Thu Jul 3 09:01:57 PDT 2008 T0484 Make started Thu Jul 3 09:02:33 PDT 2008 Running on peep.cse.ucsc.edu Thu Jul 3 09:47:45 PDT 2008 Kevin Karplus The t2k alignment doesn't show a lot of conservation, though it has 139 sequences (almost all of unknown function). The t06 alignment, with 205 sequences, shows substantial conservation from E38 on. Most of the homologs are bacterial, but there is at least one mitochondrial one. The HMMS do not seem to be coming up with consistent top hits. Thu Jul 3 10:58:26 PDT 2008 Kevin Karplus best e-value from HMMs is 23.8 for 1hdfA, a gamma-Crystallin-like 8-strand greek-key fold, but T0484 is too short for this fold, so this seems unlikely to be a good hit. I think that this is a template-free modeling problem---I just hope we can get some decent pieces to build from. Thu Jul 3 13:15:58 PDT 2008 Kevin Karplus The try1-opt3 model has bad breaks and looks a bit junky. I'll try closing the gaps with try2, and I use the same cost fcn in try3 do make another model from alignments and fragments. Thu Jul 3 15:03:12 PDT 2008 Kevin Karplus The try2 model doesn't manage to close the gaps, but the try3 model is interesting. I'll make a try4 costfcn with the try3 long helix (P13-L33) and try3 sheet, and do another optimization from alignments. Thu Jul 10 10:59:55 PDT 2008 SAM-T08-MQAO hand QA T0484 Submitted Thu Jul 10 10:59:55 PDT 2008 SAM-T08-MQAU hand QA T0484 Submitted Thu Jul 10 10:59:56 PDT 2008 SAM-T08-MQAC hand QA T0484 Submitted Thu Jul 17 16:33:13 PDT 2008 Josue Samayoa So looks like try4-opt3 ran OK. It scores best with the try4 cost function. The model looks a bit foamy. We will do a try5 do try to make it less foamy. It is odd that there appear to be very few hydrophobic residues. Thu Jul 17 17:21:58 PDT 2008 Josue Samayoa We used the try4 sheet and helix constraints We increased the weight on both constraints sets We upped the dry8 and dry12 costs The goal was to optimize burial and force the formation of both the helices and the sheets We looked at the rr constraints and they all looked satisfied try5 is currently running on lopez. Fri Jul 18 13:39:51 PDT 2008 Josue Samayoa I noticed that earlier in the README, kevin mentioned high conservation from the T06 alignments for residues 38-62. I also noticed this is a with very poor secondary structure predictions I am going to try and search using only residues 35-62. so far I have found some poor hits(E-values > 10) to structures in the PDB. I want to check if these match the hits in T0484.t06.top_reported_alignments.html If they are new matches I may try to incorporate constraints from these structures into future try* runs. Fri Jul 18 14:01:15 PDT 2008 Firas Khatib try5 packs better and switched the positioning of the strands compared to try4 Josue noted that the near-backbone-11 predictions hardly have any hydrophobic predictions (which might be why dry8, dry12 are scoring so poorly) I'm going to launch a run using pred_cb14_back instead of pred_nb11_back, just to see if that changes anything (since the burial predictions have more hydrophobic predictions) try6 will have try5's helix and sheet constraints and similar costs to try5 except for the burial. try6 is running on shaw Sat Jul 19 01:01:16 PDT 2008 Firas Khatib I like the packing on try6-opt3, it's the least foamy of all the models so far. I'll let Josue fiddle with the conservation of 32-62, but I think try6 is a good starting point. Sun Jul 20 17:33:46 PDT 2008 Kevin Karplus I don't see any discussion of the metaserver results. MQAU favors the SAM-T08-server, RBO-Proteus_TS1, BAKER-ROBETTA_TS2, GS-KudlatyPRED_TS2, ... MQAC favors SAM-T08-server_TS1, BAKER-ROBETTA_TS2, GS-KudlatyPRED_TS2, ... It doesn't look like metaserver runs have ever been done for this target, so I'll start MQAU1, MQAC1, and MQAY1 (omitting SAM servers). Sun Jul 20 17:40:39 PDT 2008 Kevin Karplus I'll also start MQAU4, MQAC4, and MQAY4 with the try4 costfcn. Sun Jul 20 18:09:02 PDT 2008 Kevin Karplus What models are the MQA* runs working on? MQAU1 SAM-T08-server_TS1 MQAC1 SAM-T08-server_TS1 MQAY1 RBO-Proteus_TS4 MQAU4 Pcons_dot_net_TS1=?BAKER-ROBETTA_TS2 MQAC4 BAKER-ROBETTA_TS2 MQAY4 RBO-Proteus_TS4 So it looks like three different models are being worked on. Sun Jul 20 21:54:06 PDT 2008 Kevin Karplus The metaserver runs are done costfcn favorite models try6 try6-opt3 try5-opt3 try4-opt3.gromacs0 MQAU4-opt3 MQAC4-opt3 try3-opt3 MQAY4-opt3 try5 try5-opt3 try6-opt3 try4-opt3 MQAU4-opt3 MQAC4-opt3 try3-opt3 MQAY1-opt3 MQAY4-opt3 try4 try5-opt3 try4-opt3 try6-opt3 try3-opt3 MQAU4-opt3 MQAY4-opt3 MQAC4-opt3 MQAY1-opt3 try2-opt3 try3 MQAU4-opt3 MQAY4-opt3 MQAY1-opt3 MQAC4-opt3 try5-opt3 try3-opt3 try6-opt3 try4-opt3 try2-opt3 try2 MQAU4-opt3 MQAY4-opt3 MQAY1-opt3 MQAC4-opt3 try5-opt3 try3-opt3 try6-opt3 try4-opt3 try2-opt3 try1 MQAY1-opt3 MQAC1-opt3 try2-opt3 MQAY4-opt3 MQAU1-opt2 try1-opt3 try5-opt3 try6-opt3 try3-opt3 rosetta MQAU4 MQAY4 MQAC4 MQAY1 try3 try4 try6 try5 try2 try1 MQAC1 MQAU1 I noticed that all the runs that Josue did were starting over from alignments, rather than polishing existing models. It might be worthwhile to do a polishing run (perhaps with the try6 costfcn) to optimize all the models based on 1o06A. I didn't care much for the weights in try6.costfcn, so I put together a new try7.costfcn that tries to get the sheets and helices of try6. try7.under is based on try2.under, to polish existing models. I think that the try7 model will supersede all the try3/4/5/6 models, though they are a bit different from each other in how the short parallel sheet is connected to the long helix. try1 and try2 are too awful to consider keeping. Sun Jul 20 23:07:02 PDT 2008 Kevin Karplus Both MQAU4-opt3 and try3-opt3 show some potential for forming rather different 3-strand sheets: try3-opt3 M1 T5 G38 R42 Hbond G40 G38 R42 G57 D61 Hbond I39 MQAU4-opt3 D3 E6 D61 I58 Hbond A59 G51 K55 F62 I58 Hbond I60 Perhaps we should try optimizations with these constraints (or maybe 3 for each set of constraints: from alignments (like try3,4,5,6), from existing models (like try2,7), or metaserver (like MQAY1 and MQAY4). Mon Jul 21 00:36:20 PDT 2008 Firas Khatib I like MQAU4-opt3 and it would be nice to get a 3-strand sheet there, since I haven't seen that in any of our models or any of the server predictions that I have looked at. I do think that T0484.MQAY4-opt3.pdb is pretty ugly. Of the 5 models that are currently in best-models.pdb & superimpose-best.under, it is the only one that is unable to form any kind of sheet. I think we might have more luck forming a 3-strand sheet with MQAU4-opt3. Mon Jul 21 00:51:16 PDT 2008 Firas Khatib I'll try putting Kevin's MQAU4-opt3 sheet constraints into a try8 run based on the existing model: MQAU4-opt3 try8 is running on shaw Mon Jul 21 08:30:14 PDT 2008 Kevin Karplus try8 seems to have improved the putative MQAU4 sheet somehwat, but not enough to actually form any hbonds. I think that we have to let breaks and clashes happen in order to get any movement, then try to remove them afterwards. I've started try9 (from all existing models) with low clash and break weights, to try to get more movement. Mon Jul 21 08:38:24 PDT 2008 Kevin Karplus I also started an MQAY9 metaserver run with the try9 costfcn, and with the try10.costfcn=try9.costfcn, try10 starts from alignments. We need to do the same three jobs for the try3 constraints. Mon Jul 21 08:44:57 PDT 2008 Kevin Karplus OK try11 starting from existing models, try12 starting from alignments, and MQAY11 starting from server models are all running with the handtry3 sheets and try3 helices. RATS! I forgot to check for typos. All three of the try11, try12, MQAY11 runs are junk, because I typed G38 when I should have had E38. (Probably misread GLU as GLY). I'll fix the typo, then try to find the moai jobs that are erroneous and restart them. Mon Jul 21 08:53:43 PDT 2008 Kevin Karplus OK, restarted. Mon Jul 21 10:56:24 PDT 2008 Kevin Karplus MQAY9 is based on Pcons_dot_net_TS1=?BAKER-ROBETTA_TS2 MQAY11 is also based on BAKER-ROBETTA_TS2 The try10 costfcn likes try9-opt3 (from try8-opt3), MQAY9-opt3, try8-opt3, MQAU4-opt3, ... The try12 costfcn likes try11-opt3 (from try7-opt3), try12-opt3 (from alignment to 1o06A), try7-opt3, try6-opt3, MQAY11-opt3. Mon Jul 21 11:16:26 PDT 2008 Kevin Karplus None of the runs made a 3-strand sheet, though some came closer than others. I've updated superimpose-best.under to reflect my current favorites. Firas and Josue can try playing with this some more---I'm running out of ideas. Tue Jul 22 09:34:02 PDT 2008 Kevin Karplus I believe that Josue and Chirag looked at this target yesterday afternoon and decided not to waste time on trying anything different on it, so I have to pick out 5 models to represent our efforts. Of the BAKER-ROBETTA_TS2 models, I'll pick MQAY9-opt3 (closest to forming a third strand) and MQAU4-opt3.gromacs0.repack-nonPC (best rosetta energy). I like these models beter than the 1o06A models, which have parallel instead of antiparallel strands. Of the 1o06A models, I'll pick try11-opt3 and try3-opt3.gromacs0.repack-nonPC As a 5th model, I'll toss in MQAY4-opt3, based on RBO-Proteus_TS4, though it doesn't form a sheet with predicted strand I58-I60. So I'll submit ReadConformPDB T0484.MQAY9-opt3.pdb # < Pcons_dot_net_TS1 =? BAKER-ROBETTA_TS2 # closest to adding 3rd strand ReadConformPDB T0484.MQAU4-opt3.gromacs0.repack-nonPC.pdb # < Pcons_dot_net_TS1 =? BAKER-ROBETTA_TS2 # best rosetta energy ReadConformPDB T0484.try11-opt3.pdb # < try7-opt3 < try6-opt3 < align(1o06A?) # good score ReadConformPDB T0484.try3-opt3.gromacs0.repack-nonPC.pdb # < align(1o06A?) # best rosetta of the 1o06A models ReadConformPDB T0484.MQAY4-opt3.pdb # < RBO-Proteus_TS4 # another metaserver model, though it doesn't have sheet for I58-I60. Tue Jul 22 10:01:03 PDT 2008 Kevin Karplus submitted with comment SAM HMMs did not find any strong hits (best was 1hdfA with E-value > 23, but the target is too short for an 8-strand greek-key fold). I had help from Firas Khatib and Josue Samayoa on this target. We ended up with two meta-server models based mainly on BAKER-ROBETTA_TS2, two SAM+undertaker models based mainly on 1o06A, and one metaserver model based on RBO-Proteus_TS4. The first 4 models were selected mainly because they formed bits of beta structure, as we had predicted a beta strand for I58-I60. All the models agreed on the helix from P14 to L33, and to a lesser extent, to the residues three on each side of the helix. There is a possibility that not much more than that will be resolved in the NMR also. Model 1 T0484.MQAY9-opt3.pdb # < Pcons_dot_net_TS1 =? BAKER-ROBETTA_TS2 # closest to adding 3rd strand 2 T0484.MQAU4-opt3.gromacs0.repack-nonPC.pdb # < Pcons_dot_net_TS1 =? BAKER-ROBETTA_TS2 # best rosetta energy 3 T0484.try11-opt3.pdb # < try7-opt3 < try6-opt3 < align(1o06A?) # good score not from metaserver 4 T0484.try3-opt3.gromacs0.repack-nonPC.pdb # < align(1o06A?) # best rosetta energy of the 1o06A models 5 T0484.MQAY4-opt3.pdb # < RBO-Proteus_TS4 # another metaserver model, though it doesn't have sheet for I58-I60.