Evaluating the Signi cance of Sequence

Sdiscover is a tool capable of nding subsequences, possibly separated by arbitrarily long gaps, in a set of sequences. These subsequences are referred to as motifs. This paper proposes a method to evaluate the signiicance of the sequence motifs found by Sdis-cover. The method is based on the minimum description length principle and Shannon's coding theory. The equivalence of the proposed method to the Bayesian inference is also discussed. 1 Introduction As the Human Genome Project 4] is expected to complete in a few years, research focus has been shifted from sequencing the biological data to mining and interpreting these data 1, 7, 9]. The interesting patterns to be mined range from genes 3], to DNA or protein sequence motifs 2, 10], to protein and RNA structure motifs 5, 9]. In this paper, we consider the problem of evaluating the signiicance of sequence motifs found by our pattern matching tool, Sdiscover 10]. Given a set of sequences, the motifs of interest are in the regular ex