Comparative evaluation of DNase-seq footprint identification strategies

DNase I is an enzyme preferentially cleaving DNA in highly accessible regions. Recently, Next-Generation Sequencing has been applied to DNase I assays (DNase-seq) to obtain genome-wide maps of these accessible chromatin regions. With high-depth sequencing, DNase I cleavage sites can be identified with base-pair resolution, revealing the presence of protected regions (“footprints”), corresponding to bound molecules on the DNA. Integrating footprint positions close to transcription start sites with motif analysis can reveal the presence of regulatory interactions between specific transcription factors (TFs) and genes. However, this inference heavily relies on the accuracy of the footprint call and on the sequencing depth of the DNase-seq experiment. Using ENCODE data, we comprehensively evaluate the performances of two recent footprint callers (Wellington and DNaseR) and one metric (the Footprint Occupancy Score, or FOS), and assess the consequences of different footprint calls on the reconstruction of TF-TF regulatory networks. We rate Wellington as the method of choice among those tested: not only its predictions are the best in terms of accuracy, but also the properties of the inferred networks are robust against sequencing depth.