*************************************************************************

                Department of Mathematical Sciences                       
                   The Johns Hopkins University                           
                                                                        
                              SEMINAR                                
                                                                        
*************************************************************************

Lenore Cowen                                      February 22, 2001
Department of Mathematical Sciences               304 Whitehead Hall
The Johns Hopkins University                      Preseminar: 3:00 p.m.
                                                  Refreshments: 3:30 p.m.
                                                  Seminar: 4:00 p.m.

*************************************************************************

     PREDICTING THE BETA-HELIX FOLD FROM PROTEIN SEQUENCE DATA

*************************************************************************

                             ABSTRACT

The so-called "grand-challenge" problem associated with protein folding
is predicting a protein's three-dimensional fold when given only its amino
acid sequence.  This complex problem is of crucial importance because the
shape of the fold is the key to  understanding biological function, and
while amino acid sequences are rapidly being determined in the laboratory,
it is very difficult to determine the three-dimensional structure.

An easier, but related problem, is the "motif-recognition" problem.
Proteins are characterized into families, also called motifs, based on
their folds.  Some common motifs include those that consist primarily of
alpha helices, such as the coiled-coil motif, and those that consist
primarily of beta sheets, such as beta-helices or beta-barrels.  While
there has been much recent success in predicting alpha-helix motifs
from sequence information, predicting beta-structural motifs has
seemed to be much harder.  In beta structures, some amino acids that
are nearby in space may be far away in the linear sequence, making
finding statistical correlations a much harder computational problem.
 
We present new probabilistic techniques for incorporating these
long-distance correlations based on three-dimensional information, and
discuss how we are using them to build a predictor for a particular
beta-structural motif, namely, the beta-helix motif.

(This is joint work with Phil Bradley, Matthew Menke, Jonathan King, and
Bonnie Berger.)

*************************************************************************