Mon Jun 24 08:58:25 PDT 2002

It is clear that T0137 is a homology modeling target, with the various
b.60.1.2 domains (homologs of 1hmt) as the template.

We should probably restrict our attention to templates that have
E-values < 0.0001.

25 June 2002 Kevin Karplus

The structure in T0137/try1-opt.pdb looks very nice---it could be
perhaps a bit more densely packed, but it looks quite nice as is.
It is quite similar to the high scoring alignments to 1crb---perhaps
we should use Dali to align our current best guess to 1crb, 1hmt,
1lid, ... and see if that alignment differs from the ones we get by
sequence techniques.  There are 32 templates listed in the
T0137.best-scores.rdb file all with phenomenally good scores, so any
of the alignments should be within about the same range as we see in
the FSSP alignment of the templates (< 3 Angstrom).

We may want to pick a template with a more selective method than the
fold-recognition HMMs, since the structures are awfully close.
I put together a little script that looks at the SW scores (based on
single sequence), and it suggests that 1lid is the best of the
templates. 

There are 10 1lid alignemnts in T0137.undertaker-align.pdb, and they
don't all agree---perhaps we should simply try starting with all of
them and see what happens.

25 June 2002 Kevin Karplus

T0137.try2-opt.pdb looks quite good.  There's a bit of a loose area
where the barrel doesn't quite close (between strands 61-65 and
71-74).   I wonder if we could add constraints to get hbonds to form?
We'd pair 65 with 71, 64 with 72, 63 with 73, and 62 with 74.
The hbonds should be on 62 and 64 and 66.


26 June 2002 Kevin Karplus

The try2-opt run scores better than the try3-opt run, even with the
new constraints.  Perhaps we need to do a run using 30 or so of the
best decoys as starting points?  After that run we may want to do a
DALI alignment with some of the best possible templates---perhaps all
65 of the 1hmt homologs from FSSP.  I suspect that 1lid will come out
best, since it was used as a template, but this would help us pin down
the "right" alignment to start from.

try3-opt still has the break in the barrel between the same strands.

11 Aug 2002 Kevin Karplus

Reran make with the new template library and remade sw-best.
Added try-alignments target with best templates from best-scores or sw-best.
Updated define-score.script and undertaker.script.

Fri Aug 16 13:51:02 PDT 2002 Kevin Karplus

try4-opt-scwrl looks quite good--only one small break visible at
G33-N34 (not in the barrel).

Let's tweak the score paramters a bit to favor better packing and
penalize breaks more, and do one last cleanup run.


17 Aug 2002 Kevin Karplus

try5-try4.8.60 is new best (beating out try5-opt).  
Let's submit it, since further optimization is unlikely to make
improvements.