Tue Jul 30 03:44:50 PDT 2002
t0164

30 July 2002 Kevin Karplus

This looks like another homology modeling target, with 2bnh or 1a4yA
as the best templates (SCOP c.10.1.1).

On the try1 run, I forgot to include scwrl_each in the OptConform line.
I'll have to remeber to fix that for the try2 run.

Iteration 0 of try1 shows a very nice alternation of helices and
strands in a parallel sheet, but iteration2 shows it breaking up.
We may have to put constraints in to replace the missing hbond
function, or increase the dry weights and phobic_fit parameters, which
favor the more compact sheet.  (Let's wait though until all the
iterations are done---iteration 1 may be an anomaly.)


Tue Jul 30 18:09:02 PDT 2002 Kevin Karplus

Yes, try1-opt has destroyed the nice beta sheet.
There is no point to starting try2 until some constraints to hold the
sheet together have been added.

Wed Jul 31 10:05:01 PDT 2002	Kevin Karplus

The beta sheet in 1a4yA forms into a beautiful cylinder (well, helix
actually), but T0164 has just 5 strands and 6 helices of the repeating
unit.  

T0164 is just the C-terminal part of tropomodulin, which is believed
to be tightly folded [Kostyukova A, Maeda K, Yamauchi E, Krieger I,
Maeda Y.  Domain structure of tropomodulin: distinct properties of the
N-terminal and C-terminal halves.  Eur J Biochem 2000
Nov;267(21):6470-5 ]

It would be nice to be able to extract the hbond constraint
information directly from the T0164.t2k.undertaker-align.pdb.gz file.
If we had an Hbond detector, we could collect all the hbonds from the
alignments and average the distances to get a set of constraints. We
could also weight them by how often they occurred.

Let's add 8 Hbonds to hold the 5 short strands together:
V200.N	K228.O
V200.O	S230.N

F229.N	S256.O
F229.O	N258.N

L257.N	D284.O	
L257.O	R286.N	

L285.N	K314.O	
L285.O	G316.N

With these constraints added, the best decoy is T0164.try1.12.20.pdb,
with iterations 3, 7, 9, 16, 18, 10, 5, 8, 11, 19, 17, 15, 4, 0 also 
scoring well.  These are ones that did not blow up the beta
sheet---about 3/4 of the iterations.  Since so many of them did well,
let's start another run with all the current decoys as starting points.
The ones that blew up will get quickly discarded.

Thu Aug  1 10:08:25 PDT 2002 Kevin Karplus

try2-opt-scwrl looks pretty good, and stays the lowest cost even when
I play with the weights of the score function somewhat.  There are
still some breaks, and there is a "crack" that exposes some of the
hydrophobic core.  I'll rerun adding in a dry5 term and increasing the
penalty for breaks.  I'll also increase the constraint weight
somewhat, to keep the beta sheet from getting stretched.


1 Aug 2002 Kevin Karplus

try3-opt-scwrl is new low cost.  The C-terminal helix has wandered out
of place---we'll have to add some constraints to staple it down.

21 August 2002 Kevin Karplus

Added a couple of weak constraints for C-terminal helix, and
increased the dry6.5 and dry8 weights.  Let's try optimizing
with new score function, starting from the best-scoring current decoys:
	try3.11.40, try3.10.40, try3.14.40, try3.7.40

Thu Aug 22 10:41:24 PDT 2002

Well----in try4-opt the final helix is back where it belongs, but the
end has gotten curled and strands 4 and 5 have gotten broken.

Let's add a constraint that the final helix remain straight and up the
break penalty.  Hmm, try4-opt still looks best.

Let's reoptimize it.

Thu Aug 22 13:39:14 PDT 2002

DAMN---all that work and CASP5 has withdrawn the target!