Minimizing the Number of False Positives in Binding Site Prediction

of the regulation mechanism for gene expression.
Binding sites of transcription factors are key
elements of regulatory networks. The prediction
of binding sites therefore provides insight into
the structure of these networks. Identifying
binding sites in the promotor region may lead to
information on the function of a gene.

Usually only a few binding sites are known
experimentally. An important task is to deduce
the global binding characteristics of a protein
based on few data points. Recent approaches often
make use of Scoring Matrices. The well-known
Profile or Position Weight Matrix will be reviewed
as an example.

Then a new way of computing a Scoring Matrix
based on the maximum likelihood principle and
quadratic programming techniques will be introduced.
Our approach minimizes the number of false positives,
thus allowing substantial improvements in
specificity.