Mon Jun 12 08:53:23 PDT 2006 T0327 Make started Mon Jun 12 08:54:02 PDT 2006 Running on orcas.cse.ucsc.edu Mon Jun 12 08:57:52 PDT 2006 Kevin Karplus Moderate blast hits to 1rsgA (50% over 26 residues, 3.8e-04) and 1ypxA (47% over 38 residues, 3.8e-04). A few longer, weaker hits also. This will probably turn out to be easy fold recognition. Mon Jun 12 10:00:29 PDT 2006 Kevin Karplus The HMMs seem to be finding different things depending what HMM you look at (not a good sign). The a.4.5.* superfamily members seem to be popular with the multi-track HMMs, particularly the 3-track ones, but rather marginally with the w0.5 HMMs. Mon Jun 12 11:50:59 PDT 2006 Kevin Karplus Sure enough, all the top hits (that are in SCOP) are from the a.4.5.* superfamily. The best one is 1tbxA (E-value 0.001). Mon Jun 12 21:46:47 PDT 2006 Kevin Karplus The model and the alignments are in agreement, but the target extends a bit further at the C-terminus. It might be a good idea to make a dimer, based on 1tbxa[AB] and optimize that. Tue Jun 27 16:38:26 PDT 2006 Grant Thiltgen Pinal and I are setting up the target to run as a dimer with 1tbxA. Looking at the A2M file, there appears to be a his-tag at the end of the residue since there are six histadines in a row. I'm not exactly sure I'm doing the dimer correctly. Also, the helix needs to be extended before the dimer works. I'm starting try1 of the dimer on camano. Wed Jun 28 17:32:09 PDT 2006 Grant Thiltgen I started try2 on vashon. It attempts to make the helix with the bend (around residues 76-80) into one long helix that might be able to be used for dimerization. I also started try3 on whidbey. It is using just 1tbx as a template. Thu Jun 29 14:37:16 PDT 2006 Grant Thiltgen try3 didn't work. I didn't have the read-alignments-scwrl.under file made so it couldn't use the alignments from that source only. I made the file, and I'm restarting try3 on camano. In try2 the helix went the other direction. I am running try4 on lopez with a higher helix constraint in the C-terminal region. I also started try2 in the dimer directory, using a dimer of try1-opt2 even though the helix is broken. I'm going to see if the helix will pull away from the second chain and form better if there is another chain next to it to dimerize with. try2 is starting on whidbey. Pinal helped me with the above. Thu Jun 29 15:15:48 PDT 2006 Grant Thiltgen I took the try1-opt2 model and in proteinshop, I made the bent helix at the C-terminal into a long helix. I then dimerized this model so the helix would be in the correct orientation. Pinal helped with this. [HOW DID YOU DIMERIZE THE MODEL? GIVE DETAILS (what undertaker script, in proteinshop, ... KJK] We took the new dimer, dimer-1tbxA-try1-opt2-proteinshop and ran it through undertaker to try to optimize things. This is try3 in the dimer directory. try3 is running on camano. Fri Jun 30 10:35:59 PDT 2006 Kevin Karplus It looks like Grant did a full "make" on the dimer directory, which is usually pretty useless, unless there is a multidomain protein that has two copies of the domain. Fri Jun 30 10:39:58 PDT 2006 Kevin Karplus The unsupported helices in dimer/decoys/T0327.try3-opt2.pdb look unlikely to me. There are two main possibilities: the helices form a coiled-coil or the multimeric state is actually a 4-mer, not a dimer. 1tbxAB seems to have decent packing, so we should take our best monomer and make a dimer from 1tbxA. None of the current dimers look at all right, though the monomers in dimer/try3-opt2 look somewhat feasible. Fri Jun 30 12:41:40 PDT 2006 Grant Thiltgen The dimer in dimer/try3-opt2 seemed to split the two helices apart too much. It is also extended quite a bit and looks unrealistic, but a good part of that is the hist-tag, so I'm not worrying too much about the last 6-8 residues. For try4, I am lowering TweakMultimer to 0 in the first optimization run and raising it to 1 in the second run, to try to keep the helices on the multimer together. I also raised OneRotamer, ClashingRotamer, and ClusteredRotamer in the first opt in the try4.under file to 3. I also increased OneRotamer in the second opt in the try4.under file to 8 and Clashing Rotamer and ClusteredRotamer to 3. In the try4 costfcn, I increased the sidechain weight from 5 to 10. I included a constraint to keep the two leucines (residue 73) together in the helix also. I did notice some things looking at the individual secondary structure predicitions. A lot of the t06 predictions do not have the long helix at the C-terminal end of the structure as a full helix, some of them (str2, n_notor, o_notor, stride, alpha) have a small break in the helix. Is it possible that the region at the end is not one long helix? I am starting try5 as well, which is the same as try4 to see if we get different results. It is running on vashon. Fri Jun 30 15:49:54 PDT 2006 Grant Thiltgen I'm still concerned about the length of the helix. The template model we're basing the dimer on, 1tbx, has shorter helices than the helices on our model, so I think it is throwing off the dimerization interface somewhat. There is also one residue, residue 73, which is a leucine and predicted to be buried in the helix, which is exposed. I wonder if the helices are a little off and need to be turned in a bit to bury those residues. Sat Jul 1 08:42:55 PDT 2006 Kevin Karplus dimer/try5 is a washout---the long helices collapsed. I think that we may need some constraints, both to keep the long helix long and to pack the monomers together. Sat Jul 1 09:00:30 PDT 2006 Kevin Karplus dimer/try6 started on lopez. As in dimer/try4, we're starting from dimer/decoys/dimer-1tbxA-try1-opt2-proteinshop.pdb.gz I've added constraints to keep the helices straight and to keep them packed against each other in more or less the same way they are in the starting model. If this works to produce a decent model, I might try another run with the distance constraints for the helices shifted, to try to get the helices to pack one turn closer. That will require making sure that TweakMultimer is on during the initial OptConform (we currently have it off). I should also turn off phobic_fit in the costfcn, as it really wants to collapse the long helix. Sat Jul 1 12:24:06 PDT 2006 Kevin Karplus dimer/try6-opt2 does not look too bad. There would need to be a bit of repacking of the interface between the helices to make things fit a bit better. I think that the two blobby domains on the helices are too close together in this dimer, though, as groups like ASP4 and K3 are bumping into each other near the CB---it would be better for them to be a bit further apart so that they could form salt bridges, I think. So how do I create a dimer with the helices slid over one notch? Proteinshop is a possiblity, but I don't have it at home and haven't learned to use it yet. Another possiblity is to construct a different alignment to 1tbxA, (using just the helices E67-W92) and construct a dimer in the way we've used before. Sat Jul 1 13:02:09 PDT 2006 Kevin Karplus I made two alignments using only the final helix: 1tbxA/T0327-1tbxA.dimer1-a2m same alignment of helices as before 1tbxA/T0327-1tbxA.dimer2-a2m shifted 3 residues (1 turn) Note: I commented out the incorrect reading of the pdb files from 1tbx.pdb1, as 1tbxA and 1tbxB can be read perfectly well from PDB---we don't need a special file to generate the quaternary structure. Sat Jul 1 13:35:27 PDT 2006 Kevin Karplus I made a small change to undertaker, adding force_alignment fragment_only options to ReadFragmentAlignment, so that I could force undertaker to treat the short fragments as being a complete alignment. Sat Jul 1 14:17:22 PDT 2006 Kevin Karplus I also set up 1tbxA/T0327-1tbxA.dimer3-a2m shifted 4 residues (1 turn) since that looked like it would try to pack the aromatics W79 and Y83 together better. Sat Jul 1 14:34:58 PDT 2006 Kevin Karplus I like the dimer3 best (dimer/decoys/dimer3-1tbxA-dimer6.pdb.gz) and so will try cleaning it up with dimer/try7 (running on lopez). Sat Jul 1 18:45:52 PDT 2006 Kevin Karplus Attempting to optimize dimer3 as dimer/try7 failed miserably. Tring again with different constraints as dimer/try8 on cheep. Sat Jul 1 21:35:55 PDT 2006 Kevin Karplus dimer/try8 does bend the long helix a bit, but it actually looks pretty good, burying the hydrophobics well and exposing all the polars. It could use a little cleaning up, which I will try doing with greatly reduced constraints (just the Helix constraints and the sheet constraints). Sat Jul 1 21:50:50 PDT 2006 Kevin Karplus dimer/try9 started on lopez Sun Jul 2 09:50:59 PDT 2006 Kevin Karplus dimer/try9-opt2 looks pretty good. The dimer could be a bit more tightly packed, but it's not bad now. I wonder if there should be a salt bridge between K3 of one chain and E75 of the other. K3.NZ and E75.OE1 are currently 7.5 Ang apart, but even just rotamer changes should make them much closer. The Y83 rings could pack against each other better, as could the W79 rings, but both sets of rings look roughly right to me. I'm not so sure of Y8 and Y72 (same chain) or F15 and W92 (between chains). When Rosetta tries to repack dimer/try9-opt2 it sees big problems (clashes) with L61, L65, I14, Y59, V37. Undertaker sees clashes in dimer/try9-opt2 also: other-bump: 1.58122 Ang (T0327)L61.CD1 and (T0327)L65.CD2 threshold= 3.03525 cost= 0.980011 other-bump: 2.29472 Ang (T0327)L61.CD1 and (T0327)L65.CG threshold= 3.06417 cost= 0.823599 other-bump: 1.70257 Ang (T0327)Y8.OH and (T0327)I188.CG1 threshold= 3.105 cost= 0.97282 other-bump: 2.5878 Ang (T0327)V37.CG1 and (T0327)Y59.CE1 threshold= 3.16939 cost= 0.7037 other-bump: 2.59178 Ang (T0327)L21.CD1 and (T0327)Y59.CE2 threshold= 3.08682 cost= 0.649637 other-bump: 2.62767 Ang (T0327)I10.CG1 and (T0327)I26.CD1 threshold= 3.09055 cost= 0.622245 In dimer/try9-opt2.repack-nonPC, some of the clashes have changed other-bump: 2.49886 Ang (T0327)L61.CG and (T0327)L65.CD2 threshold= 3.13005 cost= 0.741094 other-bump: 2.68183 Ang (T0327)L61.CD2 and (T0327)L65.CG threshold= 3.13005 cost= 0.604381 other-bump: 2.6999 Ang (T0327)L61.CD2 and (T0327)L65.CB threshold= 3.07686 cost= 0.543497 other-bump: 2.81585 Ang (T0327)I46.CG2 and (T0327)L65.CD2 threshold= 3.09357 cost= 0.431277 other-bump: 2.93708 Ang (T0327)H57.NE2 and (T0327)Y59.CE1 threshold= 3.23199 cost= 0.436789 other-bump: 2.93062 Ang (T0327)H57.CD2 and (T0327)Y59.CE1 threshold= 3.21178 cost= 0.422861 other-bump: 2.82806 Ang (T0327)V37.CG2 and (T0327)Y59.CE2 threshold= 3.09655 cost= 0.419687 other-bump: 2.88891 Ang (T0327)L61.CD2 and (T0327)T66.N threshold= 3.14149 cost= 0.395232 undertaker is not seeing a bad clash for I14 in the repacked model, with the worst being other-bump: 2.90991 Ang (T0327)I14.CG2 and (T0327)L61.CB threshold= 3.0227 cost= 0.20401 Perhaps that threshold is a little too low, and I should see a worse clash from it? Typical Van der Waals radius for a CB or CG atom is 1.85 Angstroms (AMBER), so ideal distance is 3.7---at what point should we start invoking clash penalties? Rasmol uses 1.872 for "united atom" carbon models, so the ideal would be 3.75. Sun Jul 2 10:36:30 PDT 2006 Kevin Karplus I won't mess with the clash tables just yet. I've started dimer/try10 on cheep to try to pack dimer/try9 tighter while reducing breaks and clashes. Sun Jul 2 13:12:27 PDT 2006 Kevin Karplus dimer/try10 has reduced clashes a fair amount, but 3 breaks are still a problem, T0327.try10-opt2.pdb.gz breaks before (T0327)N24 with cost 3.18499 T0327.try10-opt2.pdb.gz breaks before (T0327)H57 with cost 2.26919 T0327.try10-opt2.pdb.gz breaks before (T0327)P56 with cost 1.42325 Part of problem may be strain on the backbone from the H-bond H bond: 3.37258 Ang (T0327)R55.NH2 and (T0327)E23.OE2 Rosetta still sees problems with L61, L65, I14, and L21. undertaker agrees that the repacked L61, L65 have problems, but sees only minor flaws for I14 and L21. Before repacking, undertaker sees problems with L61:L65, Y8A:I86B, I10:I26. Perhaps I should do an optimization with breaks up even higher and hbond_geom turned down. One possibility is to otpimize from the dimer/try[8,9,10] gromacs models, which are scoring terribly, but are quite close in 3-space to the try10-opt2. The terrible score gives the optimizer a chance to bounce out of the local minimum it is stuck in, while not sacrificing any of the good structure. (I think) Sun Jul 2 13:45:05 PDT 2006 Kevin Karplus dimer/try11 started on lopez, using a costfcn with no constraints and starting from just the dimer/try[8,9,10] gromacs models. Sun Jul 2 15:15:16 PDT 2006 Kevin Karplus It looks like dimer/try11 was a waste of time, as the cost function never gets close to what try10-opt2 scores with the try11 costfcn. I should probably do a try12 run with the same costfcn but starting from all try8...try11 models. Sun Jul 2 15:47:48 PDT 2006 Kevin Karplus I finally did it this year---I forgot to edit try11.under to replace try10 by try11, so I stepped on all the (probably better) try10 files. I think I can restore them from my home computer, after moving the new "try10" files to try11. Actually, the dimer/try10 pdb files were *not* stepped on, because the gzipping of the try11 ones was not forcing. Thus I could rename them and not have to restore the try10 pdb files (just the three try10-opt2.sheets .helices and .constraints files). Sun Jul 2 15:57:47 PDT 2006 Kevin Karplus Actually, the dimer/try11-opt2 model scores quite well with the try11 costfcn, and rosetta actually likes it best for repacking (finally beating out the try4-opt2 which had very straight helices and conflicts between the two monomers). dimer/try11-opt2 also scores best with the try9 and try10 costfcns, but not with try8 (which prefers dimer/try9), probably because of some of the overzealous packing constraints in dimer/try8.costfcn. Sun Jul 2 16:09:49 PDT 2006 Kevin Karplus I just noticed that our monomers seem to owe more to 1ub9A than to 1tbxA. I should try taking the best monomer (from dimer/try11-opt2 and remaking the dimer based on 1ub9. I downloaded the dimer from PQS (http://pqs.ebi.ac.uk/pqs-doc/macmol/1ub9.mmol) and saved it as 1ub9A/1ub.pdb.gz Sun Jul 2 16:18:46 PDT 2006 Kevin Karplus I made 1ub9A/T0327-1ub9A.dimer-a2m from the 80+60+80 local alignment, and used make-dimer3.under to create dimer/decoys/dimer-1ub9A-dimer11.pdb.gz Although this dimer has a lot of clashes, it is a better starting point than our other dimers made from monomers. I'll try optimizing in in dimer/try12 using the same costfcn as for dimer/try9. If it comes out as good as try9, then I'll try polishing it up using the try11 costfcn. dimer/try12 started on cheep. I think I could have save a lot of fussing if I'd noticed earlier that 1ub9A made a better dimerization model than 1tbxA. Sun Jul 2 18:08:53 PDT 2006 Kevin Karplus dimer/try12 scores almost as well as dimer/try11 and has fewer clashes and breaks. I'll do a polishing run on all the good models. Sun Jul 2 18:54:59 PDT 2006 Kevin Karplus I was in too much of a hurry (got called for dinner just as I was starting dimer/try13). I really do *not* like dimer/try12---the helices have straightened out, but they cross at a steep angle and are unsupported for most of their length. dimer/try13 is apparently based on dimer/try12, so it is likely to be junk also. Sun Jul 2 19:04:06 PDT 2006 Kevin Karplus dimer/try14 (excluding try12, but otherwise the same as try13) started on lopez. Sun Jul 2 22:01:32 PDT 2006 Kevin Karplus I submitted with the following comments: For T0327 we used Proteinshop to straighten out a helix that monomeric optimization had crumpled up (it likes to reduce radius of gyration). We then formed a dimer based on 1tbx[AB] (though 1ub9A might have been a better choice in retrospect), and optimized as a dimer. Model 1 is chain A from dimer/try14-opt2, our favorite dimer. Model 2 is chain A from dimer/try13-opt2, a rather less convincing dimer in which the helices barely interact. Model 3 is try4-opt2, a monomer optimized from alignments (mainly 1ub9A), showing the folded over helix. Model 4 is sidechain replacement by SCWRL on an alignment to 1tbxA. Model 5 is sidechain replacement by SCWRL on an alignment to 1ub9A. ------------------------------------------------------------ Thu Jul 20 13:09:13 PDT 2006 Kevin Karplus I just noticed that we had no dimer gromacs0.repack-nonPC models, so I'm running mult1 ... mult14 on the farm cluster. Thu Jul 20 14:05:37 PDT 2006 Sylvia Do I was trying to make dimer-1ub9A-try1-opt2-proteinshop.pdb (make-dimer4.under) but it doesn't seem to work. Fri Jul 21 14:57:57 PDT 2006 Kevin Karplus I don't know what Sylvia was trying, but it isn't clear to me that proteinshop is useful for this target. If we were to make a new dimer based on 1ub9, I'd want to start with our best model: dimer/decoys/T0327.try14-opt2.unpack.pdb.gz chain A Makeing the mult* models yesterday did not seem to work (the gromacs runs seems to have happened, but not the rosetta runs). One problem may have been the shared use of XXXX.pdb and XXXXA.fasta, with the parallel runs stepping on each other. Another problem may have been: decoys/grep-best-rosetta.tmp: Stale NFS file handle. grep: write error: Stale NFS file handle though the farm machines don't seem to be having trouble with NFS today. Fri Jul 21 15:23:39 PDT 2006 Kevin Karplus Rosetta likes best the dimers try13, try5, and try12, all of which have the helices unnaturally straight and unsupported. I suppose we could do another polishing run on dimer/try14, perhaps starting with the try14-opt2.unpack.gromacs0.repack-nonPC model. Fri Jul 21 15:40:29 PDT 2006 Kevin Karplus dimer/try15 started on abyss, to do a little polishing on dimer/try14-opt2.unpack.gromacs0.repack-nonPC Fri Jul 21 17:58:40 PDT 2006 Kevin Karplus dimer/try15-opt2 does not score quite as well with undertaker as try14-opt2, but rosetta likes T0327.try15-opt2.unpack.gromacs0.repack-nonPC.pdb model better than any of the others that have the helices tightly intertwined. (It likes some of the very open ones more.) Fri Jul 21 18:06:33 PDT 2006 Kevin Karplus The differences between try15 and try14 are so subtle that I don't see much point to resubmitting the monomer. (If there were dimer submissions being accepted for this target, it might be worth tweaking some more.) I declare this target done.