Identifying Potential Regulatory Elements by Transcription Factor Binding Site Alignment Using Partial Order Graphs

Abstract

Identification and functional characterization of regulatory elements in the human genome is a challenging task. A sequence feature commonly used to predict regulatory activity is the co-occurrence of transcription factor binding sites (TFBSs) in regulatory regions. In this work, we present a graph-based approach to detect frequently co-occurring TFBSs in evolutionarily conserved non-coding elements (CNEs). We introduce a graph representation of the sequence of TFBSs identified in a CNE that allows us to handle overlapping binding sites. We use a dynamic programming algorithm to align such graphs and determine the relative enrichment of short sequences of TFBSs in the alignments. We evaluate our approach on a set of functionally validated CNEs. Our findings include a regulatory signature composed of co-occurring Pbx-Hox and Meis binding motifs associated with hindbrain enhancer activity.