Wed May 10 09:37:25 PDT 2006 T0283 Make started Wed May 10 09:37:49 PDT 2006 Running on cheep.cse.ucsc.edu DUE DATE: 21 June 2006 Wed May 10 09:51:16 PDT 2006 Kevin Karplus This protein is BH3930 from Bacillus halodurans (a halophile?). The small number of similar proteins found by t2k and t04 are from other bacteria (mainly Bacillus, but also Oceanobacillus, Desulfitobacterium, Enterococcus, Lactococcus, and Listeria). None of the proteins have annotated functions in their names in the NCBI nr database. The AA-only and 2-track searches do not seem to be coming up with significant hits. This one may require new-fold techniques. It is predicted to be all helical, so the use of sheet constraints is not going to be helpful. There are 4-6 helices, depending on whether the dips in the helical prediction around S10-N12 and K95-W97 represent breaks in the helices or not. Wed May 10 10:30:46 PDT 2006 Kevin Karplus We have some conserved residues in T0283 (12 according to t2k, 6 according to t04): t2k: t04: RK23 RK23 LIVF27 YFL31 YFL31 YFWLM35 ILVMF38 YFN41 WFRYL43 WFYL43 LIVFM62 FLIVY65 G80 GA80 FYL86 FLYI86 LIMVF90 WYFL97 The t04 sequence logo shows weaker, but still interesting, conservation in other places (like P28 and E66). The t2k logo agrees on P28, but shows no conservation at E66. Wed May 10 10:58:25 PDT 2006 Kevin Karplus The template library search seems to be the slow thing for this target. All the target-model work, including alignment building, for t2k and t04 finished in 22 minutes, but the template library scoring for t2k alone took 49 minutes. Make started Wed May 10 12:02:33 PDT 2006 Running on cheep.cse.ucsc.edu Make started Wed May 10 12:05:51 PDT 2006 Running on cheep.cse.ucsc.edu The first run had a few bugs (like the near-backbone-11 network being missing from casp7/networks), so I'm running it again. I've made a number of changes in the casp7/starter-directory/Make.main file to use the new target06 script also, even though the t06 template library is far from complete. Wed May 10 12:16:18 PDT 2006 Kevin Karplus The t06 alignment finds the same 22 sequences as the t04 alignment, though the t2k alignment finds a few more. I believe that the t06 and t04 alignments are the same (except for rearranging the order of the sequences). Make started Wed May 10 12:22:58 PDT 2006 Running on cheep.cse.ucsc.edu Wed May 10 12:26:38 PDT 2006 Kevin Karplus I had to start the make again, because of a minor bug in Make.main that caused the t06 target-model scores not to be computed. None of the hits for the sequence are very strong. The best two are 1tffA (E-value 4.3) and 1b8zA (E-value 11.9). At least 1b8zA is an all-alpha fold! Make started Wed May 10 13:18:47 PDT 2006 Running on cheep.cse.ucsc.edu Another minor bug---some of the alignments that top_reported alignments was looking for were never created, and the pick_alignments script needed to be modified slightly to handle this case. Wed May 10 14:03:20 PDT 2006 Kevin Karplus try1 has finally started running. None of the top alignments generated any sheet constraints (good, since we were predicting an all-helical protein). The coloring script for rr makes the residue contact predictions a bit hard to visualize. It just breaks them into three groups: an N-terminal cluster, a mid-chain cluster, and a C-terminal cluster. These may help a little with packing the helices, but we may want to try hand-packing them. The try1-opt1 prediction has 6 helices, but the breaks don't all correspond to the predicted breaks from the neural nets. The conserved G80 does not break a helix, though it probably should. Wed May 10 14:13:21 PDT 2006 Kevin Karplus The try1-opt2 model is very similar to the try-opt1 model. It might be worthwhile to try to make F65 and F86 come close, and L62 and L90. There is also a helix-break at Y35 that doesn't belong. Both the dips in helix prediction: S10-N12 and K95-W97 are near helix breaks, with L11-D13 being a turn and M91-T94 being a rather awkward turn. I think that stronger helix constraints on the well-predicted helices would be a help for the next round. We may want to increase weights for "dry" functions to increase burial, as there is a pretty large patch of residues that are exposed, but predicted to be buried: M42, W43, L90, F86, I61, F65, L62, F58, I57, T54, T49. (Although this could be a dimeric interface, I think that repacking the helices to make a monomer is more promising.) Eventually, we may want to try moving the helices around by hand, since this may well be a helix bundle, which the current prediction is not (quite). Wed May 10 15:54:01 PDT 2006 Kevin Karplus try2-opt2 looks worse than try1-opt2. Even the tr2 cost function thinks so! G80 is nicely in a turn, but G68 and G72 are not. Neither are G59 and G60. Of course, we *do* have a helic constraint that tells undertaker that W51-A70 are a single helix. The constraints I added Constraint F65.CB F86.CB -10. 7.0 14.0 0.875476 Constraint L62.CB L90.CB -10. 7.0 14.0 0.876761 seem to have had no effect, so perhaps I should make them stronger. The constraint Constraint M42.CB I57.CB -10. 7.0 14.0 0.888313 seems to have had a large effect. For try3, I'll put back in all the rr constraints, but add much more weight to the F65-F86 and L62-K90 constraints. I'll also increase the weight of the alpha cost functions also, to try to get the 3-10 helix to form. I might need to add a 3-10 helix constraint to undertaker code. Make started Wed May 10 16:07:33 PDT 2006 Running on cheep.cse.ucsc.edu I'm doing a make with REDO_SEARCHES=1 (after updating the t06 template library a little), to see if that feature is working. There are a couple of bugs in it, so I'll have to do it again. In any event, I'll want to redo it in a few days when the t06 library is more complete---there are currently only about 1000 proteins in the t06 library and they are not a representative subset. Make started Wed May 10 16:16:08 PDT 2006 Running on cheep.cse.ucsc.edu OK, this time all the t06 searches were redone, both the target-model searches and the template-model searches. Oops, this is causing the make to redo try1-opt2. (I moved the old try1-opt2 to decoys/first-try/) Make started Wed May 10 17:11:13 PDT 2006 Running on cheep.cse.ucsc.edu I started another run with n_notor2 alphabet included in the template library. Unfortunately, I still had REDO_SEARCHES=1 turned on, so all the searches will be redone. This will take over an hour, since the t2k and t04 template-library scoring are so slow. Make started Wed May 10 17:28:36 PDT 2006 Running on cheep.cse.ucsc.edu Actually, the n_notor2 stuff is not currently acceptable to hmmscore and makelogo, so I won't be able to use it until I get updated versions of the SAM software. I'll have to turn the n_notor2 stuff off for now. I do, however have a 3-10 prediction from the n_notor2 files: Hbond L27.N A24.O 0.54 Hbond G72.N A69.O 0.49 Hbond G72.N G68.O 0.31 # Hbond S10.N M7.O 0.32 alpha helix stronger # Hbond Y31.N P28.O 0.31 alpha helix stronger Make started Wed May 10 17:49:31 PDT 2006 Running on cheep.cse.ucsc.edu I'll generate the o_notor2 predictions also, to look for other Hbonds. Make started Wed May 10 17:51:27 PDT 2006 Running on cheep.cse.ucsc.edu Make started Wed May 10 17:54:47 PDT 2006 Running on cheep.cse.ucsc.edu Make started Wed May 10 17:56:08 PDT 2006 Running on cheep.cse.ucsc.edu The o_notor2 predictions give me the following 3-10 predictions: Hbond A24.O L27.N 0.52 Hbond A69.O G72.N 0.75 Hbond A70.O K73.N 0.34 Hbond M91.O T94.N 0.30 I'll put the 3-10/turn Hbond predictions into try3.costfcn, and run try3 once the current redo of try1 is finished. Wed May 10 19:46:00 PDT 2006 Kevin Karplus I like the new try1-opt2 model, better than the old one and better than either try2-opt2 or try3-opt2. The new try1-opt2 is almost forming a 6-helix bundle: 345 126 Tucking in the C-terminal helix would probably fix it. This might be a good test case for using Protein Shop to try moving helices around. First, though, I'll try a cost function like the try3 one, but which like thr try1-opt2 model better. To do this, I'll up the dry weights a litte, and cut back the pred_alpha weights a lot. Wed May 10 22:41:53 PDT 2006 Kevin Karplus try4-opt2 is beginning to look a bit like the 6-helix bundle I was expecting. I would still like to slide F65 and L62 down a bit to bury them better, perhaps by breaking the helix at G80. Fri May 12 17:41:41 PDT 2006 Kevin Karplus I looked at the robetta models. None of them look great, but there are some parts of the robetta-model3 and robetta-model5 that look good near the G80 conserved residue. I was also wondering if R23 should Hbond with Y31 and Y35. Mon May 15 20:19:24 PDT 2006 Firas Khatib Proteinshop attempts: I took the try4-opt2 model and attempted to slide F65 and L62 down a bit, by sliding down the entire helix from 49-70. I broke the helix at G80 and then tried to rotate all the helices to optimize burial. All my attempts are in /pcep/casp7/T0283/decoys/Protoshop with the latest model (the best IMO) being: g80helix.slideMore2rotateV3.renumb.pdb Proteinshop bugs: I noticed that when you save a PDB file in Proteinshop it renumbers it. This is extremely annoying so I had to run a script after every output from Proteinshop. I called this script renumberChain.pl and it is in the same directory. This might be due to the old version of Proteinshop that I am using: /projects/paro/src/ProteinShop2.x-bin-Linux9.0/ProteinShop I am meeting with Greg to take my modified models and run Undertaker on them to relax them a bit. Tue May 16 16:43:44 PDT 2006 Firas Khatib I used Proteinshop to try and bond R23 with Y31 & Y35: Starting from g80helix.slideMore2rotateV3.renumb.pdb I tried to rotate the helix with Y31 and Y35 so that it would be closer to R23. Just rotating that helix is in: g80helix.sM2rotV3.rotateMoreY31.renum.pdb I then wanted to rotate the helix containing R23, but there wasn't enough coil for it to rotate properly. I then noticed that Leu27, which is at the end of the helix is in fact not predicted to be in a Helix in: T0283.dssp-ehl2-logo.pdf By making L27 a coiled region I was able to rotate the helix with R23 and that is in: g80h.V3.Y31.rotR23andEnd.renum.pdb I then actually moved the 2 helices (res1-26) a bit: g80h.V3.Y31.rotR23andMoveEnd1.renum.pdb and then moved them again a lot more: g80h.V3.Y31.rotR23andMoveEnd2.renum.pdb At this point I notice a problem in the helix from S10-A26. There are a bunch of predicted blue (fairly exposed residues) that all lie on the same side: K18/A19/A22/K25 according to near-backbone11 with R15 being purple on that same side. However, N12,D13,E16 are all blue or purple yet on the other side of the helix. Perhaps that end needs to point into solution? I am meeting again with Greg to take my modified models and run Undertaker on them to relax them a bit (1st meeting was during the power outage) Tue May 16 18:28:22 PDT 2006 Firas Khatib I didn't like the fact that just renaming the G80 Helix region to coil doesn't actually make it look different than a helix. In Proteinshop, you cannot modify coils without grabbing the secondary structures on either side of the coils and moving those. Since I wanted to move the coil (but leave everything else as is) this doesn't help. I noticed that you can select a residue in a coil and hit "randomize angles" and horrible stuff will occur (as one would expect). I did this a few times to try to get rid of the helix-shape of residues 75-79, by randomizing the angles of the L78. I then manually moved the other helices back into place to try to get the six-helix bundle shape back. My first attempt with this method is: g80.randomPerturbOfG80coil.renum.pdb If anyone finds a better way to modify coils that would be great! The only other method I can think of is to call it a beta-strand and then change the "beta strand shape adjustment", but this seems very silly. Wed May 17 17:30:22 PDT 2006 Kevin Karplus One possibility migh be to leave V75-L78 as a helix, but move that helix with T79 and G80 as coil. The Robetta models may give some ideas on this. Fri May 19 13:44:50 PDT 2006 Firas Khatib try5 took in g80.randomPerturbOfG80coil.renum.pdb but the proteinshop model had so many clashes that the only way to get rid of them was to pull the helices apart and it was no longer a 6 helix bundle. Tried using gromacs as well, still too many clashes... I am going to try incremental steps with proteinshop Fri May 19 15:39:11 PDT 2006 Firas Khatib I used Proteinshop just to make a coil from 79-80 and then cranked the small helix onto its side. Saved it as g79-80coil.try1.renum.pdb Running undertaker on it as try6 Mon May 22 16:02:14 PDT 2006 Kevin Karplus I downloaded the server models for T0283 and scored them with our cost functions. None of them scored as well as our models, (except SAM_T06_server_TS1, which is a similar 6-helical bundle). I looked at what Firas got for try6-opt2, and it is starting to look pretty good. Rotating the helix E81-K92 about 1/6 turn along the screw axis (so sliding and rotating) might help the burial a bit, and breaking the chain between T79 and G80 might help undertaker refill that gap with better fragments. Undertaker doesn't currrently have an operator for inserting gaps, but I think I should add one for manual use through ProteinShop. Tue May 23 15:58:14 PDT 2006 Firas Khatib try6-opt2 certainly scored better than its starting point: g79-80coil.try1 I submitted it to VAST in case there is a better structural alignment to a 6-Helix bundle. I will try rotating the E81-K92 helix for try7 before trying to align F65 with F86 better. Tue May 23 16:35:19 PDT 2006 Firas Khatib The results from VAST are here: http://tinyurl.com/jxy9p They do seem to align to different proteins than the ones that were reported in T0283.top_reported_alignments. I will look into this further. Wed May 24 15:25:50 PDT 2006 Firas Khatib I rotated the helix E81-K92 in Proteinshop and saved it as: helix80-93rotateFromTry6.renum.pdb in /decoys/Proteinshop I then notcied that starting with g79-80coil.try1.renum.pdb (where the helix was from 81-93 and I had made 79-80 coil) as input, Undertaker made residue G80 into a helix (in terms of what Proteinshop and rasmol believe) whereas we wanted a coil there (hence the idea of a break in the chain there) I changed G80 back to coil AGAIN, and then also changed D93 to coil, since T0283.t06.dssp-ehl2-logo.pdf doesn't have this residue highly predicted as helix, and Kevin mentioned E81-K92 as the helix. I then slid the helix a bit (since I already rotated it) and saved this as: helix81-92rotateFromTry6.renum.pdb helix80-93rotateFromTry6.renum.pdb has the 2 helices a bit more packed to the core of the protein and will be try7 helix81-92rotateFromTry6.renum.pdb is a little farther (in case the Proteinshop made it too packed and will be try8 I will run both these using Undertaker before adding in the rr constraints that we had turned off. Wed May 24 17:35:06 PDT 2006 Firas Khatib running undertaker on helix80-93rotateFromTry6.renum.pdb as try7, turned off this constraint: Constraint L62 L90, because it doesn't seem possible at this point. Next run (after try8 on helix81-92) I want to add this constraint in: Constraint I61.CB V83.CB -10. 7.0 14.0 0.628027969049 Thu May 25 14:05:14 PDT 2006 Firas Khatib so try8-opt2 and try7-opt2 both score better than try4-opt2, which is good (since that was the starting point of the ProteinShop process). I am now going to put in George's rr constraints and put the L62 L90 constraint back in (I was mistaken in thinking that it was too far (currently 11 A apart) try9 will be based on try7 with only the rr constraints added. Thu May 25 14:52:21 PDT 2006 Kevin Karplus I rather like try7-opt2, the packing is starting to look good, and the only improperly exposed residues seem to be the ones around I61, F65, and V75. If we could slide the part before G47 relative to the part after G47, by about one turn of the helix, so that Y31 lines up more with F65 than with I61, we might be able to get better burial. Thu May 25 17:27:58 PDT 2006 Firas Khatib I tried adding all the rr constraints to try8, resulting in try10, but try9 is still better than anything so far (improving slightly on try7). I will use Proteinshop on try9 to attempt to line up Y31 with F65 & I61 better. The Proteinshopped file is called liftTyr31.fromTry9.renum.pdb and running as try11. Thu May 25 21:25:29 PDT 2006 Kevin Karplus The try11-opt2 conformation looks pretty good, and is the best scoring with the try11.costfcn. It may be time to do a "polishing" run, with sidechain, soft_clashes, and break all turned up a bit (maybe to 8, 40, 100) and no constraints, starting from all existing models, but no alignments. I'm starting this are try12 on lopez. Thu May 25 21:40:30 PDT 2006 Kevin Karplus Looking at how try12.costfcn scores the different models, there is one server model that is up there with try9-opt2 and try11-opt2: SAM_T06_server_TS1 It does not have the turned part of the helix, so we may want to submit it as an alternative model. The next best server model is robetta model 5, which is not bad. We might try polishing it (as the sole input model) for an alternative prediction to submit. Fri May 26 14:50:21 PDT 2006 David Bernick A cursory look and an idea... this may be a calbindin molecule. If this is modeled as a 4-helix structure, the loop between helix 1 and 2 could be the backbone coordinated CA site, and the loop between helices 3 and 4 could be the loop that is coordinated by side-chains. Typically, this loop has concerved charged residues (Asp, or Glu). Since we know that some loops can coordinate with backbone atoms, it is conceivable that both loops are backbone coordinated. 4icb is an example of this. good luck - dlb Fri May 26 15:17:42 PDT 2006 Firas Khatib So David is talking about the coil from s10-N12 and the coil from S45-P48 in try12-opt2. Using VAST I found a decent alignment to residues 1-65 in many proteins (which include these two coils that David is talking about), you can see these here: casp7/T0283/decoys/Protoshop/VAST For example, open 1dceA.pdb.gz and select 50-77,88-123 and color it green. Then open try12-opt2 and select 1-65 and color the same. Our model looks very similar (which is a good sign) so even if there IS a calcium binding pocket among residues 1-65, at least our model is still proteinlike in those regions! Fri May 26 15:56:09 PDT 2006 Firas Khatib Looking at try12, there is still the hole around V75 that is exposed and should not be. I tried in Proteinshop to bring that small helix down (kind of like a cap), but without messing up anything too badly. I will run this through undertaker as try13 to see what happens (maybe we'll get lucky and it will help!) Fri May 26 18:11:01 PDT 2006 Kevin Karplus The try13 structure looks pretty good to me, though it is a bit surprising that try13-opt1 scores better than try13-opt2 in the score-all.try13 files (they have almost identical scores). We may want to do a polishing run with clashes turned up using all decoys as starting points. I'm a bit dubious about the EF-hand/calmodulin suggestion, as our HMMs are usually very good at finding calmodulin structures, and EF hands usually occur in pairs (or larger multiples) in proteins. Wed May 31 15:52:55 PDT 2006 Kevin Karplus I think we should submit a tentative prediction today (in case of leaks), with the option of submitting more solid predictions later. With the unconstrained costfcn, the top hits are try13-opt1 try12-opt2 try11-opt2 try9-opt2 try7-opt2 try10-opt1 SAM_T06_server_TS1 try8-opt2 try4-opt2 try3-opt2 (Note: opt1 models only reported above when they beat the corresponding opt2 model) The first non-SAM model is ROBETTA_TS5, which actually looks pretty good. We may want to try polishing it with our cost functions. The first from a different server is ROKKY_TS4-scwrl, which looks terrible (we may need more weight on phobic_fit to reject it). After that is ROKKY_TS3-scwrl, which looks pretty good. It has 6 helices (like our solutions), but has them in sort of crossing layers rather than a bundle. I like the bundle better. I'll try polishing up the ROBETTA_TS5 model (actually all 5 ROBETTA models), so that we can include it in our set. I'll start this as try14 on cheep. Wed May 31 16:28:39 PDT 2006 Kevin Karplus The try14 run seems to be inserting 2etvA.3.139 into ROBETTA_TS5 (more than once successfully), and these conformations are the ones that made it into the superiteration for try14-opt1. Wed May 31 16:48:29 PDT 2006 Kevin Karplus The try14-opt2 run gets costs very close to the try12 and try13 runs. We may want to do a run from the ROKKY_TS3 model (SCWRLing it) to see if that improves as much. Wed May 31 16:57:43 PDT 2006 Kevin Karplus I can't polish the ROKKY models, because the are backbone-only models and are missing the oxygen atom from the last residue, resulting in even the scwrled models being missing 8 atoms, so that there were no complete models to seed the conformation pool. It would require some changes to undertaker to be able to complete an incomplete conformation (we could have a command that inserted it into a random conformation, for example, then ran scwrl to fill in any residues that were missing sidechain atoms). The alignments from undertaker are so gappy, that there is no point in submitting them. I'll just arbitrarily pick two other models to submit: try13-opt2 try14-opt2 try7-opt2 try10-opt1 try4-opt2 Wed May 31 17:21:13 PDT 2006 Kevin Karplus I've submitted these five models, but we can resubmit if further work is done (particularly if we can get some other, different topology). Tue Jun 13 18:01:23 PDT 2006 Firas Khatib starting from crazyCapHelix75-77.pdb I tried to tighten the small helix at 75-77 so that it's not as exposed. I will input crazyCapHelixTake3.renum.pdb for try16 (on shaw) Wed Jun 14 17:59:42 PDT 20 Firas Khatib Dang! I can't seen to get that cap to come down with Undertaker, I might have to add some constraints! I actually though that crazyCapHelixTake3.renum.pdb would work! Fri Jun 16 15:46:45 PDT 2006 Firas Khatib Ok, I added 2 contraints so that the helix will stay down (hopoefully) to try17.costfcn! Fri Jun 16 17:24:59 PDT 2006 Firas Khatib that didn't work, I have increased the weights and am trying again for try18 Mon Jun 19 18:52:22 PDT 2006 Firas Khatib try19 was using actually using the constraints (since previously it had weight of zero!) but using read-pdb.under which included crazyHelix try20 is the same but only using these files as input: ReadConformPDB T0283.try11-opt2.pdb.gz ReadConformPDB T0283.try12-opt2.pdb.gz ReadConformPDB T0283.try13-opt2.pdb.gz ReadConformPDB T0283.try14-opt2.pdb.gz ReadConformPDB T0283.try16-opt2.pdb.gz ReadConformPDB T0283.try17-opt2.pdb.gz ReadConformPDB T0283.try18-opt2.pdb.gz ReadConformPDB T0283.try19-opt2.pdb.gz Tue Jun 20 10:52:27 PDT 2006 Kevin Karplus Bad idea---try20 just tried tweaking up try19, which was only good in that it had the desired constraint. You wanted to start from models that were *not* based on crazyCap. I'll start try21, same as try20, but excluding try19 and later. I'll include all our "try" models before try19, but not the robetta models (though robetta3 scores well with the try21 costfcn). I'll also turn up the pseudocounts on the CrossOver and CrossAndInsert operators, and turn down InsertSpecificFragment, since crossover might be a good thing to get the the good models tweaked up. (try21 started on camano) Tue Jun 20 14:57:15 PDT 2006 Firas Khatib so try21 took try3 as its starting point and that does not look good at all. I have run try22 using just: ReadConformPDB T0283.try11-opt2.pdb.gz ReadConformPDB T0283.try12-opt2.pdb.gz ReadConformPDB T0283.try13-opt2.pdb.gz ReadConformPDB T0283.try14-opt2.pdb.gz ReadConformPDB T0283.try16-opt2.pdb.gz ReadConformPDB T0283.try17-opt2.pdb.gz ReadConformPDB T0283.try18-opt2.pdb.gz and it's running now on orcas. I have gone through them all (to make sure that they are good starting points) and they all seem good except try14 (which looks VERY different because it came from inserting 2etvA.3.139 into ROBETTA_TS5), so I will run try23 using all those listed above EXCEPT try14 (I will finally get this right, I swear) and maybe I will change the constraints so that the sidechains can fit better instead of using the CBs. Tue Jun 20 15:12:50 PDT 2006 Firas Khatib try23 will use these constraints instead: Constraint V75.CG2 F65.CE1 -10. 4.2 6.0 80 Constraint V75.CG1 P28.CB -10. 6.2 9.0 100 try23 is running on orcas Tue Jun 20 16:47:30 PDT 2006 Firas Khatib it still is foamy. I will try inputting crazyCapHelixTake3.renum.pdb to see if Undertaker can fix it. This is try24 on orcas. Tue Jun 20 18:00:15 PDT 2006 Firas Khatib I am trying try23 again with less strict contraints to try to bring it down a tiny bit. This is try25 on lopez. Tue Jun 20 18:05:57 PDT 2006 Firas Khatib I am running try26 similar to try25 but with these differences that Kevin had added for try21: diff try25.under try26.under | grep -v try < InsertSpecificFragment 20 \ --- > InsertSpecificFragment 3 \ < CrossOver 3 \ < CrossAndInsert 5 \ --- > CrossOver 15 \ > CrossAndInsert 10 \ the costfcn is the same as try21's try26 is running on whidbey Tue Jun 20 18:44:10 PDT 2006 Firas Khatib try24 basically scores similarly to try23... sigh... Tue Jun 20 19:29:41 PDT 2006 Firas Khatib well, try26 sure is different, it managed to bring down the helix by rotating the rest of the chain! try25 bent the long helix a bit to get satisfy the weaker constraints that I had put in, but doesn't even score better. I think we are stuck with try18 unless there are any other ideas on how to make V75 come down more with messing up the model. Tue Jun 20 19:41:29 PDT 2006 Firas Khatib I will try one last attempt starting with try26, but adding a constraint to keep the N-terminus from rotating Tue Jun 20 19:58:35 PDT 2006 Firas Khatib try27 has as input a proteinshoped try26 called rotNtermFromTry26.renum.pdb try28 has as input a proteinshoped try26 called rotNtermLessFromTry26.renum.pdb last chance run, try29, will be the same inputs as try26, but with a constraint to keep the N-terminal helix from rotating (running on camano with try28) Tue Jun 20 21:08:28 PDT 2006 Firas Khatib Holy cow! I like try27! Tue Jun 20 21:13:58 PDT 2006 Kevin Karplus Well, try27 looks ok. The model is a bit foamy and E21 is buried without a compensating charge. But it may be the best we're going to get. Firas, can you list your 5 favorites for me, so I can submit tonight? A history of which model came from which other (so we don't just submit tiny variants of the same model) would also be useful. try27 <- rotNtermFromTry26 <- try26 <- try16 <- try13-opt1 <- try12 <- try11 <- liftTyr31.fromTry9 <- try9 <- fromTry7-opt2 <- try7 <- helix80-83rotateFromTry6 <-g79-80coil.try1.renum <- try5??? <- g80.randomPerturbOfG80coil.renum <- try4 <- alignments (2fvhA) Rosetta likes repacking first_try-opt2 and try4-opt2 best. unconstrained costfcn likes try28 best, but not by much. Tue Jun 20 21:25:37 PDT 2006 right, try28 is good, we'll see how try29 is and then I'll give you my top5 Tue Jun 20 21:27:54 PDT 2006 Kevin Karplus I made a new costfcn for helping pick tightly packed ones "dry.costfcn" It sees try18 and try28 as nearly equivalent, and try29 as slightly better. Tue Jun 20 21:32:10 PDT 2006 Firas Khatib so I guess try29 will be the best (of all those I tried bring down V75) then (good to know that the last run was worth it!) and then we should submit totally different folds, I'll give you a history of what I know. try29 <- try18 <- try17 <- try16 <- try13 <- try12 <- try11 <- liftTyr31.fromTry9 <- try9 <- fromTry7-opt2 <- try7 <- helix80-83rotateFromTry6 <- try6 <- g79-80coil.try1.renum.pdb <- try4 <- alignments (2fvhA I feel that try18 and try29 are different enough to be 2 different models. the other 3 models in best-models.pdb are the ones that you had submitted at the soft deadline: "The alignments from undertaker are so gappy, that there is no point in submitting them. I'll just arbitrarily pick two other models to submit: try13-opt2 try14-opt2 try7-opt2 try10-opt1 try4-opt2" one last thing I would add is that for try29 rasmol doesn't see a helix at V75 whereas it does for try27 & try28 (if you care about that) otherwise, i would pick try29,try18,and 3 different ones of your choice Tue Jun 20 22:14:58 PDT 2006 Kevin Karplus OK, let's try try29-opt2 Firas's favorite (also best with "dry" costfcn) try18-opt2 prior version, somewhat different try4-opt2 starting point for ProteinShop tweaking first_try-opt2.repack-nonPC rosetta's favorite try14-opt2 something different (from ROBETTA_TS5) Tue Jun 20 22:32:26 PDT 2006 Kevin Karplus So submitted. Make started Sun Feb 11 19:25:36 PST 2007 Running on orcas.cse.ucsc.edu Make started Sun Feb 11 19:28:30 PST 2007 Running on orcas.cse.ucsc.edu Make started Fri Jul 20 13:12:41 PDT 2007 Running on orcas.cse.ucsc.edu