Mon Jul 10 09:32:59 PDT 2006 T0366 Make started Mon Jul 10 09:33:42 PDT 2006 Running on shaw.cse.ucsc.edu Mon Jul 10 10:37:28 PDT 2006 Kevin Karplus BLAST has an excellent full-length hit to 2fneA (43% id over 103 residues, E-value 2.2e-20), and good hits to other templates (17 templates with E-value < 1.e-07). The HMM hits that are in SCOP are to family b.36.1.1 (occasionally a different b.36.1.*) and 2fneA comes out on top with the HMMs also. This looks like it will be fairly straightforward comparative modeling. Fri Jul 21 22:24:34 PDT 2006 Kevin Karplus In addition to building a monomer and a dimer, we need to refine the tr366 model that they gave us Fri Jul 21 22:31:36 PDT 2006 Kevin Karplus tr366 (and gromacs and rosetta-repacks of it) score awfully with try1.costfcn and unconstrained.costfcn, but rosetta likes tr366.gromacs0.repack-nonPC best of the models it is has seen so far. Of course, we should probably do most of our optimization in a dimer context. Unfortuantely, our best hit (2fneA) is a hexamer, 2fe5A is a monomer, 1tp5A is a dodecamer, and we have to go all the way to 1g9oA to get a dimer. [Thu Jul 27 05:01:16 PDT 2006 Kevin Karplus Today PQS doesn't have 2fne---perhaps I mistyped it last time. It is a monomer in PDB. ] The main differences between our try1 models and the provided tr366 model are in the termini, though our model and the tr366 model both lose some of the Hbonds of the sheets (different ones). Doing some polishing runs with sheet constraints that include the longest strand matches compatible with any of the alignments or models might be worth doing before building the dimer off of 1g9oA. Mon Jul 24 20:08:05 PDT 2006 Kevin Karplus try1-opt2.sheets SheetConstraint L11 K18 V96 G89 hbond R12 1 SheetConstraint S27 G30 A44 I41 hbond A29 1 SheetConstraint G33 S34 D38 G37 hbond S34 1 SheetConstraint G61 V65 G98 Q94 hbond R63 1 SheetConstraint V65 C68 V95 E92 hbond T66 1 SheetConstraint V65 I67 T72 T70 hbond I67 1 #sheet constraints from conformation 0 T0366-2fneA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:2fneA SheetConstraint M9 K17 G98 S90 hbond G10 33.5902 SheetConstraint S27 G30 A44 I41 hbond A29 33.5902 SheetConstraint G33 S34 D38 G37 hbond S34 33.5902 SheetConstraint D62 V65 A97 Q94 hbond R63 33.5902 SheetConstraint V65 C68 V95 E92 hbond T66 33.5902 SheetConstraint V65 I67 T72 T70 hbond I67 33.5902 #sheet constraints from conformation 1 T0366-2fe5A-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:2fe5A SheetConstraint T13 K18 Q94 G89 hbond V14 32.7163 SheetConstraint S27 G30 A44 I41 hbond A29 32.7163 SheetConstraint T66 C68 Q94 E92 hbond T66 32.7163 SheetConstraint V65 I67 T72 T70 hbond I67 32.7163 #sheet constraints from conformation 2 T0366-1tp5A-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:1tp5A SheetConstraint E15 K18 E92 G89 hbond M16 32.5741 SheetConstraint S27 G31 A44 P40 hbond A29 32.5741 SheetConstraint G61 I64 G98 V95 hbond R63 32.5741 SheetConstraint V65 I67 T72 T70 hbond I67 32.5741 #sheet constraints from conformation 3 T0366-1g9oA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:1g9oA SheetConstraint L11 K18 V96 G89 hbond R12 31.6457 SheetConstraint S27 G30 A44 I41 hbond A29 31.6457 SheetConstraint D62 V65 A97 Q94 hbond R63 31.6457 SheetConstraint T66 C68 Q94 E92 hbond T66 31.6457 SheetConstraint V65 I67 T72 T70 hbond I67 31.6457 #sheet constraints from conformation 4 T0366-2fcfA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:2fcfA SheetConstraint E15 K17 E92 S90 hbond M16 31.5858 SheetConstraint G25 S27 H47 M45 hbond S27 31.5858 SheetConstraint S27 G30 A44 I41 hbond A29 31.5858 SheetConstraint D62 I64 A97 V95 hbond R63 31.5858 SheetConstraint V65 I67 T72 T70 hbond I67 31.5858 #sheet constraints from conformation 5 T0366-1qavA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:1qavA SheetConstraint L11 K18 V96 G89 hbond R12 31.2414 SheetConstraint S27 G30 A44 I41 hbond A29 31.2414 SheetConstraint D62 V65 A97 Q94 hbond R63 31.2414 SheetConstraint T66 C68 Q94 E92 hbond T66 31.2414 SheetConstraint V65 I67 T72 T70 hbond I67 31.2414 #sheet constraints from conformation 6 T0366-2h2bA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:2h2bA SheetConstraint S8 K18 G99 G89 hbond G10 30.0817 SheetConstraint D62 V65 A97 Q94 hbond R63 30.0817 SheetConstraint V65 C68 V95 E92 hbond T66 30.0817 SheetConstraint V65 I67 T72 T70 hbond I67 30.0817 #sheet constraints from conformation 7 T0366-2h3lA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:2h3lA SheetConstraint S8 R12 G99 V95 hbond G10 28.0914 SheetConstraint E15 K18 E92 G89 hbond M16 28.0914 SheetConstraint S27 G30 A44 I41 hbond A29 28.0914 SheetConstraint D62 I64 A97 V95 hbond R63 28.0914 SheetConstraint V65 I67 T72 T70 hbond I67 28.0914 #sheet constraints from conformation 8 T0366-1wf7A-t2k-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:1wf7A SheetConstraint V14 G19 V95 S90 hbond M16 26.8387 SheetConstraint I28 G30 I43 I41 hbond A29 26.8387 SheetConstraint G31 V32 V39 D38 hbond G31 26.8387 SheetConstraint V65 I67 T72 T70 hbond I67 26.8387 #sheet constraints from conformation 9 T0366-2f0aA-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m:2f0aA SheetConstraint L11 M16 V96 I91 hbond R12 26.6706 SheetConstraint I28 G30 I43 I41 hbond A29 26.6706 SheetConstraint D62 V65 A97 Q94 hbond R63 26.6706 SheetConstraint V65 C68 V95 E92 hbond T66 26.6706 SheetConstraint V65 I67 T72 T70 hbond I67 26.6706 I merged these to get the following constraints for try2: SheetConstraint S8 K18 G99 G89 hbond R12 11 SheetConstraint S27 G31 A44 P40 hbond A29 5 SheetConstraint G33 S34 D38 G37 hbond S34 2 SheetConstraint G61 V65 G98 Q94 hbond R63 5 SheetConstraint V65 C68 V95 E92 hbond T66 4 SheetConstraint V65 I67 T72 T70 hbond I67 3 Mon Jul 24 20:22:21 PDT 2006 Kevin Karplus try2 started from alignments on cheep. When it finishes, we should probably do a polishing run with a similar costfcn (maybe a bit lower on the constraints) to all the existing models. Wed Jul 26 09:57:15 PDT 2006 Kevin Karplus try2-opt2 does not do as well as try1-opt2 with the try2 costfcn, but rosetta likes it better. try3 has been started on shaw as a polishing run starting from all existing models. I'll do try4 with the same cost function, but starting from just decoys/T0366.try2-opt2.gromacs0.repack-nonPC.pdb.gz Wed Jul 26 10:06:48 PDT 2006 Kevin Karplus try4 started on cheep. Wed Jul 26 21:13:55 PDT 2006 Kevin Karplus try3-opt2 scores best with the try3=try4 costfcn, but rosetta likes best try4-opt2.gromacs0.repack-nonPC Wed Jul 26 21:22:18 PDT 2006 Kevin Karplus I'll try making a dimer from try3-opt2. Since there are no notes by Martin anywhere in the README file, I'm assuming that he has not done and does not plan to do any work on this target. In the interest of keepin the records straight, I'm removing his name from this target on the targets list. Wed Jul 26 21:32:51 PDT 2006 Kevin Karplus oops. I have to get the dimeric form of 1g9o from PQS. http://pqs.ebi.ac.uk/pqs-doc/macmol/1g9o.mmol Wed Jul 26 21:43:56 PDT 2006 Kevin Karplus OK, I built dimer/decoys/dimer-try3-1g9oA.pdb.gz but it does not look too promising as a starting point. Perhaps I should look as some of the other multimers, and see if part of one of them looks like a better dimer. Too tired tonight to keep working (it is nearly 1am here in Vermont). Thu Jul 27 05:04:33 PDT 2006 Kevin Karplus Looking at 1tp5.mmol http://pqs.ebi.ac.uk/pqs-doc/macmol/1tp5.mmol It looks like we could try making a dimer from 1tp5A and 1tp5G One possible problem is that dimerization interface appears to be on the C-terminal part of the 1tp5A chain, which is not matched in our target. I set up 1tp5A/T0366-1tp5A.dimer-a2m from T0366-1tp5A-t06-local-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m and make-dimer2.under to use it to create dimer/decoys/dimer-try3-1tp5A.pdb This dimer looks feasible (better than the one made from 1g9o), but there are some exposed isoleucines (I26 and I28) that could pack against each other, I think, so for dimer/try1, I'll try moving them close to each other. Thu Jul 27 05:22:15 PDT 2006 Kevin Karplus dimer/mult1 started on cheep Thu Jul 27 05:28:11 PDT 2006 Kevin Karplus oops--forgot to make decoys/read-pdb.under in the dimer directory, so dimer/mult1 failed. Restarted on cheep. Thu Jul 27 07:31:20 PDT 2006 Kevin Karplus dimer/try1-opt2 reduced the I26/I28 cross-monomer distance to 10.6 (from 15.6). There is still a pretty big void around I26, I28, L24, L84. I think I'll do another dimer run with the constraints off, but with the dry terms turned up. Thu Jul 27 07:40:28 PDT 2006 Kevin Karplus dimer/try2 started on cheep to try to improve packing of dimer/try1-opt2 Fri Jul 28 14:20:58 PDT 2006 Martin Madera b.36 is PDZ-like, currently only one superfamily, PDZ (b.36.1), with four known families. Don't see anything interesting in what Pfam has to say. Multimeric status of 2fne: it has three chains, A, B, C, but they have different lengths (chain A is the shortest, chain C is the longest). Actually the SEQRES records for all three chains are identical, but the ATOM sequences start at different positions at the N-terminus (the C-termini being identical). It looks like the N-terminal His tag / cloning region may be disordered in A and B. Importantly, the C-terminal beta strand forms the edge strand of the beta sheet in the next domain. It looks vaguely like a hexamer to me, but clearly each of the subunits has a slightly different structure. The packing is very loose, the sole connection is provided by the final beta strand. *However*, 2fne is a fragment of a larger multi-domain protein (MPDZ_HUMAN in Uniprot) which has at least 13 PDZ domains in a tandem repeat. So it looks to me like they got the domain boundaries slightly wrong and it happened to form a (slightly distorted) hexamer, but this isn't biologically relevant. 2fe5 is a monomer, but that's because the C-terminal beta strand got chopped off; the structures look very similar to 2fne, so my guess is that the biological domain does have that extra strand. Again it's a fragment (of DLG3_HUMAN), it's the second PDZ domain from a tandem repeat of three (or maybe four, though neither Pfam nor SUPERFAMILY can see the putative third domain). 1uju is an NMR structure, skipped. 2byg is a monomer, similar situation to 2fe5. 1tp5 is actually a hexamer, even though PQS says it's a dodecamer -- if you look at the file PQS provides (saved in martin/1tp5_PQS.pdb.gz) the peptide ligands account for six out of the twelve chains. Note that the ligand acts as the final strand in the beta sheet! So maybe the domain boundaries in 2fne were actually correct. 1tp5 is a trimer of dimers, and the dimer interface is large and quite well packed. This could be a good template! OK, time to get more systematic: all BLAST hits better than 1e-6: PDB PQS ---- --- 2fne - 2byg 1 1uju NMR 2fe5 - 1vj6 NMR 1gm1 NMR 1d5g NMR 1iu0 NMR 1rgr NMR 1kef NMR 1ozi NMR 1x5q NMR 1q7x NMR 2fcf - 1wha NMR 1wfv NMR 1um1 NMR 1qlc NMR 1zok NMR 1qav 2 1z87 NMR 2db5 NMR 1z86 NMR 2pdz NMR 1mfg 2 1wf8 NMR 2h3l 1,1 1i16 NMR So the structures to look at are 1qav and 1mfg, and possibly 2h3l. (I've looked at 2fcf but it seems to be a monomer with lots of chain breaks.) 1qav is a dimer in the PDB file, 1mfg is a monomer, 2h3l is a dimer but the interface is so small that this almost certainly isn't biological (and PQS says they're two monomers). I also got the PQS files for 1qav and 1mfg and saved them in martin/; the file for 1qav agrees with the PDB file. The 1mfg file also agrees, it's a monomer and the dimer is due to the peptide ligand being counted as the second subunit. -------------------------------------------- So far I have found two good interfaces: 1qav martin/1tp5_PQS.pdb.gz (between *I,*J,*E,*F) -------------------------------------------- I think we'll need to look further down the list of BLAST hits: PDB PQS ---- --- 1uez NMR 1n7e 2 1wf7 NMR 1wi2 NMR 1x6d NMR 2daz NMR 1x5n NMR 1um7 NMR 1uf1 NMR 1pdr 6 1xz9 NMR 1uhp NMR 1t2m NMR 1tp5 12 1uew NMR 1be9 6 1x5r NMR 1v62 NMR 2csj NMR 1x45 NMR 1ujd NMR 1ihj 2,2 PDB: 1n7e ... monomer 1pdr ... monomer 1be9 ... monomer with a ligand 1ihj ... two monomers with ligands Looking at the PQS files (saved in martin/), 1n7e really is a dimer 1pdr is a trimer of dimers, but the packing is foamy (because no ligand?) 1be9 is a trimer of dimers, good packing 1ihj are two monomers, each with a ligand 1n7e is a new interface: the C-terminal strand of each domain acts as a ligand for the other domain. And 1be9 looks very similar to 1tp5. So, having looked at all the BLAST structures up to E=0.001 (plus 1tp5), the plausible interfaces are: ----------------------------------------------------------- - 1qav (between *A,*B) - martin/1n7e_PQS.pdb.gz (between *A,*B) - martin/1be9_PQS.pdb.gz (between *K,*C with *L,*D ligands) martin/1tp5_PQS.pdb.gz (between *I,*E with *J,*F ligands) ----------------------------------------------------------- Fri Jul 28 20:25:51 PDT 2006 Kevin Karplus I believe that the dimer interface in 1tp5 that Martin mentions is the same one as I was using (1tp5A, 1tp5G). It is standard for PDZ domains to bind a peptide as the last strand of the beta-sheet---that is how PDZ domains work to selectively bind proteins. Since we were not given the ligand peptide, we have to assume that it is missing. Fri Jul 28 21:57:53 PDT 2006 Martin Madera Yes, 1tp5 A-G is the same as 1tp5 I-E. I think it's the best interface out of the three. The other two require an extra C-terminal strand (or two, for 1qav). The question is, do we have that strand? For 1n7e, the chain ends at S761 and the last residue in the beta sheet is K751, so we need 10 extra residues. In T0366.try1-opt2, the last residue in the beta sheet is A97, the last residue on the chain is V106... well we're one residue short, but it could work. I've just noticed that 1qav actually contains two different proteins and hence is asymmetric. Plus our C-terminus isn't long enough for this mode of binding, so we can forget about this one. My comments on what Kevin has done: First, what's the interface for 1g9o? Ah, it's the same idea as for 1n7e; the last residue in the sheet is V91, the last residue is L99, which is actually one residue shorter than what we have. However the relative orientation of the two domains is different from 1n7e (see the respective PQS files in martin/). -------------------------------------------------------------------- Hmm, we shouldn't ignore the self-binding dimer: we have just the right number of residues on the C-terminus! -------------------------------------------------------------------- Dimer-try3-1g9oA.pdb.gz looks mostly OK, but there are two problems: the C-terminal bit needs to be flipped out of the way (which should remove most of the clashes), and the N-terminal helix needs to be unwound. I'll edit the monomer (try3-opt2) in ProteinShop and then re-run the dimerization script. Sat Jul 29 00:56:40 PDT 2006 Martin Madera I've generated dimer-try3-1g9oA-take2.pdb.gz, which already looks much better, but I didn't get the N-terminal strand quite right. (Difficult to do blind!) I'll give it one more shot. Sat Jul 29 01:40:39 PDT 2006 Martin Madera dimer-try3-1g9oA-take6.pdb.gz is my final attempt -- it's far from perfect, but it should be a good starting point. Sat Jul 29 08:21:31 PDT 2006 Kevin Karplus The dimer costfcns, including unconstrained.costfcn, like dimer-try3-1g90A-take3 best of the 1g9oA dimers. I'll do two polishing runs: one from all the 1g9oA dimers, and one from just take6, which Martin liked best. Sat Jul 29 08:27:28 PDT 2006 Kevin Karplus dimer/try3 (all 1g9oA dimers) and dimer/try4 (dimer-try3-1g9oA-take6 only) started on lopez. Sat Jul 29 09:21:30 PDT 2006 Kevin Karplus There was a typo in dimer/try4.under that caused it *not* to start with take6, and to work from a random model. I killed try4 partway through, and restarted it with the typo fixed. The already-output try4-opt1 was renamed to dimer/decoys/T0366.bad-try4-opt1.pdb.gz and T0366.bad-try4-opt1-scwrl.pdb.gz The try3 run seems to be doing ok, but it ended up preferring to use take2, not take3 or take6. Sat Jul 29 11:12:06 PDT 2006 Kevin Karplus OK, we now have 4 dimers we can submit: 2 based on 1tp5[AG] and two based on 1g9o. The 1tp5 dimers are preferred by both undertaker and rosetta. I could probably improve the 1g9o dimers slightly by adding some inter-monomer hbonds Hbond L36.O D144.N Hbond D38.N L142.O I'll try polishing dimer/try3 and dimer/try4 with this added constraint, then decide whether any of the monomers from the dimer models should be submitted. We have to submit monomers and dimers for T0366. Sat Jul 29 11:28:06 PDT 2006 Kevin Karplus dimer/try5 started on lopez, polishing try3 and try4 with added Hbonds between the monomers. Sat Jul 29 13:52:56 PDT 2006 Kevin Karplus I think we are ready to submit both monomers and dimers. I just need to write up the methods files. Sat Jul 29 16:24:05 PDT 2006 Kevin Karplus Monomers submitted with comment We had excellent fold recognition hits to PDZ domains, including 2fneA as our top hit. Since this was also BLAST's top hit, with an E-value of 2.2e-20, we are quite confident of the fold recognition. The guts of the fold remained essentially unchanged during our optimization, so we are mainly sampling minor variations on the termini and the small loops. We tried making dimers with 1tp5 and 1g9o as templates for the dimerization, and optimized the monomers in the dimeric context. The try3-opt2 monomer that we started with had some conflicts in the termini when superimposed on the 1g9o dimer, so ProteinShop was used to move the termini out of the dimerization interface. We do not believe that the dimerization in the crystal is biologically meaningful, and we did not see much signal for it in the multiple alignments or monomer models. Because there was not much signal to work from, we did not spend a lot of time trying to optimize the dimers. Model 1 is try3-opt2, our best-scoring monomer, the result of two rounds of polishing the automatic with undertaker. Model 2 is try4-opt2.gromacs0.repack-nonPC, the monomer that rosetta liked best of those it repacked sidechains for. It was optimized by undertaker from try2-opt2.gromacs0.repack-nonPC, then reoptimized with gromacs and with sidechains repacked by rosetta. Model 3 is chain A of dimer/try5-opt2, a dimer based on the monomer try3-opt2 superimposed on the 1g9o dimer, then reoptimized by undertaker in the dimer context. Model 4 is chain A of dimer/try2-opt2, a dimer based on the monomer try3-opt2 superimposed on the 1tp5 dimer, then reoptimized by undertaker in the dimer context. Model 5 is sidechain replacement by SCWRL on an alignment to 2fneA. Sat Jul 29 16:35:45 PDT 2006 Kevin Karplus Dimers submitted with comment For T0366, we used only one monomer (try3-opt2) as the basis for our dimer submissions, but we superimposed on two different dimers of PDZ domains: 1tp5 and 1g9o. The superpositions were then optimized by undertaker. For 1g9o dimers, we had to use ProteinShop to modify the termini of the monomer, to avoid bad clashes in the dimerization interface. The 1tp5 dimers scored better with the undertaker costfcns and with rosetta's cost function, when rosetta was used to repack sidechains. Dimer 1 is try2-opt2, our best-scoring with undertaker cost functions, based on 1tp5. Dimer 2 is try1-opt2.unpack.gromacs0.repack-nonPC, rosetta's favorite of the dimers it repacked sidechains for. It is also based on 1tp5, but the undertaker-optimized model (try1-opt2) was reoptimized by gromacs and had sidechains repacked by rosetta. Dimer 3 is try5-opt2, our best scoring dimer based on the 1g9o superposition. We deliberately attempted to introduce some hydrogen bonds between the monomers at L36.O and D38.N. Dimer 4 is is try3-opt2.unpack.gromacs0.repack-nonPC, the 1g9o-based dimer that rosetta liked best. Dimer 5 is try4-opt2, the remaining 1g9o-based dimer, included only because it is slightly different from either try3-opt2 or try5-opt2. ------------------------------------------------------------ Mon Jul 31 09:55:58 PDT 2006 Kevin Karplus To do the refinement they requested, I'll make a dimer from tr366.gromacs0.repack-nonPC using the 1tp5 template, then optimize with undertaker (high clash and break penalties) in the dimer context. It may never get quite as good in undertaker's eyes as our own models, but it will meet the spirit of the experiment. Mon Jul 31 10:05:29 PDT 2006 Kevin Karplus dimer/mult6 started on cheep to do refinement. Mon Jul 31 13:08:26 PDT 2006 Kevin Karplus dimer/try6-opt2 looks ok, but rosetta doesn't like it. For dimer/try7, I'll start from try6-opt2.gromacs0 and try6-opt2.gromacs0.repack-nonPC and optimize with a costfcn that has clashes and breaks turned up. Mon Jul 31 15:36:57 PDT 2006 Kevin Karplus The dimer/try7 run produces rosetta's favorite so far of the dimer models, but there are still a few breaks, so I'll do another polishing run from try7-opt2.gromacs0.repack-nonPC Mon Jul 31 15:46:27 PDT 2006 Kevin Karplus Started dimer/try8 (still with helix and sheet constraints) to polish dimer/try7-opt2.gromacs0.repack-nonPC on cheep. Mon Jul 31 17:45:29 PDT 2006 Kevin Karplus dimer/try8-opt2 slightly outscores dimer/try7-opt2, but rosetta perfers to repack try7-opt2.gromacs0 The differences in the backbone are so tiny that I can barely see them with rasmol. There are a few differences in rotamers, but they are not worth more polishing runs for. We probably *should* do an optimization run for dimer-tr366-1g9oA (which has not been made yet), since that will have a clash and require reshaping one of the termini. Mon Jul 31 23:06:36 PDT 2006 Kevin Karplus I set up make-dimer-refine2.under to make dimer-tr366-1g9oA, and started dimer/mult9 to optimize it using the same costfcn as for dimer/try7. As expected, dimer-tr366-1g9oA has horrible clashes and breaks, but there is a chance that undertaker can fix it by moving the C-terminal bit out of the way, but it is more like to mess up the dimer than fix the tail. If it does mess up the dimer, I'll have to add constraints to hold the dimer together, since copying the tail from a different model would not be in the spirit of the refinement experiment (though I might be tempted to do it anyway). Tue Aug 1 09:54:40 PDT 2006 Kevin Karplus dimer/try10 beats dimer/try9 on the try10 and unconstrained costfcns, but rosetta still prefers decoys/T0366.try7-opt2.unpack.gromacs0.repack-nonPC decoys/T0366.try9-opt2.unpack.gromacs0.repack-nonPC The try10 model moved the N-terminal helix, so non-trivial reshaping of the models is happening. Perhaps I should do another run (like try9 but with the try10 costfcn) from the dimer-tr366-1g9oA model, to see if there is different way to get the C-terminus out of the dimer interface. Tue Aug 1 11:16:23 PDT 2006 Kevin Karplus try11 did not score quite as well as try10, and rosetta doesn't care much for it, but in some ways I like it better, as it comes closer to retaining the original dimer relationship. In fact, try11 comes close to forming an inter-monomer sheet: SheetConstraint L36 P40 P146 L142 hbond D38 10 Tue Aug 1 11:37:09 PDT 2006 Kevin Karplus decoys/try12 started to try to force the inter-monomer sheet. This will probably be my last run. Tue Aug 1 13:09:48 PDT 2006 Kevin Karplus dimer/try12 looks pretty good (better than try10) so I will submit as refinement models: ReadConformPDB T0366.try8-opt2.unpack.pdb chain A # from tr366, 1tp5 ReadConformPDB T0366.try7-opt2.unpack.gromacs0.repack-nonPC.pdb chain A # from tr366, 1tp5 ReadConformPDB T0366.try12-opt2.unpack.gromacs0.repack-nonPC.pdb chain A # from tr366, 1g9o ReadConformPDB T0366.try9-opt2.unpack.pdb chain A # from tr366, 1g9o ReadConformPDB T0366.try10-opt2.unpack.pdb chain A # from tr366, 1g9o There are 3 different N-termini and 3 different C-termini, but the core is essentially the same in all models. Tue Aug 1 14:01:23 PDT 2006 Kevin Karplus Refinement models submitted with comment We started out by cleaning up tr366, running it through gromacs and repacking the sidechains with rosetta. This improved its score with undertaker cost functions without moving anything very far. We then superimposed that model on two different dimers (1tp5 and 1g9o) and did all further optimizations in the dimer context. The original model was compatible with the 1tp5 dimerization, but had serious clashes in the 1g9o dimerization, so that the optimization needed to reshape the C-terminus. Here is the record of optimizations tried: try6-opt2: optimized by undertaker from dimer-tr366-1tp5 try6-opt2.repack-nonPC: sidechains repacked by Rosetta on backbone of try6-opt2 try6-opt2.unpack.gromacs0: optimized by gromacs from try6-opt2 dimer try6-opt2.unpack.gromacs0.repack-nonPC: sidechains repacked by Rosetta on backbone of try6-opt2.unpack.gromacs0 try7-opt2: optimized by undertaker from try6-opt2.unpack.gromacs0 try7-opt2.repack-nonPC: sidechains repacked by Rosetta on backbone of try7-opt2 try7-opt2.unpack.gromacs0: optimized by gromacs from try7-opt2 dimer try7-opt2.unpack.gromacs0.repack-nonPC: sidechains repacked by Rosetta on backbone of try7-opt2.unpack.gromacs0 try8-opt2: optimized by undertaker from try7-opt2.unpack.gromacs0.repack-nonPC try8-opt2.repack-nonPC: sidechains repacked by Rosetta on backbone of try8-opt2 try8-opt2.unpack.gromacs0: optimized by gromacs from try8-opt2 dimer try8-opt2.unpack.gromacs0.repack-nonPC: sidechains repacked by Rosetta on backbone of try8-opt2.unpack.gromacs0 try9-opt2: optimized by undertaker from dimer-tr366-1g9o try9-opt2.repack-nonPC: sidechains repacked by Rosetta on backbone of try9-opt2 try9-opt2.unpack.gromacs0: optimized by gromacs from try9-opt2 dimer try9-opt2.unpack.gromacs0.repack-nonPC: sidechains repacked by Rosetta on backbone of try9-opt2.unpack.gromacs0 try10-opt2: optimized by undertaker from try9-opt2.unpack.gromacs0 try10-opt2.repack-nonPC: sidechains repacked by Rosetta on backbone of try10-opt2 try10-opt2.unpack.gromacs0: optimized by gromacs from try10-opt2 dimer try10-opt2.unpack.gromacs0.repack-nonPC: sidechains repacked by Rosetta on backbone of try10-opt2.unpack.gromacs0 try11-opt2: optimized by undertaker from dimer-tr366-1g9o try11-opt2.repack-nonPC: sidechains repacked by Rosetta on backbone of try11-opt2 try11-opt2.unpack.gromacs0: optimized by gromacs from try11-opt2 dimer try11-opt2.unpack.gromacs0.repack-nonPC: sidechains repacked by Rosetta on backbone of try11-opt2.unpack.gromacs0 try12-opt2: optimized by undertaker from try11-opt2 try12-opt2.repack-nonPC: sidechains repacked by Rosetta on backbone of try12-opt2 try12-opt2.unpack.gromacs0: optimized by gromacs from try12-opt2 dimer try12-opt2.unpack.gromacs0.repack-nonPC: sidechains repacked by Rosetta on backbone of try12-opt2.unpack.gromacs0 Two of our models came from the 1tp5 dimerization. We put them first because they are closer to the original that we were refining. Refinement 1 is chain A of try8-opt2, the best-scoring of the models based on 1tp5 dimerization, at least with cost functions with high break penalties. (try7-opt2 does better when clashes are more important.) Refinement 2 is chain A of try7-opt2.unpack.gromacs0.repack-nonPC, the dimer model that rosetta liked best of those it repacked. Three of our models came from the 1g9o dimerization: Refinement 3 is chain A of try12-opt2, which scored the best (in several costfcns) of the 1g9o dimerizations. It may have been helped by almost forming an antiparallel sheet between the two monomers (L36-P40 to P40-L36). Refinement 4 is chain A of try9-opt2.unpack.gromacs0.repack-nonPC. It was the version of the backbone that rosetta most liked repacking among the 1g9o-based dimers. Refinement 5 is chain A of try10-opt2. It did not score quite as well as some of the other 1g9o dimers, but it provides a different N-terminus than any of the others. ----------------------------------------------------------------------