Mon Jul 17 09:35:52 PDT 2006 T0383 Make started Mon Jul 17 09:37:50 PDT 2006 Running on cheep.cse.ucsc.edu Mon Jul 31 17:01:23 PDT 2006 Sylvia Do I fetched the server models and made the score-all+servers.unconstrained.pretty.I have not done snything with it yet. SAM_T06_server_TS1 is the best scoring one. There is a break between D46 and N49, which is probably a hairpin turn. Perhaps moving the P32-N49 strand forward by one will fix this. Another break, before N85. I'll just up the break constraints to 100 to fix the N85 break first. If I'm lucky,the other one will close as well. try2 is running on abyss Tue Aug 1 11:59:41 PDT 2006 Sylvia Do It looks like the two breaks from before have closed in try2; but in the process, another one formed. And it's a pretty bad break at that: before (T0383)Q50 with cost 2.26491 This is close to the 46-49 break, so it probably just "moved" the break. I just put InFilePrefix decoys/servers/ ReadConformPDB SAM_T06_server_TS1.pdb #ReadConformPDB SAM_T06_server_TS1-scwrl.pdb ReadConformPDB ROBETTA_TS1.pdb into the superimpose-best file. The Robetta model has a good point - it has one of the sheets be a helix, and that covers the exposed residues on the other side of the beta sheets. The helix that Robetta has is S1-N27.It is not predicted to be a helix. Robetta tends to like helices, I believe, so I'll keep that in mind. These are the changes to try1.costfcn, for try3.costfcn: #include T0383.dssp-ehl2.constraints #include T0383.undertaker-align.sheets #include rr.constraints //from try1-opt2.helices HelixConstraint (T0383)N91 (T0383)R100 1 HelixConstraint (T0383)L98 (T0383)T114 1 HelixConstraint (T0383)V113 (T0383)L120 1 //added this HelixConstraint (T0383)S1 (T0383)N27 2 This is an attempt to make the helices that Robetta has. try3 is running on lopez. try4 will just try to move the 32-45 sheet over by one and close the gap: //from try1-opt2.sheets SheetConstraint L6 E9 G81 H78 hbond K7 1 SheetConstraint Q10 S14 H78 S74 hbond Q10 1 SheetConstraint Q15 A20 S74 I69 hbond H17 1 SheetConstraint Q50 I62 D80 V68 hbond V51 1 SheetConstraint M61 V63 V68 K66 hbond V63 1 //changed this SheetConstraint P32 H45 I62 Q50 hbond E34 1 try4 is running on camano. Tue Aug 1 13:51:09 PDT 2006 Sylvia Do try3 has too many helices. On the plus side, no big breaks. I'm going to try again, but with the sheet constraints included as well: //from try1-opt2.sheets SheetConstraint (T0383)P32 (T0383)H45 (T0383)V63 (T0383)Q50 hbond (T0383)E33 1 SheetConstraint (T0383)Q50 (T0383)I62 (T0383)D80 (T0383)V68 hbond (T0383)V51 1 SheetConstraint (T0383)M61 (T0383)V63 (T0383)V68 (T0383)K66 hbond (T0383)V63 1 try5 is running on peep. Tue Aug 1 14:53:13 PDT 2006 Sylvia Do try4 is no good. Looks like both undertaker and rosetta like try5. Rosetta likes try3 more, but it likes helices. Interestingly, try4 is second best scoring for undertaker, even with its bad breaks and clashes. try6 is try4, with soft clashes upped to 30, backbone_clashes upped to 3, and breaks upped to 100. try6 is running on abyss. Wed Aug 2 15:50:18 PDT 2006 Sylvia Do, George Shackelford George is helping me out with this target. Using Alphabetmatch, using the ehl2 and the burial alphabets, we got 1dwnA as our top result. We've added these: 1dwnA 1mvaA 1aq3A 2gx9A; to the MANUAL_TOP_HITS. try7 is using InfilePrefix 1dwnA/ and InfilePrefix 1mvaA/ for its .under file. The costfcn file is altered from try 1 by: //include T0383.dssp-ehl2.constraints //include T0383.undertaker-align.sheets try7 is running on peep. Wed Aug 2 16:55:49 PDT 2006 Sylvia Do, George Shackelford There were some disputes between t2k and t06 abouta hleix/strand position. Using the t2k's aphabetmatch, 2gx9A is now the top scoring one (though its score is not that great). try8 will use 2gx9A in its .under file, and will have the three constraints off. try8 is running on peep. I will also try to run try7 as try9, with the rr constraints off, to see how that will do. try9 is running on shaw. 1-37-ish have some helix issues (try 7,9). They should be sheets, but they are curling.I think those will need some sheet constraints. Hopefully it will curl under like in 1dwn. I'll try some sheet constraints (and ProteinShopping, if the sheets won't bend around. I was looking at try8-opt1, and that looks pretty bad. Thu Aug 3 12:59:59 PDT 2006 Sylvia Do Added SheetConstraint L6 E9 G81 H78 hbond K7 1 SheetConstraint Q10 S14 H78 S74 hbond Q10 1 SheetConstraint Q15 A20 S74 I69 hbond H17 1 SheetConstraint P32 H45 V63 Q50 hbond E33 1 SheetConstraint Q50 I62 D80 V68 hbond V51 1 SheetConstraint M61 V63 V68 K66 hbond V63 1 to try7's costfcn. Try10 is running on shaw. Thu Aug 3 15:44:57 PDT 2006 Sylvia Do Wow. That was ... bad. Try10 did not work very well with the sheet constraints. For try11, I will use: //from try7-opt2.sheets SheetConstraint F12 S14 N22 A20 hbond V13 1 SheetConstraint V36 Q44 L58 Q50 hbond D37 1 SheetConstraint H45 D46 Q50 N49 hbond D46 1 SheetConstraint E48 F60 S88 V76 hbond Q50 1 SheetConstraint M61 K66 S74 I69 hbond I62 1 //added this SheetConstraint N2 V13 P32 R21 hbond N2 1 try11 is running on peep. Try11 is looking good so far. try12 is 11 with drys and soft clashes constraints upped. try12 is running on peep. Thu Aug 3 17:37:59 PDT 2006 Sylvia Do Going to try to make it look prettier. changed the added constraints to //added this SheetConstraint N2 R8 H17 E11 hbond N2 1 SheetConstraint A20 N27 D37 G30 hbond A20 1 try13 is running on orcas. try12 is messing up all my efforts to get rid of those helices! Fri Aug 4 11:50:41 PDT 2006 Sylvia Do Try13 has the helices back. Try14 will have //SheetConstraint F12 S14 N22 A20 hbond V13 1 SheetConstraint V36 Q44 L58 Q50 hbond D37 1 SheetConstraint H45 D46 Q50 N49 hbond D46 1 SheetConstraint E48 F60 S88 V76 hbond Q50 1 SheetConstraint M61 K66 S74 I69 hbond I62 1 //added this SheetConstraint N2 R8 H17 E11 hbond N2 3 SheetConstraint A20 N27 D37 G30 hbond A20 3 try14 is running on lopez. Fri Aug 4 12:24:44 PDT 2006 Cynthia Hsu try14-opt1 has finished, and it doesn't look very good. The C-terminal helix (approx Q92-F127) is protruding off into the environment, and her desired sheets are still being extremely stubborn about forming. They exist as coils at the moment. The curious thing is that in Sylvia's sheet constraints, she leaves two residues (28 and 29) between them. That does not seem to be enough to form a hairpin turn, which may be why it keeps trying to form a sheet and twist back on itself. Because of this, I've changed the SheetConstraint as follows: SheetConstraint A20 N27 V38 A31 hbond A20 3 //used to be from D37-G30. I've also raised "constraints" to 12. Because I didn't like the orientation of the helix, I felt like I ought to include some rr constraints to hold it in place. Constraint V76.CB I123.CB -10. 7.0 14.0 0.5 Constraint V76.CB L125.CB -10. 7.0 14.0 0.5 Constraint V56.CB V113.CB -10. 7.0 14.0 0.5 try16 is currently running on camano. Fri Aug 4 12:54:05 PDT 2006 Sylvia Do Gah. Yeah. I took the number of residues into consideration, but I guess I had an off-by-one error. -_-;; Anyway, before I ran off to lunch, I was seeing if I can improve try4. Try15 is running on peep. Try17 will have //from try7-opt2.sheets //SheetConstraint F12 S14 N22 A20 hbond V13 1 SheetConstraint V36 Q44 L58 Q50 hbond D37 1 SheetConstraint H45 D46 Q50 N49 hbond D46 1 SheetConstraint E48 F60 S88 V76 hbond Q50 1 SheetConstraint M61 K66 S74 I69 hbond I62 1 //added this SheetConstraint N2 R8 F18 F12 hbond N2 3 SheetConstraint A20 N27 V38 A31 hbond A20 3 (minor changes - Cynthia didn't catch my other "two-residue hairpin.") try17 is running on lopez. Fri Aug 4 13:57:11 PDT 2006 Sylvia Do I used ProteinShop to move the helix of try11 -opt2 in order to cover up some of the exposed residues.This is try18. try18 is running on shaw. I just used ProteinShop to move the helix of try11-opt2 to look more like that of 1dwn. This will be try19. try19 is running on peep. Fri Aug 4 14:59:07 PDT 2006 Sylvia Do, Cynthia Hsu In try17, the correct residues are now properly aligned, but they refuse to stay as sheets and instaed become strands. We've added a strand constraint from H17-D19, with a weight of five, to hopefully create a hairpin turn where there is currently a single turn of a helix. The weight on the relevant sheet constraints was also raised to 8, and "maybe_ssbond" was removed. Following the suggestion of Grant, we also commented out rr constraints, then in the try20.under file, we raised "ReduceConstraint" to 25 and "OptSegment" to 5. try20 is currently running on camano. Fri Aug 4 16:15:49 PDT 2006 Sylvia Do try20 is looking good so far. I'm just going to try to get rid of the breaks in try15. The creak cost has been upped to 100 for try 21. try21 is runnin on lopez. try22 is try20, but with the break costs up to 200 and having the distance constraints back on. try22 is running vashon Sat Aug 5 01:53:25 PDT 2006 Kevin Karplus The o_sep predictions have one fairly strong hairpin: SheetConstraint V76 I79 N85 R82 hbond I79 but other neural nets don't agree. The n_sep and o_sep alphabets suggest a helix N22 N29 with C-capping Hbonds Hbond N27.O G30.N Hbond E26.O A31.N The n_notor2 and o_notor 2 alphabets sort of agree on the capping motif, but not on the helix. The notor alphabets don't want the C-terminal helix to be as long as try20 makes it---they have P101-L109 and N91-R100, but the o_sep alphabet does go for a longer helix----we are not seeing much consensus on the secondary structure here! Sat Aug 5 02:58:47 PDT 2006 Kevin Karplus Is superimpose-best.under up to date? try11-opt2 looks terrible, but try15, try5, and try1 look ok. None of them seem to have a consistent dry-side/wet-side pattern for the sheet. The helix should probably be against the dry side. Are there better models? Sat Aug 5 09:55:57 PDT 2006 Kevin Karplus Looking at try20, strand V76-E86 looks like it is clearly upside down, burying charges. (Actually, E86 may be ok, but H78, D89, R82 seem pretty clearly wrong.) Do we have *any* models with the main sheet reasonable? Sat Aug 5 10:10:48 PDT 2006 Kevin Karplus At first, thought that try15-opt2 looked like it has a reasonable sheet, though the helix needs a quarter turn to put its charged side out. This might be doable with ProteinShop, if undertaker can't handle it. Unfortunately the strand T52-I62 seems upside down in this one. (try21 looks similar). So I want try15-opt2, but with T52-I62 flipped over and slid by 1. try5-opt2.sheets: SheetConstraint Q10 S14 H78 S74 hbond Q10 1 SheetConstraint Q15 A20 S74 I69 hbond Q15 1 SheetConstraint P32 H45 V63 Q50 hbond E33 1 SheetConstraint Q50 I62 D80 V68 hbond V51 1 SheetConstraint I62 V63 F67 K66 hbond V63 1 Change to (for try23) SheetConstraint Q10 S14 H78 S74 hbond Q10 1 SheetConstraint Q15 A20 S74 I69 hbond Q15 1 SheetConstraint P32 H45 I62 N49 hbond E33 1 SheetConstraint N49 M61 D80 V68 hbond Q75 1 SheetConstraint I62 V63 F67 K66 hbond V63 1 or (for try24) SheetConstraint Q10 S14 H78 S74 hbond Q10 1 SheetConstraint Q15 A20 S74 I69 hbond Q15 1 SheetConstraint P32 H45 F64 V51 hbond E33 1 SheetConstraint V51 V63 D80 V68 hbond Q75 1 SheetConstraint I62 V63 F67 K66 hbond V63 1 Let's add acouple of packing constraints for the helix also: Constraint V95.CB L42.CB -10 4.8 8 0.5 Constraint L106.CB I73.CB -10 8 12 0.5 Constraint L103.CB L54.CB -10 5 8 0.5 Sat Aug 5 10:39:07 PDT 2006 Kevin Karplus try23 started on cheep and try24 started on lopez. Sat Aug 5 10:57:28 PDT 2006 Kevin Karplus try23 and try24 started from alignments. I also started try25 (on shaw) and try26 (on peep) with the same costfcns but starting from try15-opt2.gromacs and try15-opt2.gromacs.repack-nonPC. I turned up the ShiftSubtree and ShiftSegment operators, which were *intended* to do this sort of strand flipping, but I don't think they really work. Sat Aug 5 12:58:30 PDT 2006 Kevin Karplus SUCCESS! try23 managed to build a sheet with consistent polarity, but try24, try25, and try26 did not. I'll do a polishing run starting from try23 (no constraints). Sat Aug 5 13:05:29 PDT 2006 Kevin Karplus try27 started on cheep. Sat Aug 5 18:46:18 PDT 2006 Kevin Karplus although try27 doesn't score quite as well as try25 or try26 with the try27 unconstrained costfcn, rosetta likes it much better (almost as well as try3). Sat Aug 5 19:21:29 PDT 2006 Kevin Karplus The helix still needs to be turned a little. Perhaps I could put L109-I73, I116-F18 constraints in. Sat Aug 5 20:06:55 PDT 2006 Kevin Karplus Polishing run try28 started on peep (polishing try27-opt2.gromacs0.repack-nonPC). Sat Aug 5 20:57:55 PDT 2006 Kevin Karplus I'll submit ReadConformPDB T0383.try28-opt2.pdb ReadConformPDB T0383.try15-opt2.pdb ReadConformPDB T0383.try5-opt2.gromacs0.repack-nonPC.pdb ReadConformPDB T0383.try1-opt2.pdb ReadConformPDB T0383.try21-opt2.pdb Sat Aug 5 21:26:08 PDT 2006 Kevin Karplus submitted with comment Neither BLAST nor our HMMs got hits for T0383, but we were able to assemble a sheet and helix that look moderately plausible. We relied on burial predictions to get the phase of the strands of the sheet, and they may be off by 2. Burial is not as good as we would like, but we ran out of time to play with the models. There is not really time even to go through the README file and figure out the history of the models. Model 1 is try28-opt2, polished from try27-opt2, from try23-opt2, created from alignments, the most recently added of which is 1xnzA. Sheet constraints and helix constraints were used in the selection of the alignment and optimization of the model. Model 2 is try15-opt2, an earlier model that was used to get sheet constraints for the try23-opt2 optimization (but the constraints for strand N49-I62) were shifted by 1 for the try23 optimization, to flip the strand over). Model 3 is try5-opt2.gromacs0.repack-nonPC, one of rosetta's favorites of the undertaker-produced backbones that it repacked, but with polarity problems on the sheet, and with one predicted strand wound up into a helix. Model 4 is try1-opt2, the fully automatic model. Model 5 is try21-opt2, one of the many models that had polarity problems with the sheet, but otherwise scored reasonably well. ---------------------------------------------------------------------- Sun Aug 20 20:15:43 PDT 2006 Kevin Karplus Our best submitted was model1 (try28-opt2), which was about as good as the SAM_T06_server_TS1 model. try23 was slightly better, but the RAPTORESS_TS2 model was much better. If we had picked among the server models using our unconstrained costfcn, we would have picked RAPTOR-ACE_TS2-scwrl, which was much better than any we generated. Mon Aug 21 11:44:26 PDT 2006 Kevin Karplus Looking at GDT scores, our initial alignment to 1qynA is better than any of our complete models. We seem to have acquired some bad strand misalignments, particularly for S53-F64. Our best model (try23-opt1) has the right fold, but is rather awkwardly misaligned. Tue Aug 22 10:12:53 PDT 2006 Cynthia Hsu The reasons for not picking the RAPTORESS_TS2 model seem to be associated with our unconstrained costfcn. Looking at score-all+servers.unconstrained.pretty, we find that Raptoress is very poorly ranked in comparison to our own generated models, such as tries 25, 28, 15, 27, 24, 3, 26, 20, 9, 7, etc. Looking at the cost function, it is impossible to isolate specific columns that would have indicated the value of the RAPTORESS_TS2 model over our own.