Fri Jun 30 09:42:08 PDT 2006 T0354 Make started Fri Jun 30 09:44:03 PDT 2006 Running on lopez.cse.ucsc.edu Fri Jun 30 09:46:35 PDT 2006 Kevin Karplus Rats! I accidentally stepped on the README file from yesterday by redoing the new-target command. Oh, well, there wasn't much in it. BLAST finds nothing in pdb: best is all C-terminal stuff that mainly matches the HIS tag, and gets E-values of 0.014 or worse. HMMs didn't find much either: best is 1y2qA at E-value 0.6. Fri Jun 30 12:14:48 PDT 2006 Kevin Karplus The alignments aren't finding much (scraps of supersecondary structure) and they don't agree, so we'll have to treat this as a new-fold prediction. Sat Jul 15 15:30:28 PDT 2006 Pinal Kanabar I looked at summary.html - as kevin wrote eariler there is no good match. In my try2.costfcn - I have increased HelixConstraint N45 E61 1 (inc.) HelixConstraint I3 D16 0.4 (decreased) Goal : 3 models agree for helix around residue N45 to E61 so I want to see inc wt make any difference. In '1y2qA' It starts with strand so I want to see that if I decreasing wt ... converts it to strand? commented #include T0354.dssp-ehl2.constraints #include rr.constraints try2 is running on camano Mon Jul 17 10:31:58 PDT 2006 Pinal Kanabar My try2 is scoring less than try1 but lookes little better than try1. try1 had 3-4 breaks. score-all+serverscore-all+server sam scores the best : SAM_T06_server_TS1 SAM_T06_server_TS1-scwr1 looking at SAM_T06_server_TS1 has break - and looks preety. Looking at T0354.t06.str2-logo.pdf there are more than few ways strand can go around. There is strong prediction from 76- 90 for strand but the server model has helix. Mon Jul 17 12:21:26 PDT 2006 Pinal Kanabar try3.under : try3.costfcn #Keeping some alignments from try2.under which are matching the server model HelixConstraint M1 K16 0.9 #inc wt as 3 models and 1y2q agrees for this part HelixConstraint N45 E61 1.0 SheetConstraint K20 L30 G42 S32 hbond D21 1.0 SheetConstraint S32 G42 V89 L79 hbond S32 1.0 increased break -> 100 I realy dont know whether this correct .... but i will just run it. Running on camano. Mon Jul 17 15:05:36 PDT 2006 Pinal Kanabar try3 crashed - as there was problem with sheet contrains and also try.under it could not read the server model. #Keeping some alignments from try2.under which are matching the server model HelixConstraint M1 K18 0.9 #inc wt as 3 models and 1y2q agrees for this part HelixConstraint N45 E61 1.0 SheetConstraint K20 K29 G42 L33 hbond D21 1.0 SheetConstraint L33 G42 V89 V80 hbond F34 1.0 try4 is running on whidbey Tue Jul 18 00:17:31 PDT 2006 Firas Khatib for try5, I am using the constraints from str2 T06 and most of the non-bonus rr.constraints to see what happens. Tue Jul 18 11:34:31 PDT 2006 Pinal Kanabar try4 looks bad - I could not force undertaker to take "SAM_T06_server_TS1" server model and strand from helix. try5 looks better than try4 and also scores well. Still in decoys/score-all+servers.unconstrained.pretty server model : SAM_T06_server_TS1 scores the best. I am going to try try6 with str2 alignment. recapping : try2.under->try1.under try3--->crashed try4.under->SAM_T06_server_TS1 (with extra strand contraints : goal was to convert helix in the server model to strand : residue :77-89 which were strogly prdicted as strand in str2 alphabet) try5->try1 try6.under->SAM_T06_server_TS1 and str2 constranits try6.under ->SAM_T06_server_TS1 and str2 constraints I still like ther server model and also str2 predictions are strong so I am going to use both of them together and see if i can conver the helix to strand try6. costfcn : #include T0354.t06.str2.constraints HelixConstraint I3 I17 1.0 StrandConstraint I22 D26 0.85 StrandConstraint R36 A40 0.856 HelixConstraint N45 E61 1.0 StrandConstraint S69 E70 0.2 StrandConstraint W77 D81 0.85 StrandConstraint V85 V89 0.85 HelixConstraint V94 R95 0.611 HelixConstraint Y98 A102 0.633 HelixConstraint S120 V122 0.626 Make started Tue Jul 18 12:24:17 PDT 2006 Pinal Kanabar Running on whidbey.cse.ucsc.edu Redoing make on Whidbey try6 is running on camano Tue Jul 18 14:37:23 PDT 2006 Firas Khatib try5 made some nice sheets and scores best with it's costfcn and scores 2nd best with unconstrained. I am running try8 with try5's sheets and the rr constraints again try8 is running on iris Tue Jul 18 15:41:33 PDT 2006 Pinal Kanabar I am running try7 with try5's & try6's sheet on camano it looks good Wed Jul 19 18:06:08 PDT 2006 Pinal Kanabar try9 based on try7 constraints running on camano #include sheet constrains from try7-opt2.sheets SheetConstraint (T0354)D21 (T0354)L30 (T0354)G42 (T0354)L33 hbond (T0354)I22 18 SheetConstraint (T0354)V39 (T0354)D43 (T0354)G68 (T0354)H72 hbond (T0354)A40 18 SheetConstraint (T0354)L79 (T0354)A82 (T0354)H88 (T0354)V85 hbond (T0354)V80 18 #new constraint added for try9 SheetConstraint (T0354)I22 (T0354)L25 (T0354)L79 (T0354)A82 hbond (T0354)I23 20 recapping : T0354.try5-opt2.pdb -> try1 T0354.try7-opt2.pdb -> try5 & try6 sheet constraints (but basic model was based on ->SAM_T06_server_TS1 as this model scored best) T0354.try8-opt2.pdb -> try5's sheets and rr constraints (originally based on try1) T0354.try6-opt2.pdb -> SAM_T06_server_TS1 and str2 constraints T0354.try1-opt2.pdb Wed Jul 19 18:30:28 PDT 2006 Firas Khatib so there is a soft deadline for this target tomorrow. It is due friday and we will do more work on it, but for now: How all the models score: try5 > try7 > try8 > try 6> try1 using the unconstrained costfunction try7 > try5 > try8 > try 4> try2 using try9's costfcn try5 > try8 > try7 > try6 > try4 using try8's costfcn try6 > try7 > try5 > try8 > try1 using try7's costfcn try6 > try5 > try7 > try8 > try2 using try6's costfcn Pinal is still running try9 which is based on try7, based on SAM_T06_server_TS1 For the soft-deadline (until try9 is done) I think Pinal's order is fine: T0354.try5-opt2.pdb -> try1 T0354.try7-opt2.pdb -> try5 & try6 sheet consts (model based on SAM_T06_server_TS1) T0354.try8-opt2.pdb -> try5's sheets and rr constraints (originally based on try1) T0354.try6-opt2.pdb -> SAM_T06_server_TS1 and str2 constraints T0354.try1-opt2.pdb Looking at best-models.pdb, all 5 models are diverse enough with maybe only try8 being a bit too similar (except for the N-terminus) to the rest. If try9 is worth submitting, I would probably replace try8. As for alignments: try5 used: 1wy7, 1tqj and 1wqa try7 used: 2gfg try8 used: 1v6t, 1mp9, 1wqa, 1dly, 1tqj and 1vmb try6 used: 1mp9 and 1z15 try1 used: 1vmb, 1tqj and 1wqa Wed Jul 19 22:01:29 PDT 2006 Kevin Karplus I don't care that much for try5 or try8---I think try7 looks better. try9 looks pretty useless. I'll move try7 to first position and do the submission. Wed Jul 19 22:16:03 PDT 2006 Kevin Karplus Submission done with comment Our initial search found only small scraps of secondary or super-secondary structure, and we have not yet put together a convincing model of the whole fold. Model 1 is try7-opt2, optimized from a random model with sheet constraints from try5. Only fragments were used, not alignments. Model 2 is try5-opt2, optimized from alignments with no sheet constraints. Model 3 is try8-opt2, optimized from alignments with sheet constraints taken from try5. Model 4 is try6-opt2, optimized from a random model, but an alignment to 1mp9A was used in creating the finally chosen model. No sheet constraints were used. Model 5 is try1-opt2, the fully automatic model. ------------------------------------------------------------ Wed Jul 19 22:16:45 PDT 2006 Kevin Karplus I found it a little strange to be making models with just fragment insertion and no alignments, but the alignments are so poor for this template that it makes some sense. One possible improvement to try7-opt2 would be to add sheet constraints for I22-D26 parallel to W77-D81, perhaps SheetConstraint I22 D26 W77 D81 hbond L25 to the sheet constraints and helix constraints extracted from try7-opt2, then either starting from scratch or polishing try7. Have the n_sep and o_sep predictions been looked at to get possible extra H-bonds? There seems to be C-terminal helix cap: Hbond V64.N L59.O Hbond G63.N K60.O hairpin Hbond E70.N V78.O Hbond V78.N E70.O Hbond H72.N E76.O hairpin (less clear here) Hbond V80.N V87.O Hbond V87.N V80.O Thu Jul 20 11:56:38 PDT 2006 Pinal Kanabar Firas found big spelling mistake in my try4/6/7/9 all them had wrong spelling of ReadConformPDB so basically it never read the server model as its template basically it ignored that command - so I dont know what that models are based on. I will discuss with Firas and see if I should rerun certain try's I started try10 with following changes : try10.under InfilePrefix decoys/ ReadConformPDB T0354.try7-opt2.pdb try10.costfcn #try7-opt2.helices HelixConstraint (T0354)M1 (T0354)K18 HelixConstraint (T0354)N45 (T0354)A62 HelixConstraint (T0354)V94 (T0354)D99 HelixConstraint (T0354)I100 (T0354)G106 HelixConstraint (T0354)P118 (T0354)H130 #try7-opt2.sheets SheetConstraint (T0354)D21 (T0354)L30 (T0354)G42 (T0354)L33 hbond (T0354)I22 18 SheetConstraint (T0354)V39 (T0354)D43 (T0354)G68 (T0354)H72 hbond (T0354)A40 18 SheetConstraint (T0354)L79 (T0354)A82 (T0354)H88 (T0354)V85 hbond (T0354)V80 18 #new constraints added as per Kevin's suggestion. SheetConstraint (T0354)I22 (T0354)D26 (T0354)W77 (T0354)D81 hbond (T0354)L25 20 try10 is running on vashon Thu Jul 20 12:30:22 PDT 2006 Pinal Kanabar try12 I am going to start try12 with SAM_T06_server_TS1 and str2 constraints which same as try6 correct spelling of ReadConformPDB. I am running try12 on vashon Thu Jul 20 12:42:52 PDT 2006 Firas Khatib try11 is running on shaw, it uses try7 as input, as well as it's helices & sheets. I added some hbonds from sep predictions: #add in sep predictions for hairpin around 70-78 Hbond E70.N V78.O 10 Hbond V78.N E70.O 10 Hbond H72.N E76.O 10 and a sheet constraint that is consistent with this hairpin: SheetConstraint (T0354)G68 (T0354)E70 (T0354)V80 (T0354)V78 hbond (T0354)E70 20 I have also Proteinshopped this sheet constraint (manually) and will run this as try13. Thu Jul 20 13:35:26 PDT 2006 Firas Khatib try13 is running on shaw try15 is a proteinshopped version of try10-opt2 so that SheetConstraint I22 D26 W77 D81 hbond L25 will hopefully actually work. try15 is running on lopez Thu Jul 20 13:39:04 PDT 2006 Pinal Kanabar I am running try14 with try10 as input and and with try10's and sheet and helix constraints + constraints suggested by kevin. try14 is running on vashon Thu Jul 20 14:11:53 PDT 2006 Firas Khatib recapping what trys came from what, since we noticed that try4/6/7/9 we wrong: try7 -> random model using try5's constraints [alignments used: 2gfg] try8 -> used & rr constraints, str2 constraints, & try5's sheets [1wy7, 1tqj 1wqa] try10 -> used try7-opt2 and constraints, added sheet constraint try11 -> used try7-opt2 and constraints, added sep hbonds and diff sheet constraint try12 -> used SAM_T06_server_TS1 as input and str2 constraints (rerun of try6) try13 -> proteinshopped try7-opt2 model try14 -> used try10-opt2 and constraints, increased constraints from try10 try15 -> proteinshopped try10-opt2 model try16 -> SAM_T06_server_TS1 input, str2 constraints, try10's sheets constraints Thu Jul 20 14:24:35 PDT 2006 Firas Khatib drat, try13 has parallel sheets instead of anti-parallel, I got the constraints wrong! I will rerun try13 as try17 with the correct sheet direction! Thu Jul 20 15:00:25 PDT 2006 Pinal Kanabar try12 didnt work : I wanted to convert the helix to strand but I was unsucceful. It scores better than try15 (score-all.try12.pretty) and best with score-all+servers.unconstrained.pretty try16 is running on camano with different wt and SAM_T06_server_TS1 input, str2 constraints, try7's sheets constraints Thu Jul 20 15:22:06 PDT 2006 Firas Khatib try17 is running on lopez as a re-run of try13 try14 & try15 both finished and scored very well with each other's costfcns try15 scores 2nd best with unconstrained and since it is the only model that actually got the sheets the way we wanted them (by using a proteinshopped model based off try10, which was based off try7) try15 is my favorite model so far. the burial on try15 could use some work. I think that proteinshopping both ends of the chain might help, so I will try that! Thu Jul 20 15:45:56 PDT 2006 Firas Khatib recap: try7 -> random model using try5's constraints [alignments used: 2gfg] try10 -> used try7-opt2 and constraints, added sheet constraint try12 -> used SAM_T06_server_TS1 as input and str2 constraints (rerun of try6) try13 -> proteinshopped try7-opt2 model try14 -> used try10-opt2 and constraints, increased constraints from try10 try15 -> proteinshopped try10-opt2 model try16 -> SAM_T06_server_TS1 input, str2 constraints, try10's sheets constraints try17 -> proteinshopped try7-opt2 model (rerun of try13) try18 -> proteinshopped try15-opt2 model to optimize burial favorites so far: try15,try14,try12 try18 is running on shaw Thu Jul 20 16:35:57 PDT 2006 Firas Khatib try16 is garbage try17 got close, but couldn't quite form the sheet at 68-70/80-82 correctly... try18 did exactly what I wanted it to! It has very different burial than the other tries, but keeps the sheets nice. try18 and try15 are my favorites, and they even score well with unconstrained (after try12, which we don't like) Thu Jul 20 17:59:47 PDT 2006 Firas Khatib I think that these five models are different enough to submit: Final ranking from superimpose-best.under Here is the report: try18 comes from a Proteinshop model that was based on try15 -> try10 -> try7 try18 scores best with the try18 costfcn and scores 2nd using unconstrained try15 comes from a Proteinshop model that was based on try10 -> try7 try15 scores 2nd best using try18's costfcn and 3rd best using unconstrained try14 came from try10-opt2 which came from try7 try14 scores 3rd best using try18's costfcn and 5th using unconstrained try12 is the best scoring model using the unconstrained costfunction try12 is based on SAM_T06_server_TS1 and str2 constraints try17 comes from a Proteinshop model that was based on try7 try17 doesn't score incredibly good, but is different enough I feel. I would have no problem replacing try17 with another model that say, Rosetta prefers or something. Thu Jul 20 19:58:29 PDT 2006 Kevin Karplus I replaced try15-opt2 with try15-opt2.gromacs0.repack-nonPC, since that was Rosetta's favorite. Submitted #try18 comes from a Proteinshop model that was based on try15 -> try10 -> try7 #try18 scores best with the try18 costfcn and scores 2nd using unconstrained ReadConformPDB T0354.try18-opt2.pdb #try15 comes from a Proteinshop model that was based on try10 -> try7 #try15 scores 2nd best using try18's costfcn and 3rd best using unconstrained # try15-opt2.gromacs0.repack-nonPC is rosetta's favorite of the models #it repacked. ReadConformPDB T0354.try15-opt2.gromacs0.repack-nonPC.pdb #try14 came from try10-opt2 which came from try7 #try14 scores 3rd best using try18's costfcn and 5th using unconstrained ReadConformPDB T0354.try14-opt2.pdb #try12 is the best scoring model using the unconstrained costfunction #try12 is based on SAM_T06_server_TS1 and str2 constraints ReadConformPDB T0354.try12-opt2.pdb #try17 comes from a Proteinshop model that was based on try7 #try17 doesn't score incredibly good, but is different enough I feel. ReadConformPDB T0354.try17-opt2.pdb