\magnification=1200
\baselineskip=20pt
\nopagenumbers
\font\big=cmr12 scaled \magstep2

\centerline{\bf STANFORD UNIVERSITY}
\centerline{\bf DEPARTMENT OF STATISTICS}
\centerline{\big DEPARTMENT SEMINAR}
\bigskip

\baselineskip=12pt
\centerline{4:15 p.m., Tuesday, May 2, 2000}
\centerline{Sequoia Hall Rm. 200}
\centerline{(Cookies at 3:45 p.m. in 1st Floor Lounge)}

\bigskip
\baselineskip=15pt
\centerline{\sl Prof. Minping Qian}
\centerline{\sl Department of Probability \& Statistics}
\centerline{\sl School of Math, Peking University}
\bigskip
\centerline{\bf Promoter finding and linguistic}
\bigskip


According Fickett and Wasserman [1] " TFBSs stand out clearly against a
non-conserved background" , but  the problem of  promoter prediction based
on over-representation of TFBSs  remains unsolved. The difficulty is that
there are many TFBS-like patterns in non-promoters that mimic real TFBSs
and lead to many false positives in promoter prediction. Since we have more
than 2000 TFBSs, there will be more  2,ooo,ooo pairs and 2 billion of
triples of TFBSs. We introduce the relative entropy and a method for mining
in data for the biological significant words--the over-represented words,
and regulations
for combination of words into phrases by hunting the over-represented TFBS
pairs from non-conservative almost random back ground, is proposed. Our
statistical results suggest that TFBSs come in well-defined pairs and
triples, etc. In this way a stochastic regular grammar would hopefully be
obtained.

\bye