Wed Jul 12 13:59:00 PDT 2006 T0373 Make started Wed Jul 12 14:00:39 PDT 2006 Running on abyss.cse.ucsc.edu Thu Jul 13 11:13:41 PDT 2006 Kevin Karplus Although blast has only weak hits in pdb (2ethA, 32% over 73 residues, E-value 0.002), the HMMs get strong hits to a.4.5.28 domains. 2ethA does score well, but 2a61A (blast's 3rd choice) scores better. In the try1-opt2 model, the C-terminal helix seems to have been moved inward. We will almost certainly have to optimize as a dimer, as the "hole" is filled by a helix of the other monomer. We might want to make a monomer by adding some "scaffolding" to the constraints: 2ethA 2a61A A142.CB R108.CB 20.5 20.4 Q143.CB R46.CB 16.2 16.7 T139.CB R21.CB 24.0 17.4 poor choice (2ethA different) T139.CB A109.CB 25.7 21.5 T139.CB R61.CB --- 27.6 missing from 2ethA alignment T139.CB A59.CB 24.5 24.0 These should hold the helices out enough to make a monomer that can be dimerized (based on 2a61[AB]) There needs to be a better protocol for this sort of dimer, where the fold is really only compact in the multimeric state. Thu Jul 13 11:48:00 PDT 2006 Kevin Karplus I made a try2 costfcn with some scaffolding constraints. I also took helix and sheet constraints from try1-opt2, and helix and strand constrands from dssp-ehl2. I strengthened the C-terminal helix constraint since the helix had gotten broken in try1. try2 started on abyss. Thu Jul 13 15:04:46 PDT 2006 Kevin Karplus Both rosetta and undertaker like try2-opt2 better than try1-opt2, but the "scaffolding" failed to hold the center open. Undertaker just chose 1r7jA as the template instead of 1bjaA. We may want to do one of these "closed" models, but our primary submission should be an open conformation suitable for dimerizing like the top hits. So what would be the cleanest way to ensure getting the "open" conformation? Sun Jul 16 15:30:37 PDT 2006 Kevin Karplus One way might be to superimpose K110-R117 of try2-opt2 on the first alignment, and (separately) S126-E145 on the same alignment. Then I can take the two superpositions and make a chimera. After dimerizing the chimera, I can optimize to close up any bad breaks. I'll try doing the superpositions with make-2super.under Nope---that's not going to work, because undertaker will make the two different superpositions be in different reference frames. Ah, but I can use the same technique as is used for making dimers---having two different alignments in the same file. Sun Jul 16 15:48:01 PDT 2006 Kevin Karplus I put the two alignments into 2a61A/T0373-2a61A.split-a2m, aligning only the two helices of interest. The script make-2super.under Sun Jul 16 15:55:41 PDT 2006 Kevin Karplus I made decoys/chimera-2-2a61A by crossing over between the two superpositions between M119 ahd H120. Sun Jul 16 15:57:33 PDT 2006 Kevin Karplus There is a very bad collision between the N-terminus and C-terminus in chimera-2-2a61A. Back to the drawing board. It looks like try1-opt2 would avoid the conflict, so let's try again with make-1super.under. Sun Jul 16 16:04:30 PDT 2006 Kevin Karplus OK, decoys/chimera-1-2a61A.pdb.gz made, crossing over between H120 and A121. Now I need to dimerize it. Sun Jul 16 16:17:48 PDT 2006 Kevin Karplus I made the dimer with the make-dimer1.under, basing it on 2a61[AB], but there are still some bad clashes. I'll probably want to make other dimers based on other templates. First I'll try optimizing the dimer I made, to see if that cleans things up or blows up. Sun Jul 16 17:08:55 PDT 2006 Kevin Karplus Well dimer/try1 didn't blow up, though it is a little bit foamy. The N-terminal helix needs work---it covers some of the polar C-terminal residues, while V40, L41, and I44 remain exposed. I think I need to swing it around so that L24 is near L41. Sat Jul 29 19:17:23 PDT 2006 Kevin Karplus I'll try making dimers from 2fbh[AB] http://pqs.ebi.ac.uk/pqs-doc/macmol/2fbh.mmol 1lnw[AB] from PDB 1jgs[AB] http://pqs.ebi.ac.uk/pqs-doc/macmol/1jgs.mmol Sat Jul 29 19:33:55 PDT 2006 Kevin Karplus OK, I fetched the two mmol files from PQS and used make-dimer2.under to construct the dimers (using dimer/try1-opt2 chain A as the monomer). Of the forced dimers with this monomer, dimer-1-1jgsA seems to score the best, and dimer-1-2a61A the worse (so maybe 2a61 was not the best template for dimerizing). I should probably do a polishing run for each of the three new dimers that score ok. Sat Jul 29 19:45:05 PDT 2006 Kevin Karplus dimer/try2 for 1jgs started on lopez dimer/try3 for 1lnw started on lopez dimer/try4 for 2fbh started on shaw Sat Jul 29 20:59:39 PDT 2006 Kevin Karplus The new best-scoring dimer is dimer/try4-opt2 Rosetta likes best try4-opt2.unpack-gromacs0.repack-nonPC Unfortunately, I think that dimer/try2, try3, and try4 are all total crap. I went back and looked at the dimer-1-*.pdb files---they're all crap: even the 2a61A one. What a waste of time---I should have looked at them *before* trying to optimize them. I'll need to increase phobic_fit by quite a bit to separate the good models from the the trash. But first, I have to figure out why the superpositions failed so badly. Should I be building the dimers based just on E99-A129? The dimerization seems to be different in the different templates. We also don't seem to get a compact globular protein. At any rate, 2fbh seems to have a spiral groove with a lot of positive charges, so is probably nucleotide binding. (It is "transcriptional regulator PA3341", so is almost certainly DNA-binding.) To get the appropriate spiral groove may be difficult---it may take a lot of scaffolding to mimic the missing DNA. Perhaps holding the R23,R25,R26 clusters the right distance apart would help. Figuring out where the conserved L41 and I44 are buried might also help. Sun Jul 30 13:48:10 PDT 2006 Kevin Karplus All the hits for T0373 are winged-helix structures, in particular MarR-like transcriptional regulators, about which SCOP says # Class: All alpha proteins [46456] # Fold: DNA/RNA-binding 3-helical bundle [46688] core: 3-helices; bundle, closed or partly opened, right-handed twist; up-and down # Superfamily: "Winged helix" DNA-binding domain [46785] contains a small beta-sheet (wing) # Family: MarR-like transcriptional regulators [63379] The N- and C-terminal helical extensions to the common fold form the dimer interface. So we need to have both the N-terminal and the C-terminal helices properly positioned to make the dimer interface. This may be easiest if we could align the whole dimer to a template. Actually, 1p4xA may provide that opportunity: "Staphylococcal accessory regulator A homolog, SarS [88977] duplication: tandem repeat of two SarR-like domains assembled together as in the SarR dimer" I'll try doing a full make on the dimer directory, so that we can have dimer-length alignments. Of course, if there is a somewhat different dimerization than any of the tandem repeats, we won't pick it up this way. Still, it is likely to result in better models than I'd get by trying to create monomers. I really need a way to impose dimer alignments (including the inter-monomer transform) onto a model, so that we could have undertaker explore the possible dimers. One possibility is to take the multimer files (from PDB or PQS's mmol files), and to have undertaker read them running the chains together, then provide an alignment of the double-length target to the double-length template sequence. Sun Jul 30 14:45:54 PDT 2006 Kevin Karplus The dimer HMMs are not going to be very good---all the conservation is appearing in the first copy of the sequence---the second copy is getting almost no sequences aligning. The t04 alignments show a small amount of conservation in the second copy, so may have a few sequences aligning, but the conservation pattern is not a good match to that in the first copy. Sun Jul 30 15:07:10 PDT 2006 Kevin Karplus I built 4 "double" alignments by merging the A and B chains of the 4 dimer-a2m files. dimer/try5 will try inserting them into a random conformation, but still using the constraints and fragments from the monomer run. Sun Jul 30 15:27:38 PDT 2006 Kevin Karplus Starting dimer/try5 again, because the first run included read-pdb.under, causing undertaker to work on polishing existing (bad) models, rather than fixing up the inserted alignment models. The 1lnw multi-chain template is not being aligned well, since it is more than 2 monomers long. I should make a dimer-only model file and use it instead. Oh, well, I'll try that for try6. Sun Jul 30 17:37:12 PDT 2006 Kevin Karplus I'm not real impressed with dimer/try5 The C-terminal helix looks ok, but the N-terminal one has its buried face pointing out!! For try6 I'll try to increase the dry weights and maybe even phobic fit, and see if that helps. I also put a truncated version of 1lnw.pdb into 1lnwA/1lnwAB.pdb.gz, and will use that for dimer/try6 Sun Jul 30 17:50:57 PDT 2006 Kevin Karplus The dimer/try6 costfcn likes try1 best of the existing dimers, mainly because of phobic fit. dimer/try6 started on abyss. Sun Jul 30 19:41:03 PDT 2006 Kevin Karplus dimer/try6-opt2 begins to look like the model we were trying to get. It scores best with the dimer/try6 costfcn, and second best (close behind try1) with unconstrained costfcn, but rosetta doesn't like it nearly as well as the other dimers (probably because rosetta prefers foam to clashes). It looks like dimer/try6 was based on the 2a61 dimer with a 1lnwA monomer. For try7, I'll use a similar costfcn (with phobic_fit turned down a little and breaks and clashes up a bit), but I'll include the alignments for the double-length models also. Sun Jul 30 22:55:22 PDT 2006 Kevin Karplus dimer/try7 started on cheep Mon Jul 31 08:13:37 PDT 2006 Kevin Karplus although dimer/try7 is marginally plausible, dimer/try6 is still the best of the dimers we have created. There is pretty good agreement among all our dimer models for G48-L123. dimer/try6 and dimer/try7 have a different orientation for the P32-G48 helix, than the models built from the monomer. The initial helix (M1-A30) has a somewhat different orientation in try6 and try7. The Cterminal helix (E125-P147) has 3 different orientations (the version from monomer, try6 and try7). I should take the try6 and try7 chain A monomers and try making new dimers from the templates for each of them. Mon Jul 31 08:39:34 PDT 2006 Kevin Karplus I've set up make-dimer3.under to make dimers from the monomers. I just have to hope that the superposition concentrates on the common core when doing the superposition, and not on the variant positions of the terminal helices. Hmm---doing another ReadTargetPDB seems to cause undertaker to crash. It looks like the SpecificFragmentLibrary is not being properly cleared. Mon Jul 31 08:52:29 PDT 2006 Kevin Karplus Sure enough, there was no delete_specific_fragment_library() called in delete_target(). I fixed the bug and make-dimer3 seems to run ok. Mon Jul 31 09:08:11 PDT 2006 Kevin Karplus Unfortunately, none of the new dimers seem to have any interaction between the two monomers in the terminal helices, so try6-opt2 is still about the only real dimer we have. Mon Jul 31 17:37:39 PDT 2006 Kevin Karplus For dimer/try8, I ran the try7 script but with the dimer alignment T0373-2a61-twice.a2m (the one that it started from) commented out. I also made a costfcn that looks more like the try6 costfcn than the try7 costfcn. Maybe I'll do another run, starting with just the 1lnw dimer, since the 1lnwA monomer seems to be the one it favors. Mon Jul 31 17:42:22 PDT 2006 Kevin Karplus dimer/try9 started on whidbey Mon Jul 31 20:04:34 PDT 2006 Kevin Karplus dimer/try9-opt2 scores best with the try9 costfcn, then try6, try1, try7, try8 Of these, only try9, try6, and try1 have an attempt at the proper dimerization interface. dimer/try9-opt2 also is the thrd best of the starting points for rosetta repacking, and I really don't like the first 2 (try5 and try4). I don't see much hope of figuring out the dimerization interface, since the templates all seem to have different ones. I'll just submit the monomers from the top 3 dimer runs and the top 2 alignments and give up. Mon Jul 31 20:30:59 PDT 2006 Kevin Karplus Submitted (after a couple of false starts) with comment We had a lot of trouble with T0373, because the terminal helices do not pack compactly in the monomer. We had to do all our optimization in a dimer context, in order to keep the monomer from collapsing. Unfortunately, the top few templates all had different dimerizations, so there was not a simple model to copy. Eventually, we gave up on determining exactly how the winged-helix monomers dimerized, and just submitted chain A from the 3 best dimers created so far, along with models from the top two alignments. There is fair agreement on the core of the model, but the N and C terminal helices are still very much in need of work. Model 1 is chain A of dimer/try9-opt2, optimized from dimeric alignments, stressing 1lnwA and 1s3jA. Model 2 is chain A of dimer/try6-opt2, optimized from dimeric alignments, stressing 2a61A and 1lnwA. Model 3 is chain A of dimer/try1-opt2, optimized from a chimeric monomer (opening up the C-terminal helix) superimposed on 2a61 to form a dimer. Model 4 is sidechain replacement by SCWRL on an alignment to 2a61A. Model 5 is sidechain replacement by SCWRL on an alignment to 2fbhA. ------------------------------------------------------------ Sun Sep 24 14:26:33 PDT 2006 Kevin Karplus Once again the Zhang-Server did well: Zhang-Server_TS5 -143.41 Zhang-Server_TS1 -125.42 Our best submitted: model1 (dimer/try9-opt2) -107.54 Our best generated: dimer/try9-opt2.unpack.gromacs0.repack-nonPC -108.72 Our servers did poorly: SAM_-T02_AL1-scwrl -67.73 SAM_T06_server_TS1 -59.00