Thu Jul 25 14:26:24 PDT 2002
t0163

25 July 2002 Kevin Karplus

This is a very obvious homology model for 1el5A or 1l9fA, with
excellent domain hits to other templates.

I killed the first undertaker run after it had output the
Template.atoms file, because it was running on a machine that had a
wedged X server taking up half the CPU, and because I had forgotten to
include the large set of alignments to the top templates.

Hmm, I just noticed that I forgot to add "scwrl_each" to the second run.
Oh well---I'll do a third run in the morning, with the new operators I
added tonight.  Perhaps I have to turn down the probability of the new
operators a bit, as they seem to be about the only things being tried.

26 July 2002 Kevin Karplus

I rewrote undertaker to have a simplified interface to the jiggle and
opt operators, randomly trying different ways to pick the atoms and
jiggling a whole subtree or just a segment.

Tue Aug 20 18:06:54 PDT 2002

Modernized the score function.  try3-opt-scwrl and try3-opt score best.
This looks almost done---let's start polishing on try4.

21 Aug 2002 Kevin Karplus 06:20

try4.4.100 is current best score, but process is still running.  It is
about 70% done after 12 hours, so has about 5 more hours to run.

I was wondering if the polishing runs should raise the probability of
OneRotamer, as it is never tried, but might be useful in the final
polishing. 

Hmm---maybe this is not ready for final polishing, as K251-G258 seems
misaligned. 

I've added the try-heavy target and alignments, and will start another
run from scratch.  I don't expect try5 to beat try4 (which has had
heavy polishing) but I'll see whether it beats try2 and gets a better
alignment for K251-G258.  If it does, I'll polish it.

Wed Aug 21 12:41:15 PDT 2002

try5-opt-scwrl is beating all of tries 1 2 and 3.
Let's try polishing it.

21 Aug 2002 19:25 Kevin Karplus

try6-opt is new best.  It has reduced breaks somewhat from
try5-opt-scwrl, and has done a lot of OneRotamer optimizations.
I'll try another optimization run, with Optsubtree and CloseGap
turned up a bit, but I think we're getting diminishing returns.

22 Aug 2002 10:33 Kevin Karplus

try7-opt is new best---slightly better than try6-opt, but not by a
lot. We might as well give up and submit this one.