Wed Jul 12 13:59:08 PDT 2006 T0374 Make started Wed Jul 12 14:00:58 PDT 2006 Running on abyss.cse.ucsc.edu Thu Jul 13 11:51:25 PDT 2006 Kevin Karplus BLAST gets a moderate hit to 2fl4A (24% over 82 residues, evalue 0.0009) and a full length hit to 2fe7A (22% over 153 residues, e-value 0.7) The HMMs get lots of hits to d.108.1.1 domains (top is 2ge3A at 1.3e-20) All the top alignments are in agreement. There is a disagreement with the secondary-structure prediction for the C-terminal strand, which the t2k predictions want to turn into a helix. I believe that the strand (as predicted by the t06 alignments) is correct. Thu Jul 13 12:01:14 PDT 2006 Kevin Karplus I just noticed that T0372 and T0374 seem to share the d.108.1.* domain. How similar are they? Mon Jul 24 17:19:54 PDT 2006 Chris Wong Try2: We're starting it again from alignments, but including the sheet and helix constraints from try1. ( make -k T0374.do2 > & do2.log ; gzip -9f do2.log ) & started on vashon at 1720. Wed Jul 26 14:13:08 PDT 2006 Chris Wong try2 model scores same as try1. Looks the same, too. try3 will start from try2,adding helix constraint and upping constraints weight from 10 to 50 HelixConstraint (T0374)R148 (T0374)M154 ( make -k T0374.do3 > & do3.log ; gzip -9f do3.log ) & started on lopez at 1456. Wed Jul 26 15:15:48 PDT 2006 Zack Sanborn I've started a try4 where we're restarting from the alignments and using most of the sheet constraints from try2-opt2 (which had a pretty decent model), but I'm trying to remove the "bulge" in the sheet at F75-V83 by pairing up that strand in a way that better fits the secondary structure. I changed the following sheet constraints from try2-opt2.sheets: SheetConstraint (T0374)N66 (T0374)Q71 (T0374)G79 (T0374)D74 hbond (T0374)Y68 SheetConstraint (T0374)C76 (T0374)L78 (T0374)M113 (T0374)I115 hbond (T0374)C76 to SheetConstraint (T0374)F75 (T0374)L80 (T0374)R111 (T0374)S116 hbond (T0374)F75 SheetConstraint (T0374)V61 (T0374)Q69 (T0374)V83 (T0374)F75 hbond (T0374)Q69 I left the other sheet constraints and all of the helix constraints untouched. If these constraints are satisfied the bulge should be removed. If try4 is successful at fixing this sheet and Chris' try3 is successful at making that helix, we could possibly making a chimera of the two and have yet another model. We could also simply make a new run that uses the helix constraint from Chris' try3 into try4 to (probably) acheive the same effect. try4 started on vashon. Actually, I went ahead and made a try5 that did just what I described above, using all of the constraints from try4 and adding the helix constraint from Chris' try3. However, to do this, I need to remove the following sheet constraints: SheetConstraint (T0374)A137 (T0374)R141 (T0374)Q153 (T0374)V149 hbond (\T0374)A137 SheetConstraint (T0374)H142 (T0374)D143 (T0374)V149 (T0374)R148 hbond (\T0374)H142 Since they conflict with the helix we're trying to force. try5 started on woof. Wed Jul 26 17:20:18 PDT 2006 Chris Wong try3 is done. It made the helix we were going for! Now to fix the strand that got messed up in the process. We'll do a run similar to try3, except adding a sheet constraint to try to make a parallel sheet on the edge. Later, maybe try to put the last helix on the back of the sheet. The sheet constraint is: SheetConstraint (T0374)Q134 (T0374)I138 (T0374)M113 (T0374)C117 hbond (T0374)Q134 1 ( make -k T0374.do6 > & do6.log ; gzip -9f do6.log ) & started on lopez at 1723. Thu Jul 27 12:55:40 PDT 2006 Chris Wong try4 did not fix the bulge in the sheet. Do bulges occur in real proteins? try5 .... I've forgotten what it was we were going for on this one... try6 didn't quite make the sheet like we had hoped. Maybe we can try to ProteinShop the strand into position and then optimize it with some constraints. I'm not going to try to ProteinShop from home... way too painful. Fri Jul 28 18:07:27 PDT 2006 Chris Wong Okay, going to try again to fix that sheet. Using slightly different sheet constraint. Also, going to use the helix and sheet constraints from try2-opt2 because that model scored highest with the unconstrained costfcn. Modification were made so as to avoid conflicts with other constraints we thought were a good idea. Also, going to try to remove the bulge in the sheet at G79 by shifting the sheet constraint over by one residue. ( make -k T0374.do7 > & do7.log ; gzip -9f do7.log ) & started on lopez at 1814. Sat Jul 29 00:01:30 PDT 2006 Chris Wong try7 scores near the bottom of the the unconstrained list. Also, looking at try7-opt2, the bulge in the strand did not go away, and the strand at the end of the protein did not exactly turn out the way we had hoped. Sun Jul 30 16:54:24 PDT 2006 Chris Wong I think we should just leave the bulge as it is for now, expecially since some other members of the CASP team have said bulges like this actually CAN happen. ...but what to do about that strand that doesn't want to form? Oh, it looks like maybe one of the helix constraints is blocking the formation of that strand we want to make. I'm going to remove that helix constraint from the try7.costfcn, and then run a try8. ( make -k T0374.do8 > & do8.log ; gzip -9f do8.log ) & started on orcas at 521p. Mon Jul 31 00:30:15 PDT 2006 Chris Wong try8-opt2 didn't do what I hoped for. It seemed like such a simple task to make that last strand !! with try9, going to see if starting from the alignments helps. Keeping most of the constraints from try8, except increasing the weight of the strand that we want to form. Also, I'm not going to try to pull the bulge out of the other strand this time. ( make -k T0374.do9 > & do9.log ; gzip -9f do9.log ) & started on lopez at 139a. Mon Jul 31 09:50:56 PDT 2006 Chris Wong try9 scores very low.. probably because it started from alignments. There's a bad break near 48 among some other smaller breaks. The whole section from about 110 to 135 looks bad to me... a jumble of coils and breaks. Also, the last strand is trying to thread through into position as the strand next to the edge. I don't like this model. Mon Jul 31 12:21:38 PDT 2006 Chris Wong Okay, I think it's time to see what we've got as far as 5 best models go. The order of unconstrained scores is: 2, 1, 4, 5, 6, 3, 8, 7, 9. 1. try6-opt2 has the basic structure near the edge strand (136). Problem is it doesn't score the best. It doesn't have horrible breaks, either. This would be an even nicer model if the edge strand was more aligned with the rest of the sheet. 2. try2-opt2 looks nice, except that there is a sheet where a helix is predicted. 3. try3-opt2. Problem with this one is that there is a helix where I think a sheet should be (136). 4. ? 5. ? Mon Jul 31 14:09:47 PDT 2006 Chris Wong Okay, here we go again trying to make a nicer edge strand. try10 start from alignments, but with try6-opt2 helices and sheets. Going to set the constraint for that edge strand high. What edge strand constraint, and how heavy to make it? ( make -k T0374.do10 > & do10.log ; gzip -9f do10.log ) & started on lopez at 1531. Tue Aug 1 11:09:17 PDT 2006 Zack Sanborn Well, the sheet I would like to make are not in the try6-opt2. I'm going to take the sheet from try10-opt2 constraints and alter them to try to form a nice 5 stranded sheet. try10-opt2 sheets: SheetConstraint (T0374)Q2 (T0374)A8 (T0374)D58 (T0374)S52 hbond (T0374)S4 1 SheetConstraint (T0374)R49 (T0374)V56 (T0374)Q69 (T0374)L62 hbond (T0374)G51 1 SheetConstraint (T0374)A55 (T0374)H57 (T0374)L62 (T0374)Q60 hbond (T0374)H57 1 SheetConstraint (T0374)L62 (T0374)N66 (T0374)A84 (T0374)N80 hbond (T0374)F64 1 SheetConstraint (T0374)N66 (T0374)Q69 (T0374)G79 (T0374)C76 hbond (T0374)Y68 1 SheetConstraint (T0374)C76 (T0374)L78 (T0374)M113 (T0374)I115 hbond (T0374)L78 1 Altered constraints to (hopefully) remove bulge and form edge strand: SheetConstraint (T0374)Q2 (T0374)A8 (T0374)D58 (T0374)S52 hbond (T0374)S4 1 SheetConstraint (T0374)R49 (T0374)V56 (T0374)Q69 (T0374)L62 hbond (T0374)G51 1 SheetConstraint (T0374)A55 (T0374)H57 (T0374)L62 (T0374)Q60 hbond (T0374)H57 1 SheetConstraint (T0374)L62 (T0374)Q69 (T0374)V83 (T0374)C76 hbond (T0374)Y68 1 SheetConstraint (T0374)C76 (T0374)N80 (T0374)M113 (T0374)C117 hbond (T0374)L78 1 New constraint: SheetConstraint (T0374)Q134 (T0374)R136 (T0374)K114 (T0374)S116 hbond (T0374)R136 1 I will use the identical helix constraints from try10-opt2. try11, restarting from the alignments, using the described constraints started on camano. Tue Aug 1 14:23:38 PDT 2006 Zack Sanborn try11 has finished. It didn't make the strand that I wanted, nor did it get rid of the bulge. I'm beginning to suspect that the bulge must be most of the alignments used to construct our models. However, it has formed a nice extension of the beta sheet from L151-P157. It's odd, but the ehl2 coloring script says that part of this sheet was predicted to be helix, but looking at the t06.dssp-ehl2 sequence logo shows that a sheet should be there. I'm going to optimize try11 since it has some bad breaks and clashes. try11-opt2, in fact, has some huge breaks (one with 14.1). Looking at the gromacs0 optimized model, we see fewer breaks that are generally on the smaller side. So, I'll start my optimization from the gromacs0 model. try12, optimizing try11-opt2.gromacs0 for breaks and soft_clashes, started on orcas. Wed Aug 2 11:36:25 PDT 2006 Zack Sanborn Had some very silly problems getting try12 going. I should have checked the logs, but I instead ignored them until I realized that no try12 models were produced. I fixed the problems, and now try12 has finally started without error. While we wait, I'm going to do a polishing run from all models with heavy weights on breaks, soft_clashes, and the dry weights. This should good us another nice alternative model dissimilar to try12. try13, optimizing from all models, started on orcas. Wed Aug 2 14:29:15 PDT 2006 Zack Sanborn try12 and try13 have both finished. try13-opt2 is the third best scoring model, behind try2-opt2 and try1-opt2, using the unconstrained cost function. However, try13-opt2.repack-nonPC is the best scoring model using ROsetta's cost function. So, that's very good! try12-opt2 doesn't do as well, scoring out of the top 10 using an unconstrained cost function. I've started a polishing run of try12-opt2 to try to get it a higher score in try14. Also, I've started a polishing run from all models in try15. Both trys are using the unconstrained cost function. try14 started on orcas. try15 started on shaw. Wed Aug 2 16:21:40 PDT 2006 Zack Sanborn try14 and try15 have finished and try14-opt2 and try15-opt2 are the top scoring models (in that order) using an unconstrained cost functions. These models will be our top models for submission. I will (mostly) agree with Chris' earlier submission suggestions and choose the following models, adding our top models: T0374.try14-opt2 < try12-opt2 < try11-opt2.gromacs0 < alignments with modified try6-opt2 constraints T0374.try15-opt2 < all models < try12-opt2 < try11-opt2.gromacs0 < alignments with modified try6-opt2 constraints T0374.try13-opt2.gromacs0.repack-nonPC < try2-opt2 < alignments with try1-opt2 constraints T0374.try6-opt2 < try3-opt2 < try2-opt2 < alignments with try1-opt2 constraints. T0374.try3-opt2.gromacs0.repack-nonPC < try2-opt2 < alignments with try1-opt2 constraints. I will now prepare the T0374.method file with a history of each of these targets and submit them. Wed Aug 2 17:23:34 PDT 2006 Zack Sanborn I will submit T0374 with the following comment in T0374.method file: T0374 got moderate hits to 2f14A and 2fe7A, and the HMMs got lots of hits to the d.108.1.1 SCOP domain (top 2ge3A). All of the alignments were in agreement except for some disagreement over the C-terminal, where t2k was predicting a helix and all others predicting a strand in the secondary structure. In all models there is a bulge in the beta strand at residues D74-A84. We attempted to remove the bulge, but were unable to do so. Model 1 is T0374.try14-opt2, our highest scoring model using Undertaker's unconstrained.costfcn cost function. This model came from try12-opt2 < try11-opt2.gromacs0. try11-opt2.gromacs0 came from an Undertaker run deriving a model from alignments (top alignment 1tiqA), applying the sheet and helix constraints from try6-opt2, and then re-optimizing with gromacs0. We chose try6-opt2 sheet and helix constraints because it had formed a nicer beta-sheet than previous models. Model 2 is T0374.try15-opt2, our second highest scoring model using Undertaker's unconstrained.costfcn cost function. This model is a slight variation on try14-opt2, based also on the try12-opt2 model, but aimed at improving breaks, clashes, and hydrophobic packing. Model 3 is T0374.try13-opt2.gromacs0.repack-nonPC, is not a great scoring model according to Undertaker, but its Rosetta-optimized structure, T0374.try13-opt2.gromacs0.repack-nonPC, is the highest scoring model according to Rosetta. try13-opt2 came from an optimization from all models, coming up with the then best current model try2-opt2 to start its optimization, then re-optimized by gromacs and repacked (except for PRO and CYS) by Rosetta. try2-opt2 was made during an Undertaker run starting from alignments (top alignment 1tiqA) using the sheet and helix constraints of try1-opt2, which seemed like reasonable constraints. However, there was disagreement over the C-terminal strand where the t2k predicted a helix, but all others believed it to be a strand. Model 4 is T0374.try6-opt2, a not terribly great scoring model according to Undertaker, but was better liked by Rosetta after its repacking of the sidechains. try6-opt2 is included here since it was one of the first models that had the beginnings of a well-formed edge strand. try6-opt2 was made from an Undertaker optimization of try3-opt2 < try2-opt2. try2-opt2 was made during an Undertaker run starting from alignments (top alignment 1tiqA) using the try1-opt2 sheet and helix constraints. Model 5 is T0374.try3-opt2.gromacs0.repack-nonPC is included mainly for a little bit of variety. It has average scores according to both Undertaker's unconstrained.costfcn cost function and Rosetta's cost function. It has a C-terminal helix where most of our models have a strand. This helix is predicted by t2k but is predicted to be strand by the others. try3-opt2 was an optimization of try2-opt2, then re-optimized by gromacs and repacked (except for PRO and CYS) by rosetta. try2-opt2 was made by Undertaker from alignments (top alignment 1tiqA) using the sheet and helix constraints from the try1-opt2 automatic model. Wed Aug 2 17:31:54 PDT 2006 Zack Sanborn Submitted T0374. Thu Aug 3 06:44:53 PDT 2006 Kevin Karplus Still not showing up on casp7 web site. They're probably wedged. I'll submit once more and send email to casp@predictioncenter.org