Wed Aug  7 19:34:51 PDT 2002
t0173

T0173.doc.html says crystallized as monomer, dimer in solution, and
chain breaks present.
Maybe a fold-recognition target (or even ab initio).  Most likely
template is 1e0tA, fold c.1.

8 August 2002 Kevin Karplus

On first run, best score is for T0173.try1-opt-scwrl,

There are still plety of bad gaps, some strands not attached to a
sheet, and a helix that has been unwound.

I don't know if we should optimize this more, or try to find a better
alignment to work from.


24 August 2002 Kevin Karplus

I'll try a run from random start with no initial TryAllAlign

25 August 2002 Kevin Karplus

try2-opt-scwrl is new best.  I haven't looked at it yet.

25 Aug 2002 Kevin Karplus 14:15

try2-opt-scwrl is absolutely terrible!  The beta sheet has been
completely exploded.

This one looks a little difficult.  I selected the best alignment to
1et0A using superimpose.under,   and though it does not look great,
there are some beta-sheet fragments we'd want to keep.
Perhaps I should extract HBond constraints from the better parts of
this and do another optimization run from scratch.


27 Aug 2002 Kevin Karplus 14:51

I picked out the hbonds for the two sheets from T0173-1et0A-best.pdb
There are probably more constraints needed, but let's start with these.

29 Aug 2002 Kevin Karplus 17:36

Best score is now try3-opt-scwrl.

Pattern of 2ry structure is
	strand	R6-H11
	helix
	strand	Q34-T42
	helix
	strand  I85-Y86 (could be P84-G88)
	helix
	strand V132-Y136
	helix
	strand K179-V184
	helix
	strand A225-E228
	helix
	strand R255-L259

The strands are fairly strongly predicted to be parallel, with a
fairly uniform interval, so let's try to stack them
      tpRLLFVHah
     rgaQVHVVTCTlg
    tgvsapIYlggagr
      rphVVVTYdp
     tvpKFYWTVlg
      gidAVVEAde
     ptgRAAALsnn

Note: there is no apriori reason why the strands should be in this
order, but the first three are coming out that way, so let's try to
continue .

Current hbonds that match:
strand1-2	matches expected:	
	R6.O	V35.N
	L8.N	V35.O
	L8.O	V37.N

strand 2-3
	H36.O	Y86.N

But H-bonds in the second half don't match at all.

Let's do a run from just the alignments using the try4 constraints
that have CB constraints for the predicted strand-helix-strand
progression of strands in order.

Fri Aug 30 11:47:37 PDT 2002

Best score now try4-opt-scwrl
It almost forms a barrel.

We may want to slip R255-L259 back two  with respect to
A225-D230.  The non-loop charges (like D224) are coming out inside the
barrel, which seems reasonable.  E228 is twisted to the wrong side, I think.


We have only 7 strands, when 8 would be needed to close the TIM barrel.

Current arrangement of barrel is

   tpRLLFVHah
   rgaQVHVVTCTlg
    tgvsapIYlggagr
    rphVVVTYdp
     tvpKFYWTVlg
         gidAVVEAde
          ptgR AAALsnn


I'd like to fix this to be 

    tpRLLFVHah
   rgaQVHVVTCTlg
    tgvsapIYlggagr
    rphVVVTYdp
     tvpKFYWTVlg
      gidAVVEAde
     ptgRAAALsnn

This will be try5.constraints

I may want to move the third strand back 2 to get the P out of the
center of the sheet, and I may not want to move the 7th strand back 2,
as I have done.

Let's try

    tpRLLFVHah
   rgaQVHVVTCTlg
  tgvsapIYlggagr
      rphVVVTYdp
       tvpKFYWTVlg
        gidAVVEAde
         ptgRAAALsnn

as try6.constraints.

Fri Aug 30 17:40:49 PDT 2002

try5-opt makes a fairly nice barrel, with even a hint that L293
belongs in the missing 8th strand, but the last two strands seem to
twist around each other. E228 faces the wrong way.


try6-opt also makes a fairly nice barrel, but the last 2 stransd seem
to be be misaligned to the rest.


Conserved residues that may be part of active site are H13,D15,D16,E44,
R68,E71, D95, H144, D146,H147, H244, Q247, E271, 

We'd like these to be close in space.

For try5-opt, the H13-D16 cluster seems to have been pulled out of the
barrel a bit and D95 is too far past the C end of the strand to be at
the edge of the barrel next to  H144,D146,H147.

H244 and Q247 are at the N end of the barrel, so are probably not
part of the active site.  R68 and E71 are on a helix, and I don't see
how they could come into the active site either.


In try6-opt the H13 cluster comes close to  R68, but the H144 cluster
has swung out into space.

I like the D15-R68 clustering of try6-opt, but the positioning of the
H144 cluster is btter in try5-opt.

I'll have to look for hbonds in both, and try to get the best of both decoys.

Maybe I should add the D15.CG-R68.CZ closeness as a constraint.

30 Aug 2002 20:37  Kevin Karplus

I used Jonathan's new Hbond printing function to print the non-alpha
hbonds in try5-opt and try6-opt into try5.hbonds and try6.hbonds.
(There are still some bugs in the code---for example, it crashes when
trying to print the second file.)

The backbone H-bonds they have in common (omitting ones that are due to
distorted helices)

// strand 1-2
Constraint	L7.N	Q34.O	2.2 2.7 3.2
Constraint	L7.O	H36.N	2.2 2.7 3.2
Constraint	F9.N	H36.O	2.2 2.7 3.2

// strand 2-3
Constraint	T39.N	I85.O	2.2 2.7 3.2
Constraint	T39.O	L87.N	2.2 2.7 3.2

//  strand 4-5
Constraint	V134.O	K179.N	2.2 2.7 3.2
Constraint	Y136.O	W182.N	2.2 2.7 3.2
Constraint	P138.N	W182.O	2.2 2.7 3.2	// how?

// strand 5-6
Constraint	W175.N	F216.O	2.2 2.7 3.2
Constraint	Y181.N	G222.O	2.2 2.7 3.2
Constraint	Y181.O	D224.N	2.2 2.7 3.2
Constraint	T183.N	D224.O	2.2 2.7 3.2
Constraint	V184.O	L187.N	2.2 2.7 3.2
Constraint	G186.N	V227.O	2.2 2.7 3.2

// strand 6-7
Constraint	E228.N	R255.O	2.2 2.7 3.2

backbone Hbonds unique to try5-opt (omitting distored helices)

// far out on strand 2-3
Constraint	E45.O	W93.N	2.2 2.7 3.2
Constraint	G50.O	A99.N	2.2 2.7 3.2


backbone Hbonds unique to try6-opt (omitting distorted helices)

// strand 1-2
Constraint	F9.O	V38.N	2.2 2.7 3.2
Constraint	A12.N	V38.O	2.2 2.7 3.2

//strand 2-3
Constraint	V35.O	S82.N	2.2 2.7 3.2
Constraint	V37.N	A83.O	2.2 2.7 3.2

// strand 3-4
Constraint	S82.O	V133.N	2.2 2.7 3.2
Constraint	P84.O	T135.N	2.2 2.7 3.2
Constraint	P84.N	V133.O	2.2 2.7 3.2	// How?
Constraint	Y86.N	T135.O	2.2 2.7 3.2
Constraint	Y86.O	D137.N	2.2 2.7 3.2

Constraint	N139.O	G162.N	2.2 2.7 3.2

Constraint	S165.N	D212.O	2.2 2.7 3.2

Constraint	E228.O	A257.N	2.2 2.7 3.2
Constraint	A229.N	R234.O	2.2 2.7 3.2


31 Aug 2002 11:27 Kevin Karplus

Defined try7.constraints based on try5.constraints, but including
H-bonds that were compatible from either try5.hbonds or try6.hbonds.
Added hydrogen_bonds to define-score.script.

With these constraints, I expect try5-opt to score best---I'm
close:
T0173.try5-al8+T0173-2ebn-2track-protein-STR-global-adpstyle1.pw.a2m.gz:2ebn.7.40.pdb
scores best and T0173.try5-al10.8.40.pdb next best.


Did a similar thing for try8.constraints based on try6.constraints,
and the same few models came out on top.

I'll do optimization runs for both try7 and try8.

17:00 

With the try8 constraints and hydrogen bonds turned on, the best score
is for try8-opt-scwrl.  This makes a pretty barrel, but all the
conserved residues are scattered around the outside, not clustered.

With the try7 constraints and hydrogen bonds turned on, the best score
is also for try8-opt-scwrl, but try7-opt-scwrl also scores well.

If I had a potential strand somewhere after A282, it would be nice to
fit it into the barrel to complete it,  parallel to R255-L259 and
R6-H11.  The most probable position seems to start around D291:
	 ptgRAAALsnn
	  etDLLAGLg
  	  setpRLLFVHahp

Let's set up try9.constraints for the barrel:

     tpRLLFVHah
    rgaQVHVVTCTlg
   tgvsapIYlggagr
       rphVVVTYdp
      tvpKFYWTVlg
         gidAVVEAde
          ptgRAAALsnn
           etDLLAGLg
             tpRLLFVHah

which has n=8, s=8, like a TIM barrel.

try8-opt-scwrl scores best with these constraints.
Turning off hydrogen_bonds and turning up constraints to 0.2 
changes best to try8-try5-7.0.40

Let's optimize without the hydrogen bonds (to avoid crashes).
Two runs---one from scratch and one from the best few current decoys.
try9 will start from best current
try10 will start from scratch.

31 Aug 2002 Kevin Karplus 22:12

Best (with try9.constraints) is now try10-opt-scwrl.
Oops---still scoring with Jonathan's code, which is now known to be buggy.

Rescoring with try9.constraints and with proper undertaker code 
makes try9-opt-scwrl look best, and try10-opt-scwrl right after.
Both of these make somewhat raggedy but recognizable barrels.

Rescoring without constraints makes  try5-opt-scwrl look best, but the
barrel there is much more disrupted.

Let's do another optimization run, with constraints turned down a
little, starting from try9-opt-scwrl and try10-opt-scwrl.

1 Sept 2002 Kevin Karplus 09:56

Best scoring is now try11-opt, with no constraints.
If we crank up the break penalty to 25, the best is still try11-opt.

It looks like S279-E285 could be another strand of the barrel,
finishing off the barrel.

I'll try creating constraints for a barrel that looks like

     tpRLLFVHah
    rgaQVHVVTCTlg
     lgvsapIYlggagr
       rphVVVTYdp
        tvpKFYWTVlg
          gidAVVEAde
          ptgRAAALsnn
           ggsagarderg
             tpRLLFVHah

With this score function (using try12.constraints), try4-opt-scwrl looks best

We can try optimizing from there (try12), or starting from scratch.

2 Sept 2002 Kevin Karplus 11:37

Both try12-opt-scwrl and try13-opt-scwrl score better (with the
try12.constraints) than any of the earlier runs.

Both have some problems---try12 has a rather disordered barrel and
try13 has a topology error for getting from the strand ending at S260
to the one starting at G281---it runs back through the middle of hte
barrel, rather than around the outside with a helix.

Let's try doing an optimization run with both try13-opt-scwrl and try12.0.120
(I picked the first try12 run that scored worse than try13-opt-scwrl,
so that try13 would be very slightly favored, since bigger chunks of
its barrel look good to me.)


Tue Sep  3 14:32:26 PDT 2002 Rachel Karchin

best scorer (with try12.constraints) is now T0173.try14-opt-scwrl.pdb
try12.constraints are based on this alignment of the putative
TIM-barrel strands.

//     tpRLLFVHah             
//    rgaQVHVVTCTlg    
//     lgvsapIYlggagr
//       rphVVVTYdp
//        tvpKFYWTVlg
//         gidAVVEAde
//          ptgRAAALsnn
//           ggsagarderg
//             tpRLLFVHah


A number of strands have broken with these constraints.
R6-H11  looks good

Q34-T42 looks good

I85-Y86 (could be P84-G88) good (well-formed from I85-L87)

V132-Y136 broken (between V133 and Y136) but in reasonable position

K179-V184 not aligned right

A225-E228 broken (not aligned right) and
          hairpin forming between E228-A229 A233-R234

R255-L259 pretty good

G281-E285 pretty good

alignment is not right between last two strands and putative first
strand.

How to improve this alignment?
Also, A225-E228 is a real mess.  

the constraints betwen V132-T183 and A225-E228 do look good, but
they are not appearing in the decoy, which has W182 and E228
almost 11 A apart (NOT 2.7 A).

Let's try a few alternate alignments

for try15.constraints, use this alignment:

//     tpRLLFVHah             
//    rgaQVHVVTCTlg    
//     lgvsapIYlggagr
//       rphVVVTYdp
//       tvpKFYWTVlg
//        gidAVVEAde
//         ptgRAAALsnn
//          ggsagarderg
//            tpRLLFVHah



script for coloring the barrel:
  color-barrel.rasmol

also try to hold together the broken strand with a
length constraint ??

A225-E228 



with try9-constraints, best is currently try11-opt

This barrel looks much better, but it only has six strands
which are:

R6-H11
Q34-T42
I85-Y86
V132-Y136 (part)
K179-V184 (nicely aligned)
almost getting A225-E228 (actually just I223-D224)

let's try shifting A225-E228 back and aligning
R255-L259 and G281-E285 as they are in try12.constraints,
because these produced pretty good looking strands here.

try16.constraints based on:
alignment from try9-constraints with above changes
>R6-H11
..tpRLLFVHah
>Q34-T42
.rgaQVHVVTCTlg
>I85-Y86
lgvsapIYlggagr
>V132-Y136
....rphVVVTYdp
>K179-V184
...tvpKFYWTVlg
>A225-E228
....giDAVVEAde
>R255-L259
.....ptgRAAALsnn
>D291-L296
......ggsagardergwetDLLAGLg
>R6-H11-rep
........tpRLLFVHah

try9.constraints are just C-betas for some of the strand pairs
though, may want to add H-bonds.


..tpRLLFVHah
.rgaQVHVVTCTlg
lgvsapIYlggagr
....rphVVVTYdp
...tvpKFYWTVlg
....giDAVVEAde
.....ptgRAAALsnn
......ggsagardergwetDLLAGLg
........tpRLLFVHah

with try16.constraints (and before building any decoys), best scoring
decoy is T0173.try15.3.100.pdb
We'll throw this one in as one of the starting conformations.

Wed Sep  4 13:50:33 PDT 2002 Rachel Karchin

new best scores:
T0173.try16-scwrl.5.100.pdb
T0173.try16-opt.pdb

the barrel is starting to form with desired strand order.

A225-E228 strand is still broken.

try17 TO DO:

-- move out helix sitting in center of barrel 265-281
-- and add a length constraint to unwound helix piece 231-238
   and to 271-276 (central section of moved helix)

try18 TO DO:

Is D291-L296 strand flipped?  Check charges.
To flip it the right way, shift register by one with constraints.

Can any constraints be designed that use conserved residues?
Use the conserve script.

Do the barrel strands have charges on the right side?
Back-up D291-L296 strand by two?  How align it?



Wed Sep  4 17:25:31 PDT 2002 Rachel Karchin

best scorer is still try16-opt, which has the piece of helix
stuck in the barrel.  However, try17-opt (number 4 ranked)
has moved the helix out of the barrel, and the helix has
wound up nicely.  There is now a break between this piece 
and the final strand in the barrel D291-L296.

Time to do some work on D291-L296

yellow I85-Y86 and green V132-Y136 strands are not lined up right.
Need constraints here.

charges on D291-L296 look good.
for try18, we'll shift it up two.

Strand R255-L259 (strand 7) is in the wrong plane (almost perpendicular to 
the barrel).  Maybe fixing strand 8 will help.

alignment for try18:
//     tpRLLFVHah
//    rgaQVHVVTCTlg
//    lgvsapIYlggagr
//    rphVVVTYdp
//   tvpKFYWTVlg
//    giDAVVEAde
/      ptgRAAALsnn
//  ggsagardergwetDLLAGLg
//        tpRLLFVHah


try18 will start from try17-opt-scwrl

oops -- some CB constraints for G (and partners).  Fix these.

4 Sept 2002 Kevin Karplus

I noticed a typo in define-score.script ("include" missing the final
"e"), which is causing try18 to run without using the constraints.
This is probalby NOT what Rachel intended, so I'll start a try19 that
is similar, but HAS the constraints.

5 Sept 2002 Kevin Karplus

try19-opt scores best with the try18 constraints
(score-decoys-try18.rdb), but there seem to be some serious clash
problems.  Strand A83-G88 (strand 3) is not halfway between its
neighbors, and the barrel has a split between strands 1 and 8.  It
looks like strand 1 needs to be pulled back relative to strand 8, but
not relative to strand 2.  TO keep strand1-strand1 consistency, we may
need to move strands 1&2 or 1,2,3 as a unit.  Actually, the 1-2
alignment does not look great either.

What scores best without constraints?  We go all the way back to
try11-opt for that!  Strands 123 look good there, and 45 is pretty
close to being well attached also.  strand 6 is off, strand 7 is
questionable, and strand8 is missing.  Maybe we should pick up the
hbonds from try11-opt for strands 1-5, and just play with the 6,7,8
alignment.  Hmm, we already have try11.hbonds.

I manually extracted just the beta sheet hbonds into try11.beta-hbonds.

try18.constraints claims it is trying to get
    
     tpRLLFVHah
    rgaQVHVVTCTlg
    lgvsapIYlggagr
    rphVVVTYdp
   tvpKFYWTVlg
    giDAVVEAde
     ptgRAAALsnn
  ggsagardergwetDLLAGLg
        tpRLLFVHahb
but that can't be right, as there is not an advance of 8 around the barrel.
Reconstructing from the H-bonds, try18 is indeed trying to squish the
   barrel like that.

Let's start with strand 1 at the right spacing
      BEBEBEBEBE
     tpRLLFVHah
    rgaQVHVVTCTlg
      lgvsapIYlggagr
       rphVVVTYdp
        tvpKFYWTVlg
         giDAVVEAde
          ptgRAAALsnn
     rdergwetDLLAGLg
             tpRLLFVHah


With these modified (try20) constraints, try10-opt-scwrl and try11-opt
score best.  Let's optimize from both of them.


Thu Sep  5 12:33:42 PDT 2002 Rachel

This brings us back to the alignment we used at the beginning of
the week (except for shifting strand 8).  

try21-opt-scwrl.pdb
We've got five strands in the barrel, but things fall apart for 
strands six, seven and eight.


     BEBEBEBEBE
     tpRLLFVHah
    rgaQVHVVTCTlg
      lgvsapIYlggagr
       rphVVVTYdp
        tvpKFYWTVlg
         giDAVVEAde
          ptgRAAALsnn
   rdergwetDLLAGLgftasgt
             tpRLLFVHah


try22-opt is looking better. Still alot of breaks.
Strand 7 (ptgRAAALsnn) is in the wrong plane.  

Kevin has started a new run with strand 8 shifted:

//      BEBEBEBEBE
//     tpRLLFVHah
//    rgaQVHVVTCTlg
//      lgvsapIYlggagr
//       rphVVVTYdp
//        tvpKFYWTVlg
//         giDAVVEAde
//          ptgRAAALsnn
//    rgwetdllaGLGFTasgt
//             tpRLLFVHah


Thu Sep  5 17:53:12 PDT 2002	Kevin Karplus

try23-opt did not make much improvement in the looks of the barrel.
I'll score with try23.constraints, try22.constraints,
try20.constraints, and no constraints.  At this point, I think we
won't be able to get the barrel together.

Rachel, please look through the "decent" barrels and pick out which 5
we should submit.  If you want, I'll do a polishing run on the
whatever is the best of them.

Best scoring:

For try23.constraints:		try23-opt-scwrl
For try22.constraints:		try23-opt-scwrl
For try20.constraints:		try22-opt
For no constraints:		try11-opt


I'll try one more run with the try23.constraints, but turning up
optsubtree, which rarely got used on try23.

5 Sept 20:27 Kevin Karplus

The new best score with try23.constraints is try24-opt, but it looks
terrible--strand 2 is broken, strand 6 crosses strand 5 and the helix
after if has come away from the outside of the barrel and sits on top.

I think that try22-opt is the best-looking barrel so far, but try11 is
also pretty good.  I have an idea what I want the barrel to look like,
but I don't have the tools to get it there.  I give up.


model 1:	try22-opt
model 2:	try11-opt