Thu Aug 8 21:51:58 PDT 2002 t0174 9 Aug 2002 Kevin Karplus Documentation says no disulphides and family known from structure, so this sounds like a fold-recognition problem. Target2k finds only the sequence itself, no homologs, so prediction is likely to be poor. Scores with the HMMs look bad also, and this protein is much too long for ab-initio methods. I think we'll probably have to submit junk, unless we get very, very lucky. Probably we should submit 5 models for this one, with quite different templates. Best bet looks like a multi-domain protein with c.1 as one domain---1kkoA looks like a possible template. We may want to try a run with all alignments inserted, then another run with NO alignments inserted, to sample the space more thoroughly. 24 August 2002 Kevin Karplus Several things to do on this one: * assemble the robetta predictions to see what they look like (this will be try2) * run several, possibly arbitrary, subdomains to get local alignments to use. Perhaps 1-200, 101-300, 201-417. 25 August 2002 Kevin Karplus 00:57 Subdomain runs have finished, but I still have to set up "try-alignments" for them and put the alignments into the top-level undertaker script. The current best score (from robetta models) is try2-opt-scwrl. I haven't looked at it yet. 25 August 2002 16:08 try2-opt has some sheet fragments, but is very scattered. Still---that's better than try1! I'll do a run (try3) from scratch using the extra alignments from running arbitrary subdomains (1-200, 101-300, 201-417), in the hopes of picking up some super-secondary structure. After that finishes, I'll have to pick through what we've found and try to put together constraints for assembling sheets. 29 Aug 2002 10:44 Kevin Karplus I think I'll need help on this one! try3-opt looks pretty trashy also. From karplus@bray.cse.ucsc.edu Thu Aug 29 22:48:19 2002 Date: Thu, 29 Aug 2002 22:47:20 -0700 From: Kevin Karplus To: karplus@soe.ucsc.edu, rachelk@soe.ucsc.edu, weber@soe.ucsc.edu, learithe@cats.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu, rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu, oscarhur@soe.ucsc.edu Subject: T0174 I think I'll need help with T0174---another big protein with sheets to assemble somehow. 30 Aug 2002 2:30pm Jenny Draper I tried a quick literature search, since the CASP info page said the protein family was known; I couldn't find the family. T0174, however, is pretty cool -- it is the C. elegans protein "XOL-1", or "XO Lethal". It is one of the key proteins for gender determination in C. elegans! It supposedly acts as a "count" of the ratio of X-chromosomes to autosomes; if this ratio is low, XOL-1 supresses the SDC family of genes (somehow -- if only we knew), causing the worm to become male (note that most of these little guys are hermaphrodites). It's definately a "big deal" protein for worm research! 30 Aug 2002 3:00pm Jenny Draper Looked at 1kk0A. Has two parts: a TIM barrel and a weird X-shaped twisted 2-3strand sheet that fits a helix in it's dimerized form. The current plan: try to form a TIM barrel... 30 Aug 2002 3:30pm Jenny Draper A swissProt search for xol-1 shows an N-terminal domain homology to "Serine proteases, subtilase family, aspartic acid active site" This is a Prosite hit to pattern PS00136 for pattern [STAIV]-x-[LIVMF]-[LIVM]-D-[DSTA]-G-[LIVMFC]-x(2,3)-[DNH]. We should try to find a homolog here... 3 Sept 2002 Jenny Draper Trying a try4, from alignments to 1gci, which is the fssp representative for the "serine protease, subtilase family" domain (scop c.41.1.1) that the first domain of T0174 (residues 1-325) is supposed to be like. 4 Sept 2002 Jenny Draper After many crashes, finally got undertaker to run try4. It is very low-scoring; but it seems to have found a little bit of sheet. Trying to set up more sheet constriants now... 4 Sept 2002 Jenny Draper 1pm Running try5, using all alignments & try4-opt as possible starting conformations, with constraints for a bit more sheet. 5 Sept 2002 Kevin Karplus 10:56 I don't see a lot in the try5-opt-scwrl and try6.2.50 files that scored best with the try5 constraints this morning (score-decoys-try5.rdb). I tweaked the cost parameters and commented out the try5.constraints, With no constraints, try2-opt-scwrl scores best. It has a lot more strand-like stransd than try5-opt-scwrl, and looks like a much better starting place for assembling sheets. I don't think the TIM-barrel conjecture is very likely for this protein---many of the strands are predicted to occur in hairpins. I extracted the beta-bond information from try2-opt and try3-opt, hoping to combine them, but they are massively inconsistent, diagreeing totally on the bonding partners for I13-I16. The try2 ones look more promising, though H80-D84, which comes out as a strand in try3 could be one---it is predicted ot be helix, but very weakly. The "strand" F277-N288 in try3, which makes the largest collection of H-bonds, is pretty strong predicted to contain a helix. Try5 failed because it did not get any fragments from the files, though it claims to have read them. I have no explanation for this behavior, which seems to happen only to Jenny. Correction--she was not re-directing stderr, just stdout, so the echoing of the fragments did not appear in the log file. 5 Sept 2002 Jenny Draper 2:00pm Johnathan is running jtry8 (note the "j"), off of try2-opt with constraints set to bring the hairpins 295-305 and 348-370 into a sheet (something they clearly SHOULD be doing in try2 :) I am writing a constraint file from a combination of the H-bonds in the robetta models, and will then launch try9 from the robetta conformatioons & try10 from a random conformation, both using the robetta constraints. 5 Sept 2002 Jenny Draper 3:30pm Still working on sorting through the constraints, trying to choose registers for strands. 5 Sept 2002 Jenny Draper 4pm Editing the undertaker.script & starting the try9/10 runs. 5 Sept 2002 Jonathan Casper 4:10pm Started a jtry9 run (a little while ago, oops). This has the same starting point and constraints as jtry8, except for a couple of additional constraints. Basically, I'm trying to get all of the first five strands to form a sheet. This takes Jenny's initial all-antiparallel and adds the strands at residues 68 and 113. I only added one constraint for each strand, in the order 10> rvKILGidr << Jenny 29< emYTLVps << Jenny 44> apHTVHmmd << Jenny 67? << Jonathan 110? << Jonathan 5 Sept 2002 Jonathan Casper 7:12pm This is an update from home. I can't see the structures in rasmol from here. As I recall, though, the jtry8 run is doing some good stuff with the later sheet (from the 295 and 348 hairpins), and jtry9.0 has a good initial placement for the first three strands. The third strand is still in a coil right now, but it is much closer to the other two. 5 Sept 2002 Jenny Draper 11:00pm Try9 is new best-scoring using constraints. (Only try2 scores well w/o constraints, still). It finds a lot of hairpins, as one would expect from robetta constraints. :) Try10 (from random conformation) is a mess; it finds a few hairpins, but try9 and/or Johnathan's runs seem to be better to continue work on. Try9-opt-swcrl & try9.5.20 (the two top-scoring) seem to be the same model. they find hairpins and a 3-strand /\/\\/ sheet for G274-G276, G301-A306, & G295-Q299. Try9.4.20, the third-highest scoring, is a lot like try9-opt, but is more blown-apart (it might be easier to get undertaker to pull strands together in this structure -- less stuff to get in the way. I definately like the look of jtry8.4.50, which really has a decent large sheet forming. jtry9.1.50 has lost and coiled up a lot of the strands in jtry8.4.50; it appears to be heading in the wrong direction. I think I'll try an improvement run on try9-opt with sheet-forming constraint guesses as try11, and an improvement run on jtry8.4.50 as try12 tonight. 6 Sept 2002 Jenny Draper 2:30am Try11 is now running, with some attempts at pulling strands together into a few 3- or 4-strand sheets. Will need to look at results to try bigger sheets than this; there are too many options! Try12 (jtry12) is now running off of try8.4.50, with a few more constraints added by me to try to pull the big sheet that's forming more firmly into place. I will now get some sleep. I expect to be in the lab around 12:30, if people want to meet to discuss directions (I have a dentist apptmnt at 11am). 6 Sept 2002 Jenny Draper 9:30am Try11-opt new best-scoring with try11.constraints; try12.4.50 is best with jon-try12.constraints (opt not calculated yet). Neither appear to have done much to put the sheets together. I think jon's tries need H-bond constraints to tie strand352 down; it keeps twisting up. I think my tries need more time to come together, and maybe need to be run off of try9.4.20, which was really blown-apart & hopefully could fit the sheets together without bumping into stuff on the way? Try11-opt has finally made it into the top 15 of best-scoring w/o constraints (the top models are all try2). Since it has a lot of hairpins but almost no sheet & is very loose, it's still not good. I don't have enough time this morning to set up new runs; Jonathan, could you set up a run off of try11 with try11.constraints, a run off of whatever you think is best from your line, and maybe a run off of try9.4.20 with try11.constraints? Feel free to change any of the constraint files along the way, I'm guessing here. 6 Sept 2002 Jonathan Casper 10:06AM I'm starting the runs that Jenny requested. Try13 starts from try11 with try11 constraints, basically we're just giving it more time. I haven't played with the scoring weights, or with the operation pseudocounts, so we'll just see what happens. Try14 will be a run off of try9.4.20 with try11 constraints. Jtry15 will be my attempt to pull my side of things together. I think I'm going to do a little PDB file editing for this one. I like the arrangement of the first three strands in jtry9.0, and I'm hoping I can pull the third strand straight. For the later sheet, I'd like to start working with jtry8.1 or jtry8.3. I think maybe I can pull all of those strands into a sheet, with the 275 strand packing anti-parallel to the 363 strand. 6 Sept 2002 Jonathan Casper 12:43PM Trys 13 and 14 have finished (one super it, 100 gen). People with more experience should probably look at these, but I don't see a lot of improvement :P. Maybe try15 will come up with something, and then we can cut and paste from these three. 6 Sept 2002 Jenny Draper 1:45pm Undertaker seems to be having a hell of a time moving anything in these structures; I'm not sure why, but probably they are too compact & stuff is getting in the way; each time we do a run it just packs everything more instead of pulling together the sheets. OptSubTree, JiggleSubTree, InsertFragment, and InsertSpecificFragment have all been turned way up in these runs; I don't know how to solve the problem?! 6 Sept 2002 Jenny Draper & Johnathan Casper 4:45pm Running try16, which is trying to form more sheets out of try12-opt. NOTE: johnathan wrote a new script to rotate as well as translate a set of residues, which we used in try12-rot5.pdb, which is our base for try16; it's trying to get the wayward "Z" strand 275 packing into the sheet instead of cutting through it at nearly right angles. All the constriants for try16 were changed to H-bond constriants. 6 Sept 2002 Jonathan Casper 5:29pm I'm running try17, which is just like try16 except shorter. Only 5 super-iterations, and 40 generations. This is to try to give Jenny and I an idea of where try16 could be headed. 6 sept Kevin Karplus 18:27 jtry16.0.80 looks promising. 6 Sept 2002 Jonathan Casper 6:34pm Jenny and I realized that try16 and try17 had constraints that placed one strand in the opposite direction from what we wanted. We started try18 and try19 (analagous to try17 and try16, respectively), with the correct constraints. 6 Sept 2002 Jonathan Casper and Jenny Draper 8:19pm Try 18 looks like it started from a random conformation, instead of try12-rot5, which it was supposed to. We're not sure why this is the case. We verified (via the condor scripts) that we did submit the correct script (undertaker_try18.script) to undertaker. We're not sure where try19 is going, so try16.0.80 is still looking like our best shot. Try16.1.80 is doing some dumb things like trying to wrap one strand around another. We did find a few constraint mistakes, so we are correcting those. We're going to try running trys 20 and 21 now (analagous to 18 and 19), and hopefully they'll get somewhere before the submission deadline. Our best candidates for submission right now are try2-opt-scwrl, try12-opt-scwrl, and jtry16.0.80. None of them are great, but they're the best we have right now. Kevin, Jenny says she can submit targets late tonight if you want to go to sleep. However, you should still submit what you like now, so that we have something in just in case. If you want Jenny to send something in later, send her an email and make a note in this README (so that we'll be sure to notice :) ). Date: Fri, 6 Sep 2002 21:00:18 -0700 From: Kevin Karplus To: learithe@soe.ucsc.edu, jcasper@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: T0174 I looked at undertaker-try18.log to see what went wrong. I noticed one type "constraints" for "constraint" in the try16.constraint file, and several constraint errors (E374 doesn't match S374, ...). I've found it useful to do a "make score-decoys.rdb" run after editing a new .constraints file, to look for these sorts of errors. try18 did read in jtry12-rot5, but it got improved more by InsertFragment than by OptSubtree---when you have a big move to do, it helps to turn OptSubtree way up, so that the pieces get oriented and placed before they get distorted out of shape. Are you sure you wanted to start from jtry12-rot5, and not jtry12-rot3, which scores slightly better? Running make score-decoys.rdb now, I see that the typos in the constraints have been fixed. The top-scoring models are all from jtry16 (.2, .1, .0). I'll look at those and at try2-opt-scwrl and try12-opt-scwrl, which you mentioned in the README file. I'll make some notes in the README file, and send of the top 3 or 4 models. If Jenny wants to replace those models with other ones later on, that's fine with me, or she can ask me to do look at them. I'll probably be up late myself, trying to get t0183, t0185, or both sent off. Kevin Date: Fri, 6 Sep 2002 21:12:41 -0700 From: Kevin Karplus To: learithe@soe.ucsc.edu, jcasper@soe.ucsc.edu CC: karplus@soe.ucsc.edu Subject: T0174 I looked at the models scoring near the top. So far, the one I like best is jtry16.2.80. It may be worth adding some helix length constraints, pick residues 7 apart on a helix that has uncoiled and add length constraints such as constraint L95.CA I102.CA 9.5 10.5 11.5 constraint M130.CA C137.CA 9.5 10.5 11.5 constraint E149.CA N156.CA 9.5 10.5 11.5 constraint Q229.CA R236.CA 9.5 10.5 11.5 constraint Q280.CA M287.CA 9.5 10.5 11.5 constraint D315.CA A322.CA 9.5 10.5 11.5 constraint I318.CA I325.CA 9.5 10.5 11.5 (all real constraints, I think) You can do 14-residues apart on really long helices, if they are strongly predicted throughout, but I don't see any of those here. You might also want to turn the break penalty down to about 5 when trying to do massive movements, so that the constraints have more influence than trying ot keep the chain contiguous. ------------------------------------------------------------ I added these contraints to try16.constraints, calling the result try22.constraints. The best score with these constraints (and breaks turned down to 5) is jtry16.2.80. I'll start a try22 run with the helix length constraints to see if it makes any improvement, starting from several of the current best. I'll also set up the best few models to be submitted, so we have SOMETHING in. 22:02 kevin Karplus The try22 is about 1/3 done and seems to have gotten a substantial drop in cost, so I'll wait until it is done (or 11:20, whichever comes first) to submit model 1 (assuming it will be try22-opt). model 1: try22-opt (assuming it looks ok) model 2: jtry16.2.80 model 3: try2-opt-scwrl (best with no constraints) model 4: jtry12.3.50 model 5: try13.0.100.pdb (next best with no constraints) 23:18 although try22-opt scores better than jtry16.2.80, I'm not sure I like it better---I suspect that I disagree with some of the constraints it is trying to optimize. It has wound up a lot of the helices that had gotten disordered in jtry16.2.80, but it has some awful clashes. Oh, what the hell, I'll make it model 1 anyway. None of these models is very protein-like, but we're out of time. 6 Sept 2002 11:40pm Jenny & Johnathan Sounds good. I don't think there's much more we could have done quickly.