Methods for identifying regulatory grammars

Abstract

Recent advancements in sequencing technology have made it possible to study the mechanisms of gene regulation, such as protein-DNA binding, at greater resolution and on a greater scale than was previously possible. We present an expectation-maximization learning algorithm that identifies enriched spatial relationships between motifs in sets of DNA sequences. For example, the method will identify spatially constrained motifs colocated in the same regulatory region. We apply our method to biological sequence data and recover previously known prokaryotic promoter spacing constraints demonstrating that joint learning of motifs and spacing constraints is superior to other methods for this task.