Higher Order Mining of Biological Motifs

Abstract. A well-investigated problem in Bioinformatics is that of identifying regulatory functions from the genome, where pattern discovery techniques are typically adopted. In this paper we face the problem of supporting the interpretation of biological motifs by investigating their possible relationships expressed by spacer. The idea is that of discovering frequent combinations of related motifs, since significant co-occurrences of motifs suggests that their association can be important by a functional viewpoint. The proposed approach is an example of higher order mining, since a data mining step, namely frequent pattern mining, is applied to results of a previous mining step (identification of initial motifs pattern). We have experimented our approach on motifs extracted from untranslated regions (UTRs) of nuclear transcripts targeting mitochondria and preliminary results show its usefulness in supporting their interpretation by biologist.