Motivation: Transcription factor (TF) ChIP-seq datasets have particular characteristics that provide unique challenges and opportunities for motif discovery. Most existing motif discovery algorithms do not scale well to such large datasets, or fail to report many motifs associated with cofactors of the ChIP-ed TF.
Results: We present DREME, a motif discovery algorithm specifically designed to find the short, core DNA-binding motifs of eukaryotic TFs, and optimized to analyze very large ChIP-seq datasets in minutes. Using DREME, we discover the binding motifs of the the ChIP-ed TF and many cofactors in mouse ES cell (mESC), mouse erythrocyte and human cell line ChIP-seq datasets. For example, in mESC ChIP-seq data for the TF Esrrb, we discover the binding motifs for eight cofactor TFs important in the maintenance of pluripotency. Several other commonly used algorithms find at most two cofactor motifs in this same dataset. DREME can also perform discriminative motif discovery, and we use this feature to provide evidence that Sox2 and Oct4 do not bind in mES cells as an obligate heterodimer. DREME is much faster than many commonly used algorithms, scales linearly in dataset size, finds multiple, non-redundant motifs and reports a reliable measure of statistical significance for each motif found. DREME is available as part of the MEME Suite of motif-based sequence analysis tools (http://meme.nbcr.net).