Thu Jul 25 14:26:09 PDT 2002
t0162

25 July 2002 Kevin Karplus

The predictions for T0162 are from many different families, with no
strong scores.  The alignment is fairly deep though, so the signals
should be pretty strong.

I fear this may be an ab-initio or very difficult fold-recognition problem.

26 July 2002 Kevin Karplus

The first undertaker run (try1-opt-scwrl) has exploded beta
sheets---we'll have to try to add constraints to hold the sheet together.

Tue Aug 20 17:30:22 PDT 2002

Modernized score function a bit, added try-alignments target to
Makefile, and added more alignments to the initial set.  
Will try a second run to see if anything comes out.

There are several robetta models (split into subdomains).
Let's try a run to put them back together and see how they do (try3).

I think we'll see some large anti-parallel sheets forming, which we
can piece together manually with constraints (at least, I hope so).


21 Aug 2002 Kevin Karplus 05:58

The best score is for try3-robetta5.10.20 (the try3 run failed halfway
through when it was unable to read a returned SCRL file---I think I
need to improve the test for which SCWRL file to use, as SCWRL
sometimes leaves an empty file).
The try2-opt-scwrl run was also not a bad score.
Neither one is a particularly convincing fold, and both have trouble
with predicted strands being wound up into helices.

I'll do one run with initial points
	try3-robetta5.10.20
	try2-opt-scwrl
	try3-robetta3.0.20
	try3.9.20
turning up pred_alpha

Thu Aug 22 17:13:32 PDT 2002 Kevin Karplus

try4-opt is new best score.
R172-W177 are still coiled up into a helix, but there are some good
sheet pieces.  Someone needs to look at the pieces and try to add some
constraints to assemble them.

26 August 2002 Rachel Karchin

residues 140-220 are an antiparallel beta sheet as we have either four
or five consecutive strands along the chain connected by turns.  

For this region of the protein, I'll label the strands like this (where
residue ranges are very approximate).

E1 140-148
E2 151-160
E3 172-182
E4 185-199
E5 203-216

In the best scoring decoy (with current score function) try4-try3-10-scwrl.10.80
E1 and E2 are forming a nice hairpin
E3 is curling up into a helix

E4 and E5 might form a hairpin, but the turn between them is too short.

We can also observe that E4 and E5 are much longer than E1,E2, E3.

With the hypothesis that these are an anti-parallel sheet, I'll try
to guess on how the sheet might assemble to get a set of constraints. 

E1>       CTVYGKSI 
          |||||||              
E2<     EICAIITQQ
        |||||||| 
E3>   RWRSEWKFTI
      ||||||||||
E4<  EYYHVQIKLVAAVQ
        |||||||||||
E5>     NVQLVSHKDIQDSVQ

for try5, I'll use C-beta constraints at the ends of the putative
strands to try and form the anti-parallel sheet

These pairing guesses were made using the residue-specific
cross-strand potential epsilon-0 of Zhu and Brawn (ProtSci 8: 326-342 1999)

Also want to keep E3 from curling 

W173 - I181 at 28 A apart  (3.5 A /per residue on sheet)


E1-E2
E3-E4
E4-E5

are connected by short turns.

but E2-E3 is connected by a loop that appears to be around 12 residues
long.  This might be a Greek key motif.

27 Aug 2002 02:15 Kevin Karplus

I noticed that Rachel's try5 run failed because SCWRL returned a bad
file, and I still haven't put in proper checks for that.  I'll rerun
as try6 without scwrl_each, so that it will complete before Rachel checks it.

27 August 2002 Rachel Karchin

best scoring now is T0162.try6-try5-3.0.100.pdb but the sheet
has not assembled yet.

I've made a crude alignment for an anti-parallel sheet (see above)
and will put in C-beta constraints to see if this helps.

28 August 2002 Rachel Karchin

try7 constraints are doing badly, destroying the sheet
that was forming. 

Let's go back to what looks good on try6-opt-scwrl
and try to lock it in.

         
E1>       CTVYGKSi       140-148
          | | | |
E2< FQHSEICAIITQQ        163-151    
     | | | | |
E3> RWRSEWKFTIT          172-182
    | | | | | |
E4<EYYHVQIKLVAAVQA     200-186
     | | | | | | |
E5>NVQLVSHKDIQDSVQ     203-217

// this is how E4 and E5 are lined up in try6-opt-scwrl

alternate -- work on this one below if above stinks

E4<EYYHVQIKLVAAVQA     200-186
     | | | | | | |
E5>  NVQLVSHKDIQDSVQ   203-217


29 Aug 2002 Kevin Karplus 09:50

There are errors in the constraint files:
	A157, I159, S161, Q159 don't exist
These look like they should be A156,I158,S160,Q162.
E159, H161, F163.  It is part of an anti-parallel sheet:

<     gdISKGYVTCfgnp
151>     QQTIIACIESHQFqpk
<        ppTITFKWESRWRGNWFn
185>     tAQVAAVLKIQVHYYEdg
<       svQVSDQIDKHSVLQVN

Incidentally, I think the register for the strand starting at Q151 may
be wrong---I'd slide it over 2, though this may disrupt the C-C pairing.

<     gdISKGYVTCfgnp
151>   QQTIIACIESHQFqpk
<    ppTITFKWESRWRGNWFn
185> tAQVAAVLKIQVHYYE
<   svQVSDQIDKHSVLQVN

With the current constraints, try8-opt scores best, but it has
absolutely enormous breaks.

I'm also not sure we know the hbonding patter well enough to specify
it---let's try just CB constraints.

I've put together such a set of constraints (for my proposed sheet) at
try9.constraints.   With these constraints, the best score is for
T0162.try8-try6-1+T0162-3gbp-v...+T0162-1dzgI-vit-adpstyle1.pw.a2m.gz:1dzgI.4.100.pdb	

I propose doing another run, starting from a few of the good scoring
models from previous runs.

Thu Aug 29 12:25:48 PDT 2002 Rachel Karchin

Okay.  Looks like Kevin has started try9.

some comments from looking at the STR predictions:

There are two anti-parallel sheets connected by a helix,
so the last strand on sheet 1 and the first strand on sheet 2
are likely to be parallel.


roughly:

sheet 1

P64-E68  Q73-T77  G85-D89  I94-L101
   up      down      up      down  


sheet 2

F140-I148  Q151-F163  R172-T182  A186-E200  N203-H209
   down      up         down       up          down

with latest score function, we're getting alot of helix where
strand is predicted (

maybe we should turn down dry6.5 and turn up predicted alpha
which seems to match the STR predictions of strand locations
pretty well.  First, I'll try this on the existing decoys.

with dry6.5 turned down from 9 to 6 and pred_alpha turned up
from 1 to 2

T0162.try9-try8-4.0.100.pdb is the top scorer

this decoy has some sheet forming around the hoped for sheet 2
but it is quite broken up.  The predicted sheet at the c-terminal
is curling up into a helix, but this is not a very strong prediction.

The fourth strand in sheet 1 is definitely forming a helix -- and we have
VERY strong strand prediction here (although the prediction is weaker
when encoded in the ALPHA alphabet).

Let's try a run with pred_alpha set to 2
This will be try10

------------------------------
29 Aug 2002 Kevin Karplus	15:35

try9-opt is new best score (even with the pred_alpha2 turned up), but
it looks terrible.  Lots of predicted strands wound into helices...

One nice feature: C141 and C157 come close---maybe we should add an SSbond?
Of course, their coming close is a result of all the constraints on
strands E1-E2.

It looks like we need straightness constraints on the strand.



From karplus@bray.cse.ucsc.edu  Thu Aug 29 15:42:36 2002
Date: Thu, 29 Aug 2002 15:41:36 -0700
From: Kevin Karplus <karplus@soe.ucsc.edu>
To: karplus@soe.ucsc.edu, rachelk@soe.ucsc.edu, weber@soe.ucsc.edu,
   learithe@cats.ucsc.edu, yael@biology.ucsc.edu, baertsch@soe.ucsc.edu,
   rph@soe.ucsc.edu, afyfe@soe.ucsc.edu, jcasper@soe.ucsc.edu,
   oscarhur@soe.ucsc.edu
Subject: t0162


Now that t0161 is making some progress (we have something that could
be submitted, in a pinch), I looked a bit at t0162.

This seems to be a bit more desperate, as our strands are curling up
into helices.

I'll try adding some straightness constraints and start a try11 run.

Thu Aug 29 17:08:55 PDT 2002 Rachel Karchin

The decoys builds are proceeding slowly . . .

In the meantime, I did a bit of research about F-actin capping protein
alpha-1 subunit.

Actin-capping proteins are heterodimers with an alpha and beta
subunit.  In vertebrate cells, there are three alpha-subunit isoforms
and three beta-subunit isoforms.  The actin binding appears to be
in the beta-subunit.  (Function is to cap the barbed ends of the
thin filaments and anchor them to the "Z-line").

Since T0162 is not a monomer, it may be particularly hard for undertaker 
to get it right.

I'll try and find out if the structure of any of the beta-unit forms
has been solved.  Hmm.  Doesn't look like it.


29 Aug 2002 19:51 Kevin Karplus

Best score (with try11.constraints) is try11-opt.
This is broken into many pieces---perhaps because JiggleSubTree and
OptSubtree were called a lot, and CloseGap  was called rarely.

Let's do a "healing" run with insertfragment and closegap turned up
and Jigglesubtree turned way down.  (this will be try12)

The big question now is whether our constraints are at all correct, or
whether adjusting some of the strand registers or pairings would
produce a more compact, more contiguous model.


Thu Aug 29 20:31:52 PDT 2002  Rachel Karchin

try8.constraints seem to have worked better than the try11.constraints
try8 is also broken into many pieces, but sheets are beginning to
form in the try8-opt decoy -- looks more promising than try11-opt.

In many ways, we've gone steadily downhill since try6-opt, which is
not at all compact, but has extensive sheet formation.  the try8
constraints were based on the sheets observed in try6-opt.
try6-opt has some pieces of curly helix, but lots of good sheet and
all these sheets formed with minimal constraints!  (see try5-constraints)
Also, try6-opt has the two cysteines C141 and C157 positioned to form a
disulfide bridge -- they are nicely lined up on two anti-parallel
strands that are part of a hairpin, pointing at each other and
touching.  And this is not the result of one of our constraints.

I really do think we want to keep these cysteines in the same register 
when we put in constraints for the beta strands.  The try11.constraints
currently have them shfited away from each other by two.

try11-opt has the cysteines close, but not pointing the right way to
form a bridge.

try8-opt has the cysteines pointing towards each other and touching.


29 Aug 2002 21:27 Kevin Karplus

With the try8-constraints, 
T0162.try8-try6-1+T0162-1a2vA-2track-protein-EBGHSTL-local-adpstyle1.pw.a2m.gz:1a2vA.5.100.pdb
scores best.

With no constraints, try7-opt-scwrl scores best.

With the try7 constraints, try10-try8-4.3.100 scores best.

With the try11.constraints, try12.0.60 scores best (try12 run not
finished yet).

Date: Thu, 29 Aug 2002 21:40:16 -0700
From: Kevin Karplus <karplus@soe.ucsc.edu>
To: rachelk@soe.ucsc.edu
Subject: t0162


Have you looked at try7-opt-scwrl?  It scores best without constraints now.

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

29 Aug 2002 22:40 Kevin Karplus

try12 finished.  It is still rather a mess, but it scores the best
with the try11 constraints.


Fri Aug 30 08:43:52 PDT 2002	Kevin Karplus

try10 finished.

Although optimized with the try9 constraints, it does not score as
well with those constraints as try12-opt.  Try10-opt-scwrl does score
best with the try7 constraints, though try7-opt-scwrl still scores
best without constraints.

(Scoring results in score-decoys-try7.rdb, score-decoys-try9.rdb,
score-decoys-try11.rdb, and score-decoys-no.rdb)

try10-opt-scwrl does have the helices clustered around the strands and
parallel to them---it just hasn't formed a clean sheet.

I'm starting a short polishing run (try13) with no constraints from
try6 through try12.  I'll start a similar short run with the
try7.constraints (try14).


From: Rachel Karchin <rachelk@soe.ucsc.edu>
Date: Fri, 30 Aug 2002 09:18:00 -0700
To: Kevin Karplus <karplus@soe.ucsc.edu>
Subject: Re: t0162

Kevin Karplus <karplus@soe.ucsc.edu> writes:
> 
> Have you looked at try7-opt-scwrl?  It scores best without constraints now.

Looks pretty broken up, and the cysteine pairing is not that good.

-- rk


--------------------------------------------------
From karplus@bray.cse.ucsc.edu  Fri Aug 30 09:35:32 2002
Date: Fri, 30 Aug 2002 09:35:31 -0700
From: Kevin Karplus <karplus@soe.ucsc.edu>
To: rachelk@soe.ucsc.edu
CC: karplus@soe.ucsc.edu
Subject: t0162


Pick your favorite set of constraints, or make up new ones from the
best pieces you've seen in different models, and start a new run on t0162.
I've go no strong feelings on this protein---I like to have the
helices clustered around the sheet, but other than that ...

--------------------------------------------------
Fri Aug 30 10:57:25 PDT 2002 Kevin Karplus

try14-opt is rather foamy. Like most of our models, it has a large
hydrophobic face.  This could be the oligomerization interface, or it
could be a bad mistake.

The strands aroung V191-H197 and W177-F179 are rather broken up.
Could the try7 constraints be inconsistent in this area?


Fri Aug 30 11:09:55 PDT 2002 Rachel Karchin


Best scoring with the minimal try5 constraints (which produced
our most sheet-forming decoy) is T0162.try13-try7.0.120.pdb

try10-opt-scwrl.pdb has some compact features.
Maybe try a run that starts with this one and includes constraints 
that match good sheet structure we have in existing decoys

try6-opt has good hairpins for 
P64-E68  Q73-T77

G85-D89  I94-L101

F140-I148  Q151-F163

and a hairpin with two very long strands from
A186-E200  N203-H217


try10-opt-scwrl has good hairpins

T142-K146  T153-C157

"almost hairpin"
V196-Y199  N203-H209

Since try6-opt has so much good sheet, let's just
go back to the constraints that mimic them
try8-constraints, check for mistakes and rerun only from
try6-opt conformation.

try15-constraints will only have try8-constraints that are found
in try6-opt (no guesses).  Plus a constraint for the disulfide
bond between C141 and C157.  Also try InserSSBond and ImproveSSBond
set to 0.1


Also try a simultaneous run, try16  with no disulfide constraint
and disulfide scoring functions turned off.

These runs have produced conformations that are pretty much
indistinguishable from try6-opt (so far)

I'll start another run, try17, and include other conformations in addition
to the constraints (try15-constraints)


?? S147 near I181
   Q151 near D99

try18 will have weak constraints to try and pull together
the two putative sheets (otherwise same as try17)

constraint I148.CB I181.CB 3 5 9
constraint R102.CB Q151.CB 3 5 9


try19  

need more constraints to pull together this sheet

parallel 

I181.CB G145.CB  5.12
T180.CB Y144.CB   "
F179.CB V143.CB   "
K178.CB T142.CB   "
W177.CB C141.CB   "
E176.CB F140.CB   "

straightness constraints for curly helix
W173-F179  20 21.9 24
S175-I181  "   "   "

break off curly helix piece

161-171  (move from 171 coords to 139 coords)
about halfway



shorter runs -- quick results (about an hour)

gen 100 iter 1

jiggle subtree down
opt subtree up

try19 does not use the second set of c-beta constraints
(which tries to line up strand 98-102 and strand 151-155)
but try20 does:

constraint R102.CB Q151.CB  3 5.12 9
constraint L101.CB  Q152.CB 3 5.12 9
constraint H100.CB  T153.CB 3 5.12 9
constraint D99.CB   I154.CB 3 5.12 9
constraint F98.CB   I155.CB 3 5.12 9

This is the only difference between them

No decoys yet from try19 or try20:

best scorers with try20-constraints:
T0162.try15.2.120.pdb
T0162.try15.3.120.pdb
T0162.try17-try6.3.120.pdb

best scorers with try19-constraints
T0162.try17-try6.2.120.pdb
T0162.try16.1.120.pdb

best scorers with no constraints

T0162.try17-try6.0.120.pdb


From: Rachel Karchin <rachelk@soe.ucsc.edu>
Date: Fri, 30 Aug 2002 20:24:59 -0700
To: karplus@soe.ucsc.edu
Subject: no decoys yet for try19 or try20 

I've had two tries (19 and 20) running for over an hour, which
implement the various constraints and moves we discussed
(details in the README).  Neither has produced any decoys yet.

Everything is detailed in the README.

For existing decoys, I've listed best
with no constraints, and constraints from my last two tries.  
Please look at the README for explanation of what is difference 
between 19 and 20.

I've got to meet Lew at a restaurant in 10 minutes.

If you can wait another hour or so, it would be quite worthwhile
to see if try19 or try20 produce any good scoring decoys.

If you don't want to do this:

The current best decoys look very similar to me, I'd just go
with the best scorer (perhaps the no constraints one).

T0162.try17-try6.0.120.pdb

My second pick, for an alternate fold, would be:

try10-opt-scwrl

-- rk


30 Aug 2002 Kevin Karplus 21:27

Best with try19.constraints:

T0162.try19-opt		extends pair F87-I155 to R86-I154
		strand E175-T180 almost part of sheet.

Best with try20.constraints:

T0162.try20-opt.	strand E175-T180 nowhere near reasonable position.


Best scorer now (21:42) with no constraints:

T0162.try16.4.120	pairs long and short hairpins at F87-I155

Upping the break cost and tweaking other parameters a bit
moves try5.1.80 to the top, which I think looks better than try16.4.120.

Of the "other" series, we should look for somthing that scores well
with try7, try8, or try11 constraints:

With try7.constraints:	try14-try10.0.120	strands pretty shattered
With try8.constraints: T0162.try8-try6-1+T0162-1a2vA-2track-protein-EBGHSTL-local-adpstyle1.pw.a2m.gz:1a2vA.5.100.pdb	
	The try8 model looks better to me than the try14-try10.0.120
	
With try11.constraints:	try12.2.60
With try19.constraints: still try19-opt
With try20.constraints:



model1:	try5.1.80
model2:	try19-opt
model3: T0162.try8-try6-1+T0162-1a2vA-2track-protein-EBGHSTL-local-adpstyle1.pw.a2m.gz:1a2vA.5.100.pdb	
model4:	try14-try10.0.120