During FY96, we expanded our linear Hidden Markov Model software into 
a functioning protein threader.  Using this threader as our primary
tool, 
we are competing in the fold recognition section of the Second Critical
Assessment of Techniques for Protein Structure Prediction (CASP2)
contest.
We have built a model library that contains models for over four
thousand 
protein families, and based on this library we have submitted X
structure predictions.  Contest results will not be available until
December; however, we have learned the actual fold of two targets on
which we predicted, and in both cases we predicted the correct fold.

To suport our entry in this contest, we have enhanced our tool suite 
to include a graphical, RASMOL-based alignment viewer, and scoring
options
to adjust for model length.  Upcoming features include a weighting
scheme which builds on phylogenetic relationships and emphasizes the
alignment columns most critical to the structure or function of the
protein.  We presented posters on this weighting scheme at ISMB96 and
the 1996 Protein Society Symposium.

We have continued development of a neural net method for modeling beta
sheet interactions.  Based on the protein sequence, plus other physical
properties including chain separation and beta strand membership, we
estimate the likelihood that two beta strands will be in contact.  We
plan to incorporate this into our threader during FY97.