Spectral Sequence Motif Discovery

Sequence discovery tools play a central role in several fields of computational bi-ology. In the framework of Transcription Factor binding studies, motif finding algorithms ofincreasingly high performances are required to process the big datasets produced by new high-throughput sequencing technologies. Most existing algorithms are computationally demandingand often cannot support the large size of new experimental data. We present a new motif dis-covery algorithm that is built on a recent machine learning technique, referred to as Method ofMoments. Based on spectral decompositions, this method is robust under model misspecificationand not prone to locally optimal solutions. We obtain an algorithm that is extremely fast anddesigned for the analysis of big sequencing data. In few minutes, we can analyse datasets ofmore than hundred thousand sequences and produce motif profiles that match those computed byvarious state-of-the-art algorithms.