Wed May 24 08:59:29 PDT 2006 T0300 Make started Wed May 24 08:59:42 PDT 2006 Running on camano.cse.ucsc.edu Wed May 24 20:28:44 PDT 2006 Kevin Karplus The pdb.blast run finds no really close homologs, and the iterated searches don't end up with any PDB files. The sequence logos for the three iterated searches look identical, though t06 has 24 sequence and t2k has only 22. All the sequences are from herpes viruses, but viruses mutate fast enough that there is some conservation signal. The best hits are very weak (1debA with E-value 0.4), so this is basically ab-initio. I suspect that this is a three-helix bundle. Perhaps Firas could shape it into one with ProteinShop? The strand predicted at the C-terminus confuses me. I'll try a run without the rr constraints and with the phobic_fit turned down and the constraints turned up (to keep the helices from crumpling). Wed May 24 21:32:42 PDT 2006 Kevin Karplus I like the coiled-coil that try2-opt2 creates, except that one of the helices needs a half turn to bury the hydrophobics. Thu May 25 15:46:11 PDT 2006 Kevin Karplus I'd like to see what Firas can do to this one with ProteinShop. Thu May 25 16:05:28 PDT 2006 Firas Khatib This one is tricky, just rotating helix 46-79 doesn't work because then the coil 80-81 is so short that it can't flex without moving the rest of the chain. I might have to trim some residues from the ends of the helix (ie residue 79, this shouldn't be a problem since A79 is not predicted to be Helix in T0300.dssp-ehl2 and even K78 is very weak Helix) For now, I just let 80-102 spin as I rotated the helix and then manually tried to move 80-102 back to where they should be (hopefully Undertaker will bury it back). My first attempt in: ~casp7/T0300/decoys/Proteinshop/ it's called: rotateFromTry2attempt1.renum.pdb and will be inputted for try3 Thu May 25 16:42:45 PDT 200 Firas Khatib try3 is running using rotateFromTry2attempt1 as input I tried Proteinshop again with rotateFromTry2attempt1 as a starting point and will launch try4 using rotateFromTry2attempt2.renum.pdb next as I think the burial is better. Thu May 25 18:42:33 PDT 2006 Kevin Karplus I like try3-opt2 better than try4-opt2. Perhaps the dry8 and dry12 weights need to be increased. Even try3 needs a bit more rotation to better bury the residues. Sat May 27 21:54:42 PDT 2006 Firas Khatib rotateFromTry2attempt2.renum.pdb has more rotation than what was inputted to try3, but it's result was try4... I will take try3 as the starting point and rotate more there. Tue May 30 15:28:36 PDT 2006 Firas Khatib starting with try3-opt2 I tried to rotate helix 46-79 more. we now have: rotateFromTry3attempt1.renum.pdb rotateFromTry3attempt2.renum.pdb rotateFromTry3attempt3.renum.pdb which I will input to Undertaker as try 5, try6, and try7 respectively I am really surprised as to how awful try4 looks, when it used rotateFromTry2attempt2.renum.pdb as input, and the burial is decent in that one. I think Undertaker had trouble with the 2 turns at D40 and L44 in rotateFromTry2attempt2.renum.pdb, causing the entire 46-79 helix to rotate in the wrong direction. Hopefully try5 through try7 will be better (or at least 1 of them). Tue May 30 18:51:05 PDT 2006 Firas Khatib damn... try5-try7 are all much worse! try5-opt2 would look ok if it didn't have a horrible chainbreak at E82 (try5-opt2 does not show up on score-all.try7.pretty, I guess because it wasn't done yet... I ran them all at similar times) Thu Jun 1 17:20:04 PDT 2006 Firas Khatib I noticed an oddity with the M100 outputted from Proteinshop. Atom: SD 859 Group: MET 100 and Atom: CE 860 Group: MET 100 were both nowhere near the protein (and I don't know why) so I fixed rotateFromTry3attempt3.renum.pdb, and called it: rotateFromTry3attempt3.fix.pdb and am now inputting that to Undertaker as try8 Fri Jun 2 11:25:08 PDT 2006 Firas Khatib well, try8 is garbage. I don't get it, try8 and try7 are only different in their inputs, but the only different between their inputs is 2 lines: try8's input pdb file are missing these 2 lines: < ATOM 859 SD MET 100 7.616 -27.670 14.728-99.00 0.00 < ATOM 860 CE MET 100 7.513 -28.478 12.941-99.00 0.00 could that make such a drastic difference? Fri Jun 2 13:21:04 PDT 2006 Firas Khatib try9 was identical to try8, but using UNDERTAKER=/projects/compbio/bin/i686/undertaker instead (just to see if there was a difference) but it is now evident that we need to add constraints if we want it to look like try3 at all Tue Jun 6 14:33:32 PDT 2006 Firas Khatib Do we care about the predicted strand (by ehl2) from L94-V98 that we have as helix? George suggest packing that helix against the other two. So I have tried doing this with Proteinshop and saved it as georgesuggestion.renum.pdb This will be try10 and then I will add constraints as this will surely fly apart. Tue Jun 6 17:21:14 PDT 2006 Firas Khatib I added constraints and am running try11 and try12. try11 is based on rotateFromTry3attempt3.fix.pdb try12 is based on georgesuggestion.renum.pdb Thu Jun 8 14:01:14 PDT 2006 Firas Khatib try12 didn't work and I messed up the constraints on try11, I have fixed those and am re-running it as try13 Thu Jun 8 15:12:52 PDT 2006 Firas Khatib those constraints were still not enough for try13, so I have made VERY STRICT constraints for try14 to keep the coiled-coil structure from try3: Constraint V72.CB M13.CB -10. 5.0 11.0 0.9 Constraint I68.CB V17.CB -10. 5.0 11.0 0.9 Constraint A64.CB L20.CB -10. 5.0 8.0 0.9 Constraint L60.CB E23.CB -10. 5.0 8.0 0.9 Constraint S57.CB L27.CB -10. 4.0 8.0 0.9 Constraint V31.CB I53.CB -10. 4.0 8.0 0.9 Constraint E50.CB S34.CB -10. 4.0 8.0 0.9 Fri Jun 9 13:00:56 PDT 2006 Firas Khatib something is rotten in the state of try14... I think I am doing something wrong as I cannot keep the coiled coil from try3 together! I think I need some professional help on this one. Fri Jun 9 13:41:56 PDT 2006 Kevin Karplus The reason things have been failing is that the "fix" starting point was missing 2 atoms, which caused it *not* to be included in the conformation pool, so all optimizations were starting with a completely random conformation. try15 (on shaw) will try again, scwrling the conformations first to restore the missing sidechain atoms. It looks like it will be working mainly with attempt1, so try16 (on lopez) will use just attempt3. Fri Jun 9 15:56:44 PDT 2006 Kevin Karplus try15-opt2 and try16-opt2 did fix up the damage done by ProteinShop, but may need more precise constraints to pull things into tight fit. try9-opt2 (which has the short helix antiparallel to the middle one, and the N-terminal helix at the other packing angle) is not very good, but scores best with the unconstrained costfcn. Sat Jun 10 05:26:21 PDT 2006 Kevin Karplus The CASP organizers have provided the extra information that T0300 is a dimer, and that they will accept dimeric predictions as a separate target for T0300: > CASP6 participants might remember that in the previous round of our > experiment we asked predictors to submit multi-chain models for a target > that was specified by the the experimentalists as oligomeric. Last time > we got a very sparse response in this category and therefore, results > were not assessed. We decided to give another try to this initiative > this round of CASP. Target T0300 was specified by its supplier as a > dimer. This target is to expire in a week on June 16. Please, start > working on the dimeric model for this target if you plan to submit it > (obviously, this option is in ADDITION, not alternative, to the usual > CASP one-chain prediction). > Format requirements: > 1. Only TS-type predictions will be accepted for oligomers. > 2. Target record should be in the form > TARGET T0300 OLIGOMER > 3. Each monomer should be identified as a separate chain, consecutive > capital letters for chain IDs should be used (in this case A and B). > 4. Residue numbering in each chain starts from 1. > 5. No TER or END records between monomers (chains). > Please, note that we are currently working on implementing details of > this format in our format verificators, so you won't be able to submit > oligomeric prediction over the weekend. We hope to have our system ready > for accepting this kind of predictions on Monday. Sat Jun 10 14:06:46 PDT 2006 Kevin Karplus I started a VAST search Request ID: 1038598168550266886 from try3-opt2, to see if there is a 4-helix bundle that matches this pair for half. If there is, we can use it to set up a dimer for T0300 and try optimizing the dimer. Sat Jun 10 14:15:22 PDT 2006 Kevin Karplus "VAST did not find any structure neighbors." I think we'll have to find our OWN 4-helix bundle to align to. Either that, or we can make a dimer in ProteinShop by eye. Perhaps 1jm0A+1jm0B? Alignment might be difficult. Sat Jun 10 14:52:42 PDT 2006 Kevin Karplus I tried making a dimer from a rather arbitrary alignment to 1jm0A (see 1jm0A/T0300-1jm0A.dimer-a2m based on T0300-1jm0A-t2k-global-str2+near-backbone-11-0.8+0.6+0.8-adpstyle5.a2m). I am optimizing it (with no constraints between the monomers) as dimer/try1 on shaw. Sat Jun 10 15:51:48 PDT 2006 Kevin Karplus The dimer/try1-opt1 model is too loosely packed in the dimerization interface. We can try tightening it by upping dry weights and even phobic_fit, but I think that we'll need to add some constraints. Residue-residue constraints for something near 0 or 80 and something near 40 could be a cross-dimer constraint (add 102 to one of the residue numbers to get the corresponding residue in the other monomer). M13-V31, I53-V31, M13-S51 are possibles from rr.constraints, but only I53-M115=M13B seems to have a feasible geometry. Sat Jun 10 17:14:48 PDT 2006 Kevin Karplus dimer/try1-opt2 is a bit loose, but not, as I first thought, because undertaker made it loose. Rather, the looseness comes from the initial dimer built from 1jm0[AB]. Perhaps we need to find a tighter bundle to align to? Or a better alignment to make the dimer? We might also want to start with a different monomer than try3-opt2. We might want to have *more* buried residues exposed on the dimerization interface. Mon Jun 12 14:18:47 PDT 2006 Firas Khatib It is interesting that if you look at dimer/try1-opt2 or try3-opt2 with the "burial" script, almost the entire helix from 5-37 is purple, whereas the "near" script has 4 very buried residues (red) that are only green with the burial script. Yet both scripts agree that the C-terminal helices have buried residues. Could it be that the first helix isn't on the dimerization interface but residues 45-102 are? Mon Jun 12 16:04:13 PDT 2006 Firas Khatib I rotated the N-terminal helix to pack against the dimer interface better. This is: rotateNterm2FromTry3.renum.pdb and will be run as try2 in the dimer directory using dimer-rotateNterm2.pdb as input (on lopez) Mon Jun 12 18:03:55 PDT 2006 Firas Khatib I tried everything to flip the dimer interface (in case we want to try that) using Proteinshop, but to no avail. The closest I could get was: Proteinshop/dimer-flip180.fail.renum.pdb I also tried inserting a GLY between the two dimers (hoping to be able to bend that in half to get them closer) but Proteinshop crashes will my different attempts. Tue Jun 13 16:59:08 PDT 2006 Firas Khatib I was able to flip the dimer interface by 180 degrees! dimer.flipped180.renum.pdb I will use this as try3 and explain in the main README file how I was able to do this! Wed Jun 14 15:09:18 PDT 2006 Firas Khatib Looking at score-all.unconstrained.pretty, try3-opt2 is very close to try1-opt2. More work needs to be done, but both orientations of the dimer interface seem valid. We will probably want to submit the two different orientations. Wed Jun 14 15:13:05 PDT 2006 Kevin Karplus I agree. We need to move the try3-opt2 pairs a little, since the are a bit far apart at the "open" end of the helices (not the middle, but the N & C termini). We might also want to make 2 bundles with extra helix capping the other monomer. Wed Jun 14 15:26:09 PDT 2006 Firas Khatib It seems that try1 and try2's orientation is better than try3's based on rr between the 2 dimers (the reds are nowhere near each other in try3-opt2.unpack) I will try using try1's orientation but use the other side of the bundle as the dimer interface by rotating the helix in Proteinshop. Wed Jun 14 15:45:00 PDT 2006 Kevin Karplus I have started dimer/try4 on camano (no constraints) to try to get try3 to pack tighter. Wed Jun 14 15:48:16 PDT 2006 Kevin Karplus I have started dimer/try5 on camano (same costfcn as try4 constraints) to try to get try1 and try2 to pack tighter. Wed Jun 14 16:03:38 PDT 2006 Firas Khatib I am going to actually start from the non-dimer run we had started with (before I had ever rotated any helices) to do the minimal Proteinshop damage. so this will be using: casp7/T0300/decoys/T0300.try2-opt2.pdb Wed Jun 14 16:12:42 PDT 2006 Kevin Karplus try6 started on orcas to optimize dimer/try2-opt2 with inter-monomer rr constraints. (I excluded try1-opt2, although it scores better than try2-opt2, because it doesn't match the rr constraints as well.) try7 started on orcas to optimize dimer/try1-opt2 with inter-monomer rr constraints. The try7 costfcn is the same as the try6 costfcn. Wed Jun 14 18:16:59 PDT 2006 Firas Khatib Well, moving the dimer interface to the opposite end was easier said than done: they don't fit perfectly like LEGOS, so there are a few clashes: dimer/decoys/dimer.rot5.renum.pdb but hopefully Undertaker will be able to handle them without blowing up! I will run this as try8 with dimer.rot5.renum.pdb as input. Wed Jun 14 22:06:08 PDT 2006 Kevin Karplus I'll look at how everything scores and pick 3 dimers to submit---the closest packing we get in each of the three orientations we tried (maybe with an extra for the most likely orientation). anti-parallel dimer: try1-opt2, try2-opt2, try5-opt2, try6-opt2, try7-opt2 parallel dimer: try3-opt2, try4-opt2 interlocking dimer: try8-opt2 try7.costfcn favors try4-opt2, try3-opt2, try7-opt2 try5.costfcn favors try5-opt2, try7-opt2, try1-opt2 try8.costfcn favors try5-opt2, try7-opt2, try1-opt2 rosetta favors try4-opt2 I'll submit as monomors decoys/T0300.try3-opt2.pdb.gz decoys/T0300.try2-opt2.pdb.gz dimer/decoys/T0300.try4-opt2.pdb.gz dimer/decoys/T0300.try8-opt2.pdb.gz decoys/T0300.try15-opt2.pdb.gz to get as much diversity for the C-terminal end as I can. Wed Jun 14 22:48:26 PDT 2006 Kevin Karplus OK. monomers submitted. Now I have to figure out how to submit dimers. I'll submit dimer/decoys/T0300.try5-opt2.unpack.pdb dimer/decoys/T0300.try4-opt2.unpack.pdb dimer/decoys/T0300.try7-opt2.unpack.pdb dimer/decoys/T0300.try8-opt2.unpack.pdb dimer/decoys/T0300.try6-opt2.unpack.pdb Wed Jun 14 23:41:48 PDT 2006 Kevin Karplus OK, dimers submitted. Note that the submission method did not use best-models.pdb.gz, since that had problems with losing the chain ids. Instead, I called pdb2casp separately for each dimer model, and used sed to fix the header. Tue Mar 20 09:37:44 PDT 2007 Kevin Karplus On the official domain for evaluation, our best model try10-opt2.repack-nonPC, is pretty good--only one server model beats it (Pmodeller6_TS1), though a couple of the Zhang-Server models beat it if they are run through SCWRL. Our best submitted model is model4 (dimer/try8-opt2) followed by model3 (dimer/try7-opt2) and model1 (dimer/try5-opt2). Three server TS1 models beat our model1: Pmodeller6, Zhang-Server, PROTINFO. xn