Thu Jul  6 09:10:40 PDT 2006
T0362
Make started Thu Jul  6 09:12:22 PDT 2006
Running on cheep.cse.ucsc.edu

Fri Jul  7 23:56:56 PDT 2006 Kevin Karplus

Weak hit with BLAST to 2gf6A (E-value 0.002)

Strong hits with HMMs (1s5uA, 2cyeA, ...) d.38.1.1

Mon Jul 17 15:35:31 PDT 2006 Navya Swetha Davuluri , Sylvia Do

362 looks like a dimer but first the monomer has to be optimized more to fix
 the breaks before Q110 and V60. We are going to increase the weight on "break"
 to 150 and on "soft_clashes" to 40. We will also increse the dry6.5 and dry5
 weights to 25 and 20 respectively. We will also include the sheet constraints
 from try1.opt2

try2 is running on bark


Tue Jul 18 12:21:12 PDT 2006 Sylvia Do

Try2-opt2 has fixed the two major breaks. Now we are going to dimerize this. 
Okay. Got the dimer. Cynthia and Crissan helped me with this.

It seems that I did not dimerize it correctly. The dimer interface areas are
not lined up correctly. 

I asked Grant, and the problem is that I used the wrong template. So this can
be fixed by using the correct template, or ProteinShop.
Also, I used 2cyeA, not 1s5uA,since 2cyeA had the best e-value. This may be
the cause. 

ProteinShop is really hating me right now. I'll ask Firas about it later; in
the meanwhile, I'll try using 1s5u as the template.

Okay. That looks MUCH better.

There is still a big red spot when using burial though. 

Just a note: there are some extra files because I ran a make instead of 
"undertaker < make-dimer.under... "


Wed Jul 19 17:27:44 PDT 2006 Navya swetha Davuluri

The dimer made from 1s5u looks good but it h-bonds between two monomers did not form. So I put a sheet constraint between the monomers specifying the h-bond between the sheets. 

try3 is running on shaw.

I also made a tetramer out of the monomer as I felt that the protein probably is a tetramer but Dr.Karplus told to first optimize the dimer. So I am optimizing the dimer first.


Wed Jul 19 20:24:43 PDT 2006 Sylvia Do

Oops! Turns out that I had to fix the .under files for the tries. Apparently,
 they were not pointing to the correct files (needed to add the ../).
So, basically, I took the constraints for try3 and copied them onto try2
 (which I was just messing around with), and fixed the try2.under file. 

I would have fixed the try3.under file, but I could not write to it.

try2 is running on orcas.  

Thu Jul 20 01:03:25 PDT 2006 Sylvia Do

Whoa. Don't know what happened to try2. It pretty much ripped everything apart.
I tried again with try4, with the constraints written differently (A strand to 
B strand).

try4 is running on orcas.

Thu Jul 20 12:41:21 PDT 2006 Sylvia Do

Still not working. I played around more with the .under file. I think that is 
what's causing the problem. Try5 will still try to create the H-bonds between 
the monomers. 

try5 is running on whidbey.

Thu Jul 20 14:15:52 PDT 2006 Kevin Karplus

It may be worth looking at server models SAM_T06_server_TS1,
ROVETTA_TS4 and ROBETTA_TS1.

Thu Jul 20 15:50:20 PDT 2006 Sylvia Do

I killed try5 because try5-opt1 didn't look like what I was hoping for.I'll
 try again, reusing the costfcn from 2. It will still be try5.

try5 is running on whidbey.

I put the aforementioned server models into the superimpose-best.under 
and they all agree on almost the entire model. 
There are some stray areas, but it looks like it is from an insertion. 

The other models seem to agree on a strand at about VAL64, where our model
came up a little short. It has a strand, but not as long as the others.

Okay. Asked Firas about this, and so I added
Hbond H70 N D74 O  10
Hbond H70 N H74 O  10
Hbond C71 N A73 O  10
Hbond C71 N A73 O  10 
         
 to try6's constraints and commented out the sheetconstraint.

Fri Jul 21 10:58:51 PDT 2006 Navya swetha Davuluri

I think Sylvia put the wrong constraints for forming hbonds between the two monomers and I think that is the reason the protein was ripped apart. I am going to start a new run with the following sheet constraint.

SheetConstraint (T0362)H70 (T0362)D225	(T0362)D74 (T0362)H221 hbond (T0362)H70		1	 

Also Sylvia did not include "multimer 2" in OptConform commands. I think that this is also one of the reasons why the protein got ripped apart.

try7(mult) is running on camano.

Fri Jul 21 12:31:37 PDT 2006 Navya swetha Davuluri

try7 did not finish as I did not enter -k option for make. So started a new run again.

try8 running on shaw.

Fri Jul 21 15:29:24 PDT 2006 Navya swetha Davuluri   Sylvia Do

try8 did not finish yet but we wanted to start another run.

try9 started on orcas.

Fri Jul 21 15:33:05 PDT 2006 Kevin Karplus

Do you really want
	include T0362.dssp-ehl2.constraints
	include T0362.undertaker-align.sheets
	include rr.constraints
in dimer/try9.costfcn?  Those files don't exist in the dimer
directory. If you ane to include them from the superior directory,
you'd need an "InfilePrefix ../"
Also, there are no pred_alpha2k, ... cost function components being
created, so they should not be referred to in the SetCost command.

 
Fri Jul 21 17:45:53 PDT 2006 Sylvia Do

try9 does not appear to be working. Gah. I am using Crissan and Navya's 
.under file from their T0364 target since that target and this one are so
 similar. I also played around with the costfcn, one that is a copy of try9 and
another that is a mimic of the one in T0364. I'm running two tries, hoping
 that one of them will work.

try10 is running on shaw.
try11 is running on vashon.

Fri Jul 21 19:09:45 PDT 2006 Navya swetha Davuluri

Sylvia,I think you included the wrong sheet constraint for try11.costfcn. I cannot think of the reason why the protein is being ripped apart
other than using the wrong sheet constaint which I fixed in try9. Will see how try10 and try11 will do.

Fri Jul 21 20:12:26 PDT 2006 Navya swetha Davuluri

try10 and try11 also look terrible. I don't know what mistakes we have been making, but all the optimizations look really terrible. 


For todays preliminary submission:

None of the optimizations look any good. We have only optimized dimer so far since the monomer that we got from try2 looks really good.
We fixed the breaks from try1 and it scores better than other models.

From scoreall+servers.unconstrained.pretty the scores are:

 T0362.try2-opt2.pdb.gz    8.1  7.9  7.9  45.9  60.2  49.7 17.6  1.0    4.1   2.0    -22.8 -0.5  4.7 0.0  1.4 0.5  0.4  0.6 -3.1 -4.7 -12.4 -18.6    0.0  0.0  149.99
 SAM_T06_server_TS1        8.6  8.2  8.0  46.2  61.4  50.2 17.6  1.3    4.3   2.1    -23.2 -1.3  4.0 0.1  2.1 0.5  0.5  0.7 -3.1 -4.7 -13.0 -19.1    0.0  0.0  151.45
 T0362.try2-opt2.repack-n  8.3  7.9  7.9  46.6  61.2  50.3 17.6  1.0    4.1   2.0    -21.6 -0.5  3.5 0.0  1.4 0.5  0.4  0.6 -3.0 -4.7 -12.4 -18.6    0.0  0.0  152.73
 T0362.try1-opt2.pdb.gz    8.6  8.1  8.0  46.5  60.2  49.5 17.4  0.8    4.1   2.3    -23.0 -1.3  3.5 0.0  2.6 1.0  1.4  1.7 -3.0 -4.6 -13.1 -17.8    0.0  0.0  152.82
 T0362.try1-opt2.repack-n  8.7  8.1  8.0  47.3  61.3  49.8 17.5  0.7    4.1   2.3    -21.5 -1.3  2.4 0.0  2.6 1.0  1.4  1.7 -2.9 -4.6 -13.1 -17.8    0.0  0.0  155.70
 SAM_T06_server_TS1-scwrl  8.9  8.2  8.0  47.8  62.8  51.1 17.7  1.2    4.3   2.1    -24.4 -1.3  6.5 0.1  2.1 0.5  0.5  0.7 -3.1 -4.7 -13.0 -19.1    0.0  0.0  156.97
 T0362.try1-opt1.pdb.gz    8.8  8.1  8.1  46.8  60.9  50.0 17.4  0.9    4.0   2.1    -21.5 -1.1  4.8 0.1  2.3 1.3  1.7  2.1 -3.0 -4.5 -12.7 -17.4    0.0  0.0  159.29
 T0362.try1-opt2.gromacs0  8.7  8.2  8.1  47.0  60.4  49.5 17.4  0.8    3.4   4.9    -18.8 -0.6  0.6 0.0  1.7 1.7  2.1  2.7 -3.0 -4.5 -12.7 -17.2    0.0  0.0  160.38
 T0362.try2-opt1.pdb.gz    8.4  7.9  7.9  47.3  61.4  50.4 17.7  1.1    4.5   2.0    -21.7 -0.3  5.9 0.1  1.8 1.1  1.2  1.5 -2.9 -4.6 -12.2 -17.3    0.0  0.0  161.12
 T0362.try2-opt2.gromacs0  8.1  7.9  7.8  46.4  60.6  49.8 17.6  1.0    2.5   4.8    -17.3 -0.5  1.5 0.0  1.2 1.7  2.0  2.5 -3.0 -4.6 -12.1 -16.9    0.0  0.0  161.18




So the submissions has to be in the order try2-opt2<SAM_T06_server_TS1 < try2-opt2.repack-n < try1-opt2 <try1-opt2.repack -n <SAM_T06_server_
TS1-scwrl


Fri Jul 21 20:32:09 PDT 2006 Kevin Karplus

Sylvia and Navya asked me to look at why their dimers kept blowing up.

I tried looking at dimer/try11.log, and found 

# SheetConstraint # Error: strands differ in length Warning: strands too close on chain:
 (T0362)H70--(T0362)D225 (T0362)D74--(T0362)H221
# Hbond # Error: can't parse first atom name

...

# command:Warning: Couldn't open file decoys/dimer-from-try2-opt2-template-1s5u.pdb or decoys/dimer-from-try2-opt2-template-1s5u.pdb.gz for input
Trying dimer-from-try2-opt2-template-1s5u.pdb
Error: Couldn't open file dimer-from-try2-opt2-template-1s5u.pdb or dimer-from-try2-opt2-template-1s5u.pdb.gz for input
Error: can't open file dimer-from-try2-opt2-template-1s5u.pdb in ReadConformPDB

So the SheetConstraint they wanted was not correctly specfied, and
they mis-spelled the model they were trying to read in.
Reading log files looking for errors is a very useful skill for debugging!


dimer/try12.costfcn was fixed by changing the constraint to 
SheetConstraint H70 D74		D225 H221 hbond C71		1	 

dimer/try12.under was fixed by putting the missing "A" back into the
name of the model to be read.

Fri Jul 21 20:39:45 PDT 2006 Kevin Karplus

I have started dimer/ T0362.mult12 on cheep.

oops, more bugs---they still had T0284 mentioned in the .under file!!!
Fixed that, and increased weight for OptSubtree, to try to get hbonds
set right as quickly as possible.

Fri Jul 21 20:46:26 PDT 2006 Kevin Karplus

I have started dimer/ T0362.mult12 on cheep again.

Fri Jul 21 22:06:45 PDT 2006 Kevin Karplus

dimer/ try12 seems to have formed the inter-monomer Hbonds quite nicely.

Fri Jul 21 22:20:18 PDT 2006 Kevin Karplus

Preliminary submission done, with comment


    For T0362, we got excellent alignment to 1s5uA with the automatic model.
    We did not tryt to optimize the model further as a monomer, but made a
    dimer based on 1s5uA, and optimized the dimer.

    For the preliminary submission, we have done only one optimization of
    the dimer (previous attempts failed because of repeated propagation of
    typographical errors).  We hope to do more optimizations to close gaps
    and reduce clashes before the final deadline.

    The differences between models is rather slight---mainly in the
    rotamers chosen.

    Model 1 is chain A of dimer try12-opt2, the first successful
	    optimization in dimer context.

    Model 2 is chain A of dimer try12-opt2.gromacs0.repack-nonPC, which is
	    model 1 reoptimized by gromacs, then with sidechains repacked
	    by rosetta.  (This will probably be used as a basis for
	    further optimization.)

    Model 3 is try2-opt2, an optimization from the alignments with a
	    slightly different cost functon than the automatic one.  It
	    scores slightly better (in a monomeric context) than the
	    automatic model (try1-opt2).

    Model 4 is try1-opt2.gromacs0.repack-nonPC, the automatic model after
	    reoptimizing with gromacs and repacking the sidechains by rosetta.
	    It is rosetta's favorites of the monomers for which it
	    repacked sidechains (which was not very many--just try1-opt2
	    and try2-opt2, with and without gromacs optimization).

    Model 5 is just sidechain replacement by scwrl on an alignment to 1s5uA.


Fri Jul 21 22:49:07 PDT 2006 Kevin Karplus

I had to redo the submission, which was rejected by CASP.  One of the
chain ids was mistyped in the MANUAL_TOP_HITS.  That is why I now
(almost) always do a cut-and-paste operation to create the list.

Mon Jul 24 14:38:35 PDT 2006  Navya swetha Davuluri

dimer/try12-opt2 has the hbonds that I wanted to form. Since this
protein has a very strong alignment with 1s5uA which is a tetramer I
want to optimize this protein as a tetramer.

I am polishing the existing tetramer. I increased the weights on
dry6.5, dry8 and phobic-fit to pack it more tightly.

mult2 running on shaw.


Mon Jul 24 19:24:14 PDT 2006 Kevin Karplus

Nothing was produced for the tetramer, because the 4mer/try2.under
script referred in several places to files that didn't exist.  To make
it work some of the InFilePrefix commands need to be changed to
	InfilePrefix ../

The tetramer should probably be generated from the dimer, not the
monomers, if there is any point to optmizing as a tetramer anyway.
Further polishing of the dimer may be more effective.

Tue Jul 25 13:05:25 PDT 2006 Sylvia Do

Like Navya said, the model looks like it can be packed more tightly. So,
for try13 for the dimer will have higher weights on dry 6.5,  and 
phobic_fit.

From try12's costfcn, I increased dry6.5 to 30 and phobic_fit to 5.

mult13 is running on shaw.


Tue Jul 25 14:16:29 PDT 2006   Navya swetha Davuluri

As Dr.Karplus suggested, I am going to polish the dimer. try14 is a polishing run of try12-opt2. I am going to use try13's costfunction.

try14 running on shaw.


Tue Jul 25 16:24:53 PDT 2006 Sylvia Do

Some gaps that refuse to close. Problem is that there is nothing in the area
that looks like it can move to cover them. 

Maybe move 129-150 a little?

I'm just going to up the costs for dry6.5 to 40, dry5 to 25 , and dry 8 to 20
 for try 15, based off 
try13's costfcn. Also raised soft clashes to 45.


 try15 is running on cheep.
 
Wed Jul 26 06:22:51 PDT 2006 Kevin Karplus

I added the KnownBreak to dimer/unconstrained.costfcn and scored the dimers.
Currently, dimer/try14-opt2 is scoring the best, though many are quite close.
Rosetta also likes try14-opt2.unpack.gromacs0.repack-nonPC best of the
models it has repacked.

I'll do a polishing run of the dimer with no constraints, but with
dry terms, breaks, and clashes turned up a bit.  I think this may be
the last run, as this target is polished about as well as undertaker
can manage.

I started dimer/mult16 on cheep.  

Wed Jul 26 07:47:17 PDT 2006 Kevin Karplus

dimer/try16-opt2 scores best with the try16 costfcn, and rosetta likes
try16-opt2.unpack.gromacs0.repack-nonPC best by quite a bit.

The try14-opt2 costfcn slighlty prefers decoys/try14-opt2, because of
better hbond scores and better side chains.

I think that there is not much point to more polishing, so I will submit

InFilePrefix dimer/decoys/
ReadConformPDB T0362.try16-opt2.unpack.gromacs0.repack-nonPC.pdb chain A
ReadConformPDB T0362.try14-opt2.unpack.pdb chain A
InFilePrefix decoys/
ReadConformPDB T0362.try2-opt2.pdb
ReadConformPDB T0362.try1-opt2.gromacs0.repack-nonPC.pdb
InFilePrefix 
ReadConformPDB T0362.undertaker-align.pdb model 1	#t06 1s5uA

Wed Jul 26 08:05:51 PDT 2006 Kevin Karplus

So submitted, with comment

    For T0362, we got excellent alignment to 1s5uA with the automatic model.
    We did not try to optimize the model further as a monomer, but made a
    dimer based on 1s5uA, and optimized the dimer.

    The differences between models is rather slight---mainly in the
    rotamers chosen.

    Model 1 is chain A of dimer try16-opt2.unpack.gromacs0.repack-nonPC,
	    which was optimized in a dimer context with alternate rounds
	    of undertaker, gromacs, and sidechain repacking by rosetta.
	    It is rosetta's favorite of the sidechains it repacked.

    Model 2 is chain A of dimer try14-opt2, an earlier round of dimeric
	    optimization, which still has slight breaks and clashes.

    Model 3 is try2-opt2, an optimization from the alignments with a
	    slightly different cost functon than the automatic one.  It
	    scores slightly better (in a monomeric context) than the
	    automatic model (try1-opt2).

    Model 4 is try1-opt2.gromacs0.repack-nonPC, the automatic model after
	    reoptimizing with gromacs and repacking the sidechains by rosetta.
	    It is rosetta's favorites of the *monomers* for which it
	    repacked sidechains (which was not very many--just try1-opt2
	    and try2-opt2, with and without gromacs optimization).

    Model 5 is just sidechain replacement by scwrl on an alignment to 1s5uA.

Fri Aug 11 03:13:47 PDT 2006 Kevin Karplus

T0362 was selected as a refinement target, but only 1-144 are in the
solved structure.  The loop G45-E63 is quite different in tr362 from
our models.  Running gromacs does not improve tr362's score, but
repacking sidechains does. At the N-terminus, tr362 starts the strand
one residue earlier than us, which looks pretty good.

We'll have to optimize tr362 in a dimer context.

Fri Aug 11 05:39:42 PDT 2006 Kevin Karplus

make-refine.dimer.under (bad script naming on my part)
makes dimer/decoys/dimer-tr362-1s5uA.pdb from try362.repack-nonPC 

This dimer scores fairly well (about the middle of the pack for
dimer/decoys/score-all.unconstrained.pretty).  I'll try polishing it
with the unconstrained costfcn and see what happens.

Fri Aug 11 06:01:42 PDT 2006 Kevin Karplus

dimer/mult17 started on lopez, with the same costfcn as was used for
try16 (unconstrained, but with larger break and clash weights).

Fri Aug 11 07:12:55 PDT 2006 Kevin Karplus

dimer/try17-opt2.unpack.gromacs0.repack-nonPC looks pretty good to
rosetta (almost as good as try16-opt2...).

The try17 costfcn does not like it so well, putting try17-opt2 only in
about the top third of dimer models generated.  The problems with the
model ar spread out through all parts of the costfcn.

The helices have too much space between them (by about 8 Ang), which
probably measn that the monomers are not joined correctly at the
sheet.  The try17 dimer seems to be pairing C71 with C71 of the other
monomer, but maybe it should be pairing C71 with A73.

I could add 
SheetConstraint H70	D74	D225	H221	hbond	C71
to the costfcn and try again.
Perhaps the existing sheets and helices should be put in as
constraints also, to try to improve them more.


Fri Aug 11 07:31:47 PDT 2006 Kevin Karplus

mult18 started on lopez.  Sheet and helix constraints from try17-opt2,
plus the inter-monomer constraint mentioned above (which is compatible
with the try16 dimer, so is not a new idea with the refinement).

Fri Aug 11 11:20:34 PDT 2006 Kevin Karplus

dimer/try18 has still not moved the helices close enough, and the
sheet constraint is not met yet.  I should up the constraint and lower
the clash weight.

Fri Aug 11 11:28:37 PDT 2006 Kevin Karplus

mult19 started on lopez.

Sat Aug 12 04:36:03 PDT 2006 Kevin Karplus

dimer/try19 still has space between the helices, and doesn't make the
sheet constraints.  I think we need to look at dimer/try14 and see
what that did differently, perhaps tweaking the refinement model.

Although the dimer/try19 costfcn likes try19-opt2 ok (better than
try18 only), rosetta still liked try18 better.

Sat Aug 12 05:09:59 PDT 2006 Kevin Karplus

The main differences between the refined dimer/try19 and our
dimer/try14 is the G45-E63 loop.  Our models extend the helix further
than is predicted by the secondary-structure predictors, so I think
that the "better" score of our models is due to making too many helix
H-bonds. 


Sat Aug 12 09:30:11 PDT 2006 Kevin Karplus

I'll try again with the inter-monomer sheet constraint increased even
more than in try19, with clashes turned down.  If this doesn't work, we
may have to make a chimera---I'm convinced that the sheet does form
continguously, even if the helix or the edge strand has to move a little.

Sat Aug 12 09:41:28 PDT 2006 Kevin Karplus

I've started dimer/try20 starting from just the original refinement
dimer, with the constraints turned way up and the clashes and breaks
turned down.  (The try20.costfcn still likes try14 best.)  If this
succeeds in reshaping things enough to form the sheet, then I can
polish to get the refinement models.  If it doesn't, I'll have to
consider a chimera or something to fix the problem.

Sat Aug 12 19:10:59 PDT 2006 Kevin Karplus

OK! dimer/try20 did make the appropriate sheet from dimer-tr362-1s5uA,
so now I just need to polish it.

Sat Aug 12 19:18:14 PDT 2006 Kevin Karplus

dimer/mult21 started to try to polish the gromacs-reoptimized versions
of dimer/try20. Constraints were turned off, and breaks and clashes
increased (as well as dry5, dry6.5, phobic_fit, n_ca_c, and
bad_peptide).

Sun Aug 13 12:28:27 PDT 2006 Kevin Karplus

dimer/try21-opt2.unpack.gromacs0.repack-nonPC is also as good in
Rosetta's view as try16 (its favorite).  undertaker's try21 costfcn
still thinks it has a ways to go.  Main differences are hbonds,
sidechain, and bad_peptide.

We could try doing one more polishing run, but I think we're beginning
to hit diminishing returns.

Sun Aug 13 14:23:55 PDT 2006 Kevin Karplus

dime/try22-opt2 scores better than try21 (though still not as well as
try12--try16) with the try22 costfcn, but rosetta still prefers
dimer/try21-opt2.unpack.gromacs0.repack-nonPC

We may have gone about as far as we can with optimizing this model.

Sun Aug 13 17:32:30 PDT 2006 Kevin Karplus

Refinements submitted with comment

    For T0362, we did almost all refinements in a dimer context.  We first
    did a sidechain repacking using rosetta, then made an initial dimer by
    superimposing that monomer on 1s5u.  Note: we looked at tr362,
    tr362.gromacs0, tr362.repack-nonPC, and tr362.gromacs0.repack-nonPC
    (that is, optimization with gromacs, repacking of sidechains by
    rosetta, and both) and picked the monomer that scored best with the
    unconstrained undertaker costfcn.

    The dimer was optimized through a cycle:
	    undertaker (to make potentially large changes to backbone and sidechains)
	    gromacs (to remove small clashes)
	    rosetta repacking sidechains (except PRO and CYS)
    We adjusted the cost function of undertaker on each iteration,
    gradually increasing the cost of breaks and clashes.

    The first iteration was try20, the second try21, and the thrid try22.
    The changes at that point seemed to be below the resolution of the
    cost functions to discriminate between good and bad models.

    Refinement model 1 is try22-opt2 chain A, the final cycle after
	    undertaker optimization.  This is the refinement model that
	    scores best with the undertaker try22 costfcn.
    Refinement model 2 is try21-opt2.unpack.gromacs0.repack-nonPC chain A.  
	    This is the refinement model that rosetta likes best of thost
	    it repacked sidechains for.
    Refinement model 3 is try20-opt2 chain A, the first optimization by
	    undertaker that got the sheet to continue between the monomers.
	    It was optimized directly from the initial dimer, with strong
	    constraints on the inter-monomer sheet.
    Refinement model 4 is try19-opt2 chain A, our second attempt at
	    creating the continuous sheet between monomers (unsucccessful).
    Refinement model 5 is try18-opt2 chain A, our first attempt at
	    creating the continuous sheet between monomers (unsucccessful).

------------------------------------------------------------