Tue Jul 27 11:07:39 PDT 2004 T0250 DUE 15 Aug 2004 Wed Jul 28 10:26:49 PDT 2004 Kevin Karplus This has weak hits to several different superfamilies, so will probably need to be done as new fold. Thu Jul 29 02:53:32 PDT 2004 Kevin Karplus ------------------------------------------------------------ T0250 try1 is a mess It looks like there are a few RR constraints that may be helpful in pulling things together, though not very strong ones. Conserved H173, H175, and H186 may be part of a metal-binding site. and should probaby be clustered Constraint H173.NE2 H175.NE2 -10 6 9 1.0 Constraint H173.NE2 H186.NE2 -10 6 9 1.0 Constraint H175.NE2 H186.NE2 -10 6 9 1.0 The secondary structure patterns seems fairly clear in the t04 predictions: HelixConstraint T10 L18 6.018 HelixConstraint I25 E30 8.218 HelixConstraint Q92 M103 6.218 HelixConstraint T134 R141 6.726 HelixConstraint E162 L163 6.32 HelixConstraint E214 K218 6.588 HelixConstraint Y223 A233 7.254 StrandConstraint L5 Y6 6.084 StrandConstraint E45 L49 7.068 StrandConstraint K52 A56 6.802 StrandConstraint F62 L65 6.814 StrandConstraint Y73 T77 7.021 StrandConstraint S110 F118 6.565 StrandConstraint H120 M125 7.165 StrandConstraint G153 F158 6.785 StrandConstraint I174 F176 6.246 StrandConstraint H186 V192 6.438 StrandConstraint V195 Q200 6.065 StrandConstraint Q206 H208 6.913 I suspect that D193 and D194 may be a turn of a hairpin. The alignments in t2k.undertaker-align look very trashy---we almost certainly want to skip the "TryAllAlign" step and just do fragment insertion. The t04 and t2k multiple alignments look like they have almost identical conservation patterns. We might prefer the 2ry predictions from t04 then, since they have the better-trained networks. The top hits list is not really very useful---there is essentially no agreement between the t04 and t2k predictions, and they are all in the range where false positives are very likely. PREFERRED_AL_METHOD:=t04 MANUAL_TOP_HITS:= 1e49P 1k8fA 1s1dA 1azzA 1pboA 1g85A 1lnlA 1osnA 1h4rA 1us8A 1us8A 1s14A 1f2tA 1h4rA 1cr5A 1jj2B 1s72B 1vybA 1iz5A 1 huxA 1h2cA 1ud9A 1nf3C 1pyaB Thu Jul 29 09:50 PDT 2004 Kevin Karplus Observation: The reason (at least a reason) try1 was so messed up was that the alignment all-align.a2m.gz was not getting properly created before try1 was run. I moved try1 to no_align1, and am rerunning try1 after creating all-align.a2m.gz properly. Fri Jul 30 00:13:39 PDT 2004 Kevin Karplus try2 begins to get some sheet, but still nothing fold-like after Y73. The sheet looks pretty good, so we should add it as sheet constraints, perhaps adding the next strand: K83-P86 antiparallel to N78-I76, not clear about Hbonding. Strand constraints for later strands need to be increased, since some are winding up into helices. Fri Jul 30 05:22:36 PDT 2004 Kevin Karplus redone try1 looks like it may be trying to form a sandwich. rr constraints look like they may help a lot in getting topology better. Sat Jul 31 05:25:03 PDT 2004 Kevin Karplus try3 has a lot of sheet, including a fair amount that we had predicted to be helix! The weak rr constraints are not helping much. I'm not sure what to do next. Tue Aug 3 16:35:26 PDT 2004 George Shackelford I'm trying to fold up some strands into a major sheet. 109-125 folds into a sheet but I think it's a long strand with sets off of each end. First things first. TRY4 on caw. Thu Aug 5 14:25:35 PDT 2004 George Shackelford I had a problem with file permissions that I have resolved. Now I'm going to work on the sheets I have and try and use an rr constraint as a guide. I also undid some sheets that didn't look right. TRY5 running on caw. Thu Aug 5 23:58:02 PDT 2004 George Shackelford Try5 doesn't look as good as try4. I'm going to use some stretches of hydrophobics to line up more likely sheets. I'll see how they go. This resulted in some sheets being undone. TRY6 on peep. Sun Aug 8 08:13:55 PDT 2004 Kevin Karplus try5 and try6 look horrible. I think that the strand constraints may need to have more weight to keep the strands from curling up into helices. try4 at least had some sheets! In fact, try1, try2, and try4 all look better than try5 or try6. Perhaps the sheet constraints from the earlier tries should be incorporated into the cost function. (Consolidate the constraints from try1-opt2.sheets, try2-opt2.sheets, and try4-opt2.sheets, making reasonable guesses where they disagree.) I'll leave this for George to do. Sun Aug 8 23:08:40 PDT 2004 George Shackelford I am taking over now. I've got printouts of the t04.ehl2 and t04.near logos. I'm labeling the strands as follows: StrandConstraint L5 Y6 6.084 # s1a StrandConstraint I36 L49 0.0 # s1b ? StrandConstraint E45 L49 7.068 # s2 StrandConstraint K52 A56 6.802 # s3a StrandConstraint F62 L65 6.814 # s3b StrandConstraint Y73 T77 7.021 # s3c # StrandConstraint T84 P86 0.0 # s4 ?? StrandConstraint S110 F118 6.565 # s5a StrandConstraint H120 M125 7.165 # s5b StrandConstraint G153 F158 6.785 # s6 StrandConstraint I174 F176 6.246 # s7 StrandConstraint H186 V192 6.438 # s8a StrandConstraint V195 Q200 6.065 # s8b StrandConstraint Q206 H208 6.913 # s9 s1b looks iffy but it does have an rr to s3c and looks a bit like s7 as both are rather hydrophobic. In fact, the hydrophobic sections may help in matching up strands. I am combining s8a and s8b and strengthening it. TRY8 on peep. Mon Aug 9 22:13:58 PDT 2004 George Shackelford It is really trying to get somewhere. But it turned part of the helix at 92 into a strand (no way). Also I am beginning to see better matches for strands. ~36 is parallel (via str2) also ~156. (but not with each other?) Tue Aug 10 08:41:12 PDT 2004 George Shackelford It's fighting me. It keeps folding 110-125. I've put in a constraint to force it apart. It doesn't want to fold up s3a,s3b. I undid the s1||s3c and adjusted the s3a,s3b to match 56 to 62 (both phobic). I going to try matching the two long strands, 110-125 and 186-201. I hope it helps. TRY10 running on peep. Tue Aug 10 12:19:38 PDT 2004 George Shackelford Gosh. Try10 hasn't even finished running and I'm already making changes. I'm including all the t04 constraints (should have done that sooner). I'm going to make s6,s7 parallel! It looks good. Otherwise, sheets seem to be forming nicely. May need extra constraints to make the second half of the long strands s5,s8 from curling up (~185,125). TRY11 running on peep. Tue Aug 10 16:36:23 PDT 2004 George Shackelford Try11 warped most of the strands. I'm reenforcing and cranking up the hbonds. It'll have another run. TRY12 on peep. Tue Aug 10 23:48:43 PDT 2004 George Shackelford Try12 is shattered into pieces. I found helix constraints in 'try12.helices' where there should not be any. I don't know where they came from but I'm taking out some of the t04 constraints I've included leaving 'str2' and 'ehl2' only. Will that make a difference? I've also dropped the sheetconstraint for the long ones from 35.0 to 10.0. If they aren't straight, I can't make a sheet anyway. TRY13 running on peep. Wed Aug 11 11:48:43 PDT 2004 George Shackelford Try13 is much better. The sheets are many where they should be. Helix 134-141 is flat; I'm reenforcing it - again! Need to see if something is suggesting it is a strand. I have found that the Bystroff alphabet is probably the problem with curling up strands. Once the end of a defined strand is curled, it cascades and takes over (until a 'turn'). I've generally taken out the bys stuff, but I'm still getting + I've turned down the constraints for the strands since taking out the bys. + I find the s2,s3a a rather bad match wrt phobic. I'm breaking it up. + The s8b,s9 needs fixing (turn it back on and fix the hbond), + s8a needs to stay away from s8b (if we're going to match to s5). A separation constraint would help. TRY14 running on peep. Wed Aug 11 12:58:52 PDT 2004 Kevin Karplus I reformatted George's paragraphs, so that I could read them without the incredibly long lines. The best models with the unconstrained.costfcn are try13-opt2 try2-opt2 try1-opt2 The try13 costfcn prefers try11-opt2. I've put some of these into best-models.pdb.gz, so that we can see that we are getting rather different predictions. The second model in T0250.t2k.undertaker-align.pdb (5th model in best-models.pdb) is a rather pretty beta helix, but it is almost certainly wrong, as the characteristic turn signals for the beta helix are missing. Wed Aug 11 15:47:29 PDT 2004 George Shackelford Kevin doesn't use 'soft wrap.' I'll hard wrap lines after finishing my entries. So try14 finished and actually looks semi-decent. Once again the long pair s5,s8 are curling up on one end. I've decided to redefine them as 5a,5b,8a,8b and work with them that way. I'm defining two different sheets now and I'll see what this does. It means that there may be an odd bend between them. s6 and s7 are matched in 'near' and they are close to each other but not forming a parallel sheet. I'm switching to anti-parallel. I'm uncommenting the parallel match s5a||s6. So far, the conserved t04 residues are not converging on just one or two places. I need to keep an eye on that and perhaps form a group (of histidenes?) Try15 running on peep Wed Aug 11 16:36:18 PDT 2004 Kevin Karplus The unconstrained costfcn likes try13-opt2.repack-nonPC best, then try14-opt2.repack-nonPC. George's runs are all from alignments, which is the right thing to do when trying to get new conformations. At some point we'll want to do a polishing run that starts from existing models. Wed Aug 11 23:00:00 PDT 2004 George Shackelford The scoring likes try13 best? Well try15 has 'potential' but it is not clicking into place. I think I may try cranking up the constraints to force the sheets. Strand s3b needed to be straightened out and sheeted. I changed its hbond and cranked stuff up. So I do an overnight run to see if that makes any difference. Running TRY16 on peep. Thu Aug 12 06:24:34 PDT 2004 Kevin Karplus With the try16 cost function, try15 scores best. The best score with the unconstrained costfcn is still try13-opt2.repack-nonPC George, if you are really intent on polishing try15, try starting from the exiting models instead of starting over from the alignments (that is, use "include read-pdb.under" instead of "TryAllAlign" to get the starting conformations). I'll start an optimization run with the unconstrained costfcn (copied to try17.costfcn), which will probably polish up try13-opt2. Thu Aug 12 09:51:12 PDT 2004 George Shackelford I'm not so intent on polishing as I am on getting the sections to behave. We may have to fall back on try13 (and earlier decoys not, any of the later ones). The latest effort, try16, is simply more of the same. I'm going to work with try17 once it's finished. I'm starting try18 with even tighter phobic parameters but lighten up the hbond_geom_beta_pair to 10 (we're getting a bit of flat helices). I'm going to modify the read-pdb.under to remove the 14,15, and 16 pdbs. Try18 running on peep. From ggshack@pacbell.net Thu Aug 12 10:25:45 2004 From: George Shackelford To: Kevin Karplus Subject: Focusing on T0250 try17 Date: Thu, 12 Aug 2004 10:25:41 -0700 Saw your notes in the readme. Try17 is out and it really looks a lot better than tries 14,15, and 16 (which are crud). I'll start polishing and tightening it up. This is a new fold after all. Q: Can we modify the 'read-pdb.under' to ignore the pdb's for 14,15, and 16 or is that file automatically generated? I really don't like tries 14-16. ============================================================ From karplus@soe.ucsc.edu Thu Aug 12 11:32:50 2004 Date: Thu, 12 Aug 2004 11:32:50 -0700 From: Kevin Karplus To: ggshack@pacbell.net CC: karplus@soe.ucsc.edu In-reply-to: <200408121025.41843.ggshack@pacbell.net> (message from George Shackelford on Thu, 12 Aug 2004 10:25:41 -0700) Subject: Re: Focusing on T0250 try17 read-pdb.under is automatically created. If you were using a real editor, it would be a trivial matter of a few keystrokes to include it in the tryxx.under file and comment out the models you weren't interested in. If try17 scores better than try14,15,16 with the cost function you are using, then there is no need to comment them out. Thu Aug 12 17:08:22 PDT 2004 George Shackelford Try18 is back. I've cranked up the usual cost values to compress T0250 but it still has this "handle" sticking off. Should I just put some constraints in to pull that in? I'm at my wit's end (Not that that goes very far). Perhaps Kevin has some suggestions. From ggshack@cse.ucsc.edu Thu Aug 12 17:56:14 2004 From: George Shackelford To: Kevin Karplus Subject: How to squeeze T0250 down Date: Thu, 12 Aug 2004 17:55:09 -0700 I've cranked up the usual cost values to compress T0250 but it still has this "handle" sticking off. Should I just put some constraints in to pull that in? I'm at my wit's end (Not that that goes very far). - George From karplus@soe.ucsc.edu Thu Aug 12 19:44:37 2004 Date: Thu, 12 Aug 2004 19:44:36 -0700 From: Kevin Karplus To: ggshack@soe.ucsc.edu CC: karplus@soe.ucsc.edu In-reply-to: <200408121755.09837.ggshack@cse.ucsc.edu> (message from George Shackelford on Thu, 12 Aug 2004 17:55:09 -0700) Subject: Re: How to squeeze T0250 down If you want to move a piece that is sticking out, you'll probably have to add constraints, and even that might not help. Thu Aug 12 20:18:32 PDT 2004 George Shackelford Try20 running on peep. So I'm going to put some rr constraints in to push things around... Here goes. Fri Aug 13 07:19:46 PDT 2004 George Shackelford Ok, try20 shows no change. I'll crank up the rr constraints and try again. Try21 running on peep. Fri Aug 13 12:17:22 PDT 2004 George Shackelford I give up. Try21, try22... I can't get the handle out of there. The current best scoring (using the unconstrained polishing score function) selection is: Try18-opt2.pdb (NOTE: Try17-opt2.pdb actually comes here, but it is basically the same as try18; I don't know if we should be submitting something that is so similar.) Try13-opt2.pdb Try2-opt2.pdb try14-opt2.pdb try1-opt2 full auto The fact that Try14 makes the list and is a piece of trash shows how things have gone. I just wish I could have collapsed that bridge. Fri Aug 13 15:23:24 PDT 2004 Kevin Karplus The best Rosetta energy is for try4-opt2.repack-nonPC, followed by try1, try11, try17, try3, ... Rosetta sees them all as having horrendous clashes. The "handle" George is referring to involves the exposure of residues L87, L90, V95, F96, and I99, at least some of which look like they should be near V123. Unfortunately this region is anchored at each end (I76 and E102), and moving it would probably require forming some large gaps and then closing them again. We might be able to do something by adding constraints, turning the break costs WAY down (and maybe the clash costs also), then reoptimizing from the changed conformation with the weights turned back up. I'll try a crude version of this for try23. Since the try23 costfcn scores some of the early runs high, I copied in the read-pdb.under file and commmented out the high-scoring runs that came before try18 (leaving in high-scoring runs that were more recent). Fri Aug 13 16:23:41 PDT 2004 Kevin Karplus oops, it looks like I accidentally left in try1-opt2 for the try23 run. Luckily, it looks like try21 and try22 take over after a while in the optimization, though with the expected larger breaks and clashes. If try23 doesn't completely destroy the models, I'll try polishing it up with weaker (or no) constraints and higher break and clash costs. Fri Aug 13 20:32:26 PDT 2004 Kevin Karplus try23-opt2 does pack the "handle" in but is a bit foamy, and has some breaks. It scores well with the try23 costfcn, but not with the unconstrained one. Compared with the best-scorer (try22-opt2), it does better on some measures, but worse on sidechain, soft_clashes, backbone_clashes, break. The difference (about 16 points), comes about half from soft_clashes, and half from break. Oops, it looks like try23 got its constraints satisfied not by packing the residues in, but by pulling V123 out of the sheet! I should have spread the constraints around more, to different strands of the sheet, so that the sheet wasn't pulled apart. I'm not sure it is worth another attempt at this point. I'll submit try22-opt2 best score with unconstrained costfcn (and some constrained ones) try4-opt2.repack-nonPC best Rosetta energy try1-opt2 full auto try15-opt2 scores best with several of the constrained functions T0250-1k8fA-t2k-global-adpstyle5 longest alignment (2nd best template)