved around: T0246.try7-opt2.pdb.gz breaks before (T0246)D155 with cost 23.6642 T0246.try7-opt2.pdb.gz breaks before (T0246)I244 with cost 12.9165 T0246.try7-opt2.pdb.gz breaks before (T0246)D149 with cost 6.57915 T0246.try7-opt2.pdb.gz breaks before (T0246)G43 with cost 1.77888 T0246.try7-opt2.pdb.gz breaks before (T0246)N268 with cost 1.47363 Undertaker really does not like reshaping models that it has already packed! try7-opt2 DOES score the best with the unconstrained costfcn, despite the bad breaks and the loss of beta_pairs so maybe I should try polishing it with the unconstrained cost function, and see if undertaker can fix up the mess. The big gains over try1-opt2 (which is very close in score) seem to come from the dry5 and dry6.5 burial terms. Mon Aug 23 15:41:27 PDT 2004 Kevin Karplus In try8-opt2, a little of the strand gets inserted, but a big chunk is attached to the linker, which is not straigtened out. Mon Aug 23 16:38:16 PDT 2004 Kevin Karplus George looked at some of the templates and they all have something hanging out of the sheet where we have an "extra" strand. This is usually a hairpin sticking out at right angles to the sheet, but is sometimes a helix. It makes a dimer and the try1-opt2 seems to be as good as any. We need to make dimers from try1 and try2 superimposed on 1cnz to see what REALLY needs to be fixed. Foo, all that wasted effort---I should have looked at the templates FIRST. From learithe@soe.ucsc.edu Mon Aug 23 16:38:13 2004 Date: Mon, 23 Aug 2004 16:38:12 -0700 (PDT) From: Jenny Draper To: Kevin Karplus Subject: T0246 hand-edit Hi Kevin, Is this: T0246/T0246.try7-opt2.repack-nonPC_handedit.pdb what you were basically looking for? -Jenny From karplus@soe.ucsc.edu Mon Aug 23 16:48:39 2004 Date: Mon, 23 Aug 2004 16:48:38 -0700 From: Kevin Karplus To: learithe@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: Re: T0246 hand-edit Maybe, but George found out something about T0246 that makes all the effort to insert the sheet moot. See the README file. ============================================================ Mon Aug 23 16:59:30 PDT 2004 Sol Katzman I was looking at 1cnz and noticed the hairpin sticking out when George came by and said this had just been "discovered" by the rest of the team. The script pointed to by "strands" defines s1 through s12. The only ones that are perhaps inaccurate are s6 and s7, which define the hairpin in 1cnz that sticks out. Mon Aug 23 17:25:26 PDT 2004 Kevin Karplus I think that I'm just going to do an unconstrained polishing run starting from try1 and try2, ignoring my wasted intermediate effort. Mon Aug 23 22:26:26 PDT 2004 Kevin Karplus I'll submit try9-opt2 best unconstrained try9-opt2.repack-nonPC best Rosetta energy try1-opt2 full auto T0246-1cnzA-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 best template T0246-1iso-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 next best (sufficiently different) template Fri Sep 24 21:36:36 PDT 2004 Kevin Karplus Foo, it looks like robetta beat us on this one: name length missing_atoms rmsd rmsd_ca GDT smooth_GDT robetta-model1.pdb.gz 354 0.0000 20.3447 2.0571 -81.5678 -76.6670 robetta-model5.pdb.gz 354 0.0000 20.2305 1.9281 -81.4972 -76.5188 robetta-model2.pdb.gz 354 0.0000 20.2049 2.1286 -81.2147 -76.2219 robetta-model4.pdb.gz 354 0.0000 20.1970 2.1823 -80.7203 -75.9179 robetta-model3.pdb.gz 354 0.0000 20.2960 2.4909 -79.9435 -75.0405 model4.ts-submitted 354 31 20.3768 2.3909 -73.0932 -69.3816 template model2.ts-submitted 354 0.0000 20.4101 2.2092 -73.3051 -69.2043 try9-repack model1.ts-submitted 354 0.0000 20.4385 2.2092 -73.3051 -69.2115 try9 model3.ts-submitted 354 0.0000 20.4490 2.1987 -73.7994 -68.6279 full auto model5.ts-submitted 354 156 18.7556 3.1539 -64.7599 -59.9132 template Wait a second: how can we have 20 Ang RMSD with <2 Ang rmsd_ca? Maybe some of the residues of 1vlcA are backbone only, and the rmsd costfcn is clearly not handling incompleteness of the real conformation correctly, but why doesn't the evaluation report missing atoms for 1vcA? This needs to be investigated and fixed. Sat Sep 25 18:10:31 PDT 2004 Kevin Karplus RMSD bug fixed---it was caused by having some incomplete residues in 1vlcA, and an incomplete implementation of contains_atom() in AlignedFragments. name length missing_atoms rmsd rmsd_ca GDT smooth_GDT robetta-model1.pdb.gz 354 0.0000 2.6421 2.0571 -81.5678 -76.6670 robetta-model5.pdb.gz 354 0.0000 2.3708 1.9281 -81.4972 -76.5188 robetta-model2.pdb.gz 354 0.0000 2.6908 2.1286 -81.2147 -76.2219 robetta-model4.pdb.gz 354 0.0000 2.5481 2.1823 -80.7203 -75.9179 robetta-model3.pdb.gz 354 0.0000 2.8854 2.4909 -79.9435 -75.0405 model4.ts-submitted 354 31 2.7193 2.3909 -73.0932 -69.3816 T0246.try9-opt1.pdb.gz 354 0.0000 2.5541 2.2094 -73.4463 -69.3260 model1.ts-submitted 354 0.0000 2.5399 2.2092 -73.3051 -69.2115 T0246.try9-opt2.pdb.gz 354 0.0000 2.5399 2.2092 -73.3051 -69.2112 model2.ts-submitted 354 0.0000 2.5673 2.2092 -73.3051 -69.2043 model3.ts-submitted 354 0.0000 2.5642 2.1987 -73.7994 -68.6279 model5.ts-submitted 354 156 3.6578 3.1539 -64.7599 -59.9132