Thu May 15 09:26:44 PDT 2008 T0403 Make started Thu May 15 09:27:31 PDT 2008 Running on cheep.cse.ucsc.edu Thu May 15 09:55:47 PDT 2008 Kevin Karplus pdb_blast gets weak hits to 2fgrA and 2fgqX (Evalue 0.002) This is a long protein 339 residues, and a trimer. Thu May 15 13:31:07 PDT 2008 Kevin Karplus There are strong conservation patterns, and easy fold recognition for fold f.4.3.1, a porin. Membrane proteins *will* mess up some cost functions, since the inside is wet and the outside greasy! Thu May 15 14:07:50 PDT 2008 Kevin Karplus It looks from the undertaker alignments that we may be short one strand on the barrel. Figuring out how to close the barrel may be a challenge---or not, depending on how well try1 manages to handle the first and last strand. We'll probably want to increase the hbond costs a lot, and drop the near-backbone predictions way down. I think that the "trimer" nature of the crystal is a red-herring. THe porin is probably natively a monomer. Fri May 16 04:30:30 PDT 2008 Kevin Karplus I think that some of the secondary-structure prediction is wrong--- some of the strands are being predicted to be helices, but the hydrophobic alternation one would expect of a strand on a porin is present. For try2, I'll try making adjustments to the cost function based on what I think a beta-barrel membrane protein would need. I'll lean more heavily on alignments and on predicted contacts, rather than on burial. Mon May 19 22:37:58 PDT 2008 SAM-T08-MQAO hand QA T0403 Submitted Mon May 19 22:37:58 PDT 2008 SAM-T08-MQAU hand QA T0403 Submitted Mon May 19 22:37:58 PDT 2008 SAM-T08-MQAC hand QA T0403 Submitted Fri May 23 09:11:58 PDT 2008 Kevin Karplus Both MQAU and MQAC like the SAM-T08-server best. This is encouraging, though it may just reflect the biases of the align_constraints. I'll have to set up try2, though, before running the metaserver runs. Tue Jun 3 11:19:42 PDT 2008 Kevin Karplus I decided to run metaserver runs with try1.costfcn, as well as with the try2.costfcn. Looking at try1, try2, and Phragment_TS1, I see very good agreement over the body of the barrel, but disagreement over some of the "decorations". It is not clear to me whether any of the differences are ones that we can resolve. It may be best just to do a gap-closing run on the whole set of undertaker-optimized models, perhaps with dry5 and dry6.5 but not the other burial funcions. Tue Jun 3 15:43:49 PDT 2008 Kevin Karplus The top scoring models from each run with try2 are MQAU2-opt3 < BAKER-ROBETTA_TS5 MQAC2-opt3 < Pcons_multi_TS5 try2-opt3.repack-nonPC MQAC1-opt3 < Zhang-Server_TS2 try1-opt3.gromacs0.repack-nonPC MQAU1-opt1 < SAM-T08-server_TS1 THe order is slightly different with try1.costfcn: MQAU2-opt3 MQAC1-opt3 MQAC2-opt3 try1-opt3.gromacs0 MQAU1-opt1 try2-opt3.repack-nonPC (The reason for the opt1 on MQAU1 is that the undertaker run crashed wigh a get_spot assertion failure.) I don't like MQAU2-opt3, despite its good score, because it has excursions into the membrane (K168-K171 and V22-G33). try2-opt3 also has a bad excursion for V22-G33. I think that an early strand is misaligned, and would like to flip it so that we have SheetConstraint I13 V19 T315 A321 Hbond L318 SheetConstraint I13 V17 S46 A42 Hbond G45 I don't know if undertaker is capable of making this change without destroying the barrel. Actually, I'm not so sure I want this way of closing the barrel. try2-opt3 has a different approach, sticking in an extr strand: SheetConstraint (T0403)A15 (T0403)V19 (T0403)M339 (T0403)Q335 hbond (T0403)A15 1 SheetConstraint (T0403)A317 (T0403)G322 (T0403)M339 (T0403)E334 hbond (T0403)L318 1 This is more compatible with the antiparallel barrel! Tue Jun 3 16:28:07 PDT 2008 Kevin Karplus For try3, I'll do a run from alignments again, but pushing the try2 sheets. Tue Jun 3 16:33:58 PDT 2008 Kevin Karplus try3 started on shaw. Tue Jun 3 21:51:05 PDT 2008 Kevin Karplus try3 scores worse than try2, and doesn't get try2.sheets at all well. Perhaps I should try polishing using the try3 costfcn and just the "try" models. try4 started on shaw. Wed Jun 4 08:45:49 PDT 2008 Kevin Karplus try4 likes best MQAU2-opt3 try4-opt3 try2-opt3.gromacs0.repack-nonPC MQAC2-opt3 try3-opt3.gromacs0.repack-nonPC MQAC1-opt3 try1-opt3.gromacs0.repack-nonPC Maybe I should make MQAU2-opt3/try4 chimera. I like try4 better in V290-D310, V17-I48, V156-S176. Actually, V17-I48 may be better taken from MQAC1-opt3. T0403.chimera-MQAU2-try4-MQAC1 built mostly from T0403.MQAU2-opt3, but V290-D310, V156-S176 from try4-opt3 and V17-I48 from MQAC1-opt3 Wed Jun 4 09:13:23 PDT 2008 Kevin Karplus try5 started on moai cluster to fix up the chimera. Wed Jun 4 18:13:19 PDT 2008 Kevin Karplus try5-opt3 looks pretty good except for A161-R174 sticking out into the membrane. Perhaps I could copy that from somewhere else. More gap closure is needed also, since the gromacs0.repack-nonPC version is scoring best. MQAU2-opt3 is still scoring better than try5-opt3 Sun Jun 8 11:38:14 PDT 2008 Kevin Karplus I started a try6 run to try to close gaps on try5. To avoid polishing MQAU2-opt3, I left out all pdb files with MQA in the name. I'm using the try5.sheets constraints and I added "radius" as another costfcn, to try to reduce excursions out into the membrane. Sun Jun 8 20:10:19 PDT 2008 Kevin Karplus In try6-opt3, the loop from P157 to E175 is clearly wrong (sticking out into the membrane). None of the models we have handle this loop well, but try2 and try3 may do a slightly better job. I'm not sure what to do to fix this loop. Add some constraints to pull it into the center of the pore? Maybe make a beta sheet antiparallel to A200-D205? We have the pairing SheetConstraint (T0403)S176 (T0403)N185 (T0403)A197 (T0403)F188 hbond (T0403)Y177 1 so if we skip 2 residues in the turn, we'd have SheetConstraint 168 173 D205 A200 Hbond D201 We might be able to tuck K248-H260 in also with a similar trick, extending SheetConstraint (T0403)V219 (T0403)A233 (T0403)N250 (T0403)L236 hbond (T0403)D221 1 to SheetConstraint H217 A233 V252 L236 hbond D221 1 Top-scoring models with try6 are MQAU2-opt3 try6-opt3.repack-nonPC try5-opt3.gromacs0.repack-nonPC MQAC2-opt3 try4-opt3 Sun Jun 8 20:45:41 PDT 2008 Kevin Karplus I'll try using the sheet constraints listed above in try7 to tuck the loops back in place. Mon Jun 9 13:49:30 PDT 2008 Kevin Karplus try7-opt3 scores well, but P157-R174 and R21-K34 still stick out into the membrane. Tue Jun 10 09:54:22 PDT 2008 Kevin Karplus Possible problem loops are R21-S35, K289-Q299, N249-S254, R158-E175. R21-S35, K289-Q299, might be copiable from try6-opt3 I don't have a better loop for N249-S254 yet, nor for R158-E175. Thu Jun 12 21:43:23 PDT 2008 Kevin Karplus I mae a chimera consisting mainly of try7-opt3, but with R21-S35, K289-Q299 copied from try6-opt3. It has breask and clashes due to the chimera, so I am optimizing it by itself as try8, with a costfcn very similar to try7, but with sheets and helices from try7 instead of try6. Fri Jun 13 08:27:54 PDT 2008 Kevin Karplus try8-opt3 looks pretty good, except for P157-E175 sticking out. Maybe I should add a distance constraint for V164,L202, to try to pull that loop into the inside, instead of the sheet constraint I tried for try8. The best values for the new tuck_157_175 constraints come from try1-init, MQAC1, and MQAC2. I tried making chimeras of try8 with MQAC1 and MQAC2, but these did not seem to be better---I suspect the problem is that MQAC1 and MQAC2 did well not because the P157-E175 loop was tucked but because the other loop was untucked. Let me try adding more loose constraints across the barrel, between V164 and A25 and between V164 and I41, trying to cut the distances in half (down to 14 and 17 respectively. Fri Jun 13 09:29:52 PDT 2008 Kevin Karplus I'll try optimizing from the two try8 chimeras anyway, to see if the constraints can make anything move. I don't expect great results, but it is worth a try. Fri Jun 13 12:25:15 PDT 2008 Kevin Karplus try9-opt3 does not statisfy the tuck_157_175 constraints any better than try8-opt3 did, but the hydrophobic radius of gyration did go down a bit. I probably should turn up the clashes and breaks and try polishing try9. On second thought, maybe I shouldn't since the radius reduction came from peeling the frst strand off the barrel and tucking it into the center, leaving a bad hole in the barrel. Fri Jun 13 13:24:40 PDT 2008 Kevin Karplus The reason that the chimera-try8-MQAC2 initial starting point didn't work is that there are serious clashes, with K166-S173 looping around L202. Maybe I should try optimizing a try8-try1 chimera instead, with D163-E175 from try1-opt3, and P157-N162 from try9. Fri Jun 13 18:14:05 PDT 2008 Kevin Karplus try10-opt3 did not fix the loop, leaving it broken, so was not a useful run. Sun Jun 15 09:45:55 PDT 2008 Kevin Karplus For try11, I'll try again to polish chimera-try8-MQAC1, with strong constraints to try to pull the loop out of the membrane. Of course, this may end up like try10, with a badly broken loop, but I'll then try polishing try10 and try11 with breaks and clashes turned up. Sun Jun 15 12:05:40 PDT 2008 Kevin Karplus The breaks and clashes do get much worse during optimization of try11, so I don't think things will be particularly successful, but I will optimize from try10 and try11 with the tuck_157_175 turned down and breaks and clashes turned up, to see what I can get, once try11 finally finishes. Sun Jun 15 13:05:14 PDT 2008 Kevin Karplus The breaks in try11 are indeed awful, but I'll try an optimization run from try10 and try11 and see if I can close the breaks any. Sun Jun 15 19:49:50 PDT 2008 Kevin Karplus try12 does not succeed in closing the breaks, though it does reduce them. Perhaps without the stress of the tuck_157_175 constraints it would close them. I'll try that for try13. Mon Jun 16 13:00:44 PDT 2008 Kevin Karplus Wow, try13 exploded the porin into two separate halves---definitely NOT the effect I wanted. Wed Jun 18 07:31:32 PDT 2008 Kevin Karplus Just one day left on this target. Looking again at the try13 series, I see that try13-opt2 is still ok---it only explodes in opt3. It scores a little better than try12-opt3. The problem is that for a membrane protein, I can't really use the usual burial terms to hold things compact---I end up crushing the barrel, turning strands inside out, or something else bad. I have to use strand constraints to hold the barrel together, and I didn't make them strong enough in try13. Wed Jun 18 07:44:26 PDT 2008 Kevin Karplus I'm doing another short run for try14, starting with the same initial models as try13, plus the early outputs of try13 (through try13-opt2), with a costfcn that has enough weight on align_constraints, radius, and try8.sheets to make try13-opt2 score better than try13-opt3. Wed Jun 18 10:38:44 PDT 2008 Kevin Karplus try14-opt3 is better, but still doesn't close the gaps. I doubt that I'll make much improvement at this point, so I need to make decisions about which models to submit. Wed Jun 18 13:46:47 PDT 2008 Kevin Karplus I'll make one more stab at a try8-MQAC1 chimera (chimera-try8-MQAC1-a) and optimize it in try15. Wed Jun 18 17:01:16 PDT 2008 Kevin Karplus try15-opt3 scores better than try8-opt3 and the gromacs optimized version does even better. I think I'll submit ReadConformPDB T0403.try15-opt3.gromacs0.pdb ReadConformPDB T0403.try7-opt3.gromacs0.repack-nonPC.pdb ReadConformPDB T0403.MQAC1-opt3.pdb ReadConformPDB T0403.try1-opt3.gromacs0.repack-nonPC.pdb ReadConformPDB T0403.MQAU1-opt1.pdb as 5 models that haven't got too many obvious failings. I could probably polish a little more to remove gaps and clashes, but I doubt that I would make significant improvements. Wed Jun 18 17:28:17 PDT 2008 Kevin Karplus I'll submit with the following additional comments: Because undertaker has no membrane modeling, and a porin is basically inside-out from a normal globular protein, I had to rely heavily on sheet constraints to hold the protein together. Since I had only a few starting models from undertaker, I used our MQA methods to pick out good models from the servers, and did optimization with undertaker from them to get alternative loops. I did a lot of cut-and-paste from different models to try to tuck loops into the barrel and keep them out of the membrane. The models could undoubtedly be refined further, but I'm running out of time. model 1 is T0403.try15-opt3.gromacs0.pdb < chimera-try8-MQAC1-a, which is mostly try8-opt3, but N162-S176 spliced in from MQAC1-opt3 try8-opt3 < chimera-try7-try6, which is mostly try7-opt3, but R21-S35, K289-Q299 from try6-opt3 try7-opt3 < MQAU2-opt3 < BAKER-ROBETTA_TS5 try6-opt3 < try5-opt3.gromacs0.repack-nonPC < chimera-MQAU2-try4-MQAC1, which is mostly MQAU2-opt3, but but V290-D310, V156-S176 from try4-opt3 and V17-I48 from MQAC1-opt3 try4-opt3 < try2-opt3.repack-nonPC < 2j1nA MQAC1-opt3 < Zhang-Server_TS2 model 2 is T0403.try7-opt3.gromacs0.repack-nonPC.pdb try7-opt3 < MQAU2-opt3 < BAKER-ROBETTA_TS5 energy minimized with gromacs, then sidechains repacked by rosetta model 3 is T0403.MQAC1-opt3.pdb MQAC1-opt3 < Zhang-Server_TS2 model 4 is T0403.try1-opt3.gromacs0.repack-nonPC.pdb fully automatic SAM+undertaker prediction energy minimized with gromacs, then sidechains repacked by rosetta model 5 is T0403.MQAU1-opt1.pdb < SAM-T08-server_TS1