Thu Jul 29 15:55:39 PDT 2004 T0257 Due 25 Aug 2004 Sat Jul 31 03:09:48 PDT 2004 Kevin Karplus The try1-opt2 file looks pretty trashy, but the models from alignment are not much better. We may have to go with new-fold methods for T0257. There is not much overlap between the weak hits found by t2k and by t04: t04: 1jeyA 1n9pA 1gbg 1d3bA 1h4rA 1l1zA 1jb0D 1ckmA 1vcaA 1r2zA 2ayh 2rspA t2k: 1bywA 1gbg 1o9uA 2ayh 1h8fB 1h8fA 1uv5A 1ny9A 1vzoA 1cpn (1gbg and 2ayh shared). It's not clear whether rr constraints will help or not. Mon Aug 2 12:17:40 PDT 2004 Kevin Karplus try2 has selected a fold, which is sort of compatible with some of the rr constraints, but which disagrees a little with the secondary structure predictions. It is not too terrible though, so may be worth pursuing. It doesn't really cluster the conserved residues. I'll remove a few of the "bonus" keywords, and try another run, to see if the same fold comes up. Wed Aug 4 02:19:49 PDT 2004 Kevin Karplus The fold is looking pretty good, but strands involving V107, I100, and M34 are flipped over. Maybe we should ask M34 to line up with N159, and flip the strands by changing the hbond specs: Hbond Q158-S35 Hbond M34-I100 Hbond T101-L106 ?? (not the usual i-i+3 hbond) Mon Aug 16 22:00:03 PDT 2004 Martina Koeva I have started a try4 in an attempt to flip the strands over by introducing the following sheet constraints: SheetConstraint W32 Q36 L102 L98 hbond M34 7 SheetConstraint G33 S37 V160 I156 hbond S35 7 I have increased all dry and wet weights to the standard values used currently. I have also reduced sidechain back down to 1 (from 6), have increased the break weight from 20 to 40 and have increased phobic fit to 4 (from 2). Mon Aug 16 22:29:57 PDT 2004 Martina Koeva I submitted the structure to VAST to see whether we can come up with some templates that would overlap with the ones that we have: VAST ID: VS60831 Password: T0257try3 VAST came up with quite a few pretty good hits with good E-values. The top hit (with highest e-value) is 1umzA (I need to double-check ids again in dunbrack-pdbaa), 1mveA (which is another 1,3 - 1,4-beta-glucanase, just like some of our top templates 1gbg and 2ayh), 1kit4(?), 1o4yA, 1vavA, 1gv9A, 1a8d1(?), 1oq1A. Some of the alignments are still gappy, but overall it would be worth it to get those few extra ones and use them. I will do that next. I think I am starting to see an agreement in the template topologies to a jellyroll fold of some kind, which looking at try3 seems to fit pretty well. ------------------------------------------------------------------------------- Date: Mon, 16 Aug 2004 22:27:14 -0700 From: Kevin Karplus To: martina@soe.ucsc.edu Cc: karplus@soe.ucsc.edu Subject: Re: T0257 You asked > I was looking at T0257, which is one of the targets that I picked up last > week (and is due some time next week). Right now (at try3) we are > forming pretty good sheets. I noticed that you commented on the > strands being flipped and have started a try4 to fix that. > > The fold seems to fit pretty decently the templates that t2k and t04 agree > on - 1gbg and 2ayh. My only concern is that the fold does not > match really well our secondary structure predictions. What would you > advise me to do - just keep going with the current fold and try to polish > it (in which case we should be done pretty soon), or try to get the fold > to match out predictions better? I looked at the sequence logos for both t04 and t2k predictions of T0257, and the regions where the coloring is "wrong" are extremely weak predictions. I'd say that we stick with the fold, which is probably the correct one. -------------------------------------------------------------------------------- Wed Aug 18 01:11:15 PDT 2004 Martina Koeva Try4 did not manage to flip the strands over, at least from what I can see. I am wondering whether adding alignment might not help...It seemed to work on T0222, even though the case there was quite different. I am going to incorporate the extra alignments from VAST anyway. Wed Aug 18 02:18:02 PDT 2004 Martina Koeva I have started try5 on peep after updating the Makefile with the suggested structural templates in the MANUAL_TOP_HITS. I ran extra_alignments on them, as well as included all the VAST alignments. For try5 I have kept the cost function the same and running from alignments. Wed Aug 18 14:06:06 PDT 2004 Martina Koeva Try5 did not succeed in flipping the strands over. I wonder whether that is due to the sheet constraints themselves, or to the weight given to them. In order to test the latter hypothesis, for try6 I introduced much higher sheet constraint weights. If this does not help with flipping the strands, I will have to go back and slide the residues (strands) to introduce different hbonding pattern. Thu Aug 19 17:27:48 PDT 2004 Martina Koeva I have taken the sheets formed in try6 from try6-opt2.sheets and have retained the large majority of them, except for one that we really need to have to be able to force the strands to flip. I have set those up in try7. Fri Aug 20 14:28:20 PDT 2004 Martina Koeva After another unsuccessful attempt to flip the strands over, I have taken out all sheet constraints, except for the one that attempts to flip strand 33-40 over. I have also turned down the break weight from 40 to 20. Try8 is running on hoot, starting from alignments. Sun Aug 22 16:02:47 PDT 2004 Martina Koeva At Kevin's suggestion, since I have been quite unsuccessful in sliding the strands back and forth to flip them over, I put together a script called: show-align-1gbg.under, because I believe our target predictions are based on alignments to 1gbg. The output from show-align-1gbg.under has been written to T0257.1gbg.undertaker-align.pdb. From the alignments, I have included 3-track local and global str2+CB_burial-14-7 alignments, as well as the muscle alignment. It seems to me that I need to change the muscle alignment (models 3 and 6) in T0257.1gbg.undertaker-align.pdb and modify it so it can flip the strands that have residues 34,100 and 107. Sun Aug 22 17:13:04 PDT 2004 Martina Koeva I modified the muscle-t2k alignment to 1gbg (model 6 in the T0257.1gbg.undertaker-align.pdb) and saved the new hand-edited alignment as 1gbg/T0257-1gbg-muscle-hand1.a2m. I have rerun the show-align-1gbg.under script to include the model for this alignment as model #7. As I can see in the pdb file, the strands don't line up nearly as nice (to form actual hbonds), but they are aligned well to be able to possibly do that. Also, M34, I100 and V107 are now successfully pointing in. I have started a try9 only from this alignment and keeping the cost function, helix, strand and sheet constraints from the try7 cost function. I have only turned back down the break weight from 40 to 20. Sun Aug 22 19:10:28 PDT 2004 Martina Koeva Try9 looks like a big jumbled mess, but from what I see in the try9.log file, my hand-alignment was not found/deciphered/read in. So I am starting try10 and will monitor whether the hand alignment is read this time around. Ah, great! Undteraker has read in the alignment with no problem. Hopefully, try10 will produce something more interesting than the try9 models. My mistake was including on the line where I specified that I wanted to start only from this alignment, the following: ReadFragmentAlignment NOFILTER SCRWL 1gbg/T0257-1gbg-muscle-hand1.a2m For try10 I corrected that to: ReadFragmentAlignment NOFILTER 1gbg/T0257-1gbg-muscle-hand1.a2m. Sun Aug 22 19:34:59 PDT 2004 Martina Koeva I just noticed that the t04.many.frag file was never created. I will have to go back after this try10 is done and redo the initial make to create the file. Yay! Try10-opt1 is done and even though it does not look as complete as some of the earlier ones, it has flipped all three strands sucessfully (M34, I100 and V107 are now on this inside). There are quite a few breaks, but I will have wait and see first try10-opt2. Sun Aug 22 20:40:35 PDT 2004 Martina Koeva Hmm, so try10-opt1 does not score particularly well with the unconstrained function. Is it because of the large number of breaks? After try10 is done, I will redo the initial make to create t04.many.frag file and will start a try11 from existing models with an increased break weight. Try10 has just finished, so after creating the t04.many.frag file I started try11 from existing models with the same sheet constraints as try7. For try11 I have increased the break weight to 70 and the soft_clashes to 30. I have also started try12 with the sheet constraints, generated by try10 (through try10-opt2.sheets) and have retained the break weight at 40 and soft_clashes at 20. Mon Aug 23 01:28:23 PDT 2004 Martina Koeva Both try11 and try12 finished and unfortunately I requested that the tries start from existing models, so both try11 and try12 look mostly like previous models before try10. In both cases M34, I100, V107 are turned out as in all the models before try10. In an attempt to change this and to optimize the try10 models, instead of starting from all existing models, I am letting try13 and try14 start only from try10-opt2.pdb. Try13 has the same costfcn as try12, while try14 has the same costfcn as try11. Mon Aug 23 17:46:32 PDT 2004 Martina Koeva Kevin noticed at today's meeting that K60 was also flipped the wrong way, so I went back to look at the original hand alignment and create a new one, called T0257-1gbg-muscle-hand2.a2m, in which I have shifted the reisudes so that K60 points out. I have started try15 only from this alignment and after it is done, I will want to setup a try16 which increases on the breaks weight and starts only from the try15-opt2 models and tries to close the breaks on it. I am a little conscerned though, because try13 was supposed to start from the hand-alignment and from try10-opt2.pdb, but it flips residues M34,I100 and V107 back to point out, which was the whole purpose of doing the hand alignment and running try10 - so that those residues can point in (which they do in the try10 models). Mon Aug 23 17:56:06 PDT 2004 Kevin Karplus The hand1 alignment file was not read in try13, because there was no file name with an ^M on the end of it. Martina---stop using Windows editing tools if you can't remember to remove the trash ^M characters. T0257.try10-opt2.pdb was not included, because the file was specified incompletely as try10-opt2.pdb (see the try13.log.gz file for the error messages). try15 has the alignment included as desired. The try15 run did not bury any chared residues, but is still burying polar residues. It doesn't score very well on the unconstrained costfcn, but perhaps a polishing run starting from all the try15 models and using the unconstrained costfcn would help. Mon Aug 23 19:20:53 PDT 2004 Martina Koeva I have started try16 only from the try15 models (opt1,opt1-scrwl,opt2) and with an unconstrained cost function. All models and alignments have been read in correctly this time around, try16 running on ribbit. Tue Aug 24 07:48:29 PDT 2004 Kevin Karplus try16-opt2 looks ok, but seems to have lost hbonds on some of the beta strands. It may be worthwhile to create sheet constraints consistent with the way the strands are lining up and optimize with them, to get the sheets back. Particularly messy are A48-Q54 and F58-I62. I think we want SheetConstraint K50 Q54 A130 T126 Hbond N128 I'm not sure that flipping the strand to unbury K60 was such a good idea after all---it causes several polars to be buried and exposes F58. If we take back that idea, we are back to try12-opt2, I think, though that has several conserved hydrophobics on the outside. Perhaps try10-opt2, though that has some bad breaks. Maybe we should try try13 and try14 over again, but without the errors this time. That is, try optimizing with try13 and try14 costfcns, with the try10 model and all alignments. I'll try reading in the try10-opt2 model BEFORE doing the TryAllAlign, so that the alignments will be added to the try10-opt2 model. This may be a waste of time, if none of the alignments improve it. try17 uses a copy of the try13 costfcn, but with clashes and breaks turned down and constraints turned up, so that it is easier for the conformation to move toward the desired one. It still likes try12-opt2 best of the existing models. Tue Aug 24 08:54:09 PDT 2004 Kevin Karplus Sure enough, try17 alignment insertion did no good---optimization is proceeding just from try10-opt2, which may not enable strand flipping. For try18, I'll not include try10-opt2, but just optimize from alignments. For try19, I'll make similar modifications to the try14.costfcn. Tue Aug 24 12:39:52 PDT 2004 Kevin Karplus try18-opt2 seems to be doing fairly well. It is second-best (after try12-opt2) on the unconstrained cost function, and does fairly well on the burial terms (it is losing to try12 on breaks, dry5, and dry6.5). I'll do a polishing run on everything except try12 (which I expect will mainly polish try18, but may pick up something good from a different model). Tue Aug 24 13:36:37 PDT 2004 Martina Koeva Here is a summary of the differences between tries, in regards to the conserved hydrophobics (M34, I100,V107) and the charged K60: try8 and before - hydrophobics buried, K60 buried try9 - n/a try10 - hydrophobics (M34,I100,V107) buried, K60 buried try11 - hydrophobics exposed, K60 buried try12 - hydrophobics exposed, K60 buried try13 - n/a try14 - n/a try15 - hydrophobics buried, K60 exposed try16 - hydrophobics buried, K60 exposed try17 - hydrophobics buried, K60 buried try18 - hydrophobics exposed, K60 buried try19 - hydrophobics exposed, K60 buried Tue Aug 24 13:54:37 PDT 2004 Kevin Karplus It sounds like Martina is suggesting doing an unconstrained optimization run from try15 and try16, since try18 and try12 are in the same equivalence class. Maybe we should do one from try10 also. try21 will start from try15 and try16. try22 will start from try10. Tue Aug 24 15:08:00 PDT 2004 Kevin Karplus try20 has finished, and is our new top score. Martina says that she would prefer this model to be the first choice, and has no preference between the equivalence classes represented by try21 and try22. I'll probably submit try20-opt2 try21-opt2 try22-opt2 try1-opt2 T0257-1gbg-t2k-local-str2+CB_burial_14_7-0.4+0.4-adpstyle5 2nd best template Tue Aug 24 15:20:12 PDT 2004 Martina Koeva It seems to me that after all try21 might be a better model than try22. I just went back to look at score-all.unconstrained.pretty, as well as the score file for try20, and try10 (which try22 is a polishing model of) is really not well-liked by any of the score functions. It is way down on the both lists, even after some of the more random conformations that we got, due to my earlier errors (it scores even after try9, which looks horrible). Try10 also does not score well with its own cost function. It places after try8,try5,try7,try4 and try6, which are all representatives of the same equivalence class as try12, try18 and now try20. I'm thinking that no matter how much of an improvement try22 shows over try10, it probably won't be as good as the try21 (try16 is already scoring pretty high). I was a little overt in liking the try16 type of structure, because we lost a couple of the strand pairings, but it seems like all score functions seem to be pointing to it as a better structure than the try10 one. Tue Aug 24 15:47:25 PDT 2004 Martina Koeva I rescored all files with the unconstrained cost function and try20-opt2 is now at the top, as expected. Try21-opt1 is pretty high up, only behind the try20 and try12 models. Try22-opt1, as expected, is still way down the list.