A novel deep learning-based framework reveals a continuum of chromatinsensitivities across transcription factors

Abstract

The genome-wide binding of many transcription factors (TFs) depends not only on the presence of their recognition motifs, but also on the surrounding chromatin context. This raises the question of how binding specificity is achieved, which is essential for the proper activation of regulatory regions during developmental and differentiation processes. Two competing hypotheses have been proposed: either opening is a hierarchical process, whereby chromatin-insensitive TFs bind initially and make the DNA accessible for other TFs; or multiple chromatin-sensitive TFs bind cooperatively, enabling them to overcome the chromatin barrier. Here, using complementary genomic- and deep-learning-based techniques, we provide a quantitative framework to assess chromatin sensitivity of individual TFs, and hone in on the definitions and implications of such sensitivity. Our findings demonstrate that the degree of chromatin sensitivity is a continuum across TFs, showing that regulatory regions are activated neither by a purely hierarchical nor purely cooperative mechanism, but by a mixture of both mechanisms. The presented framework allows for a locus-specific as well as global, quantitative assessment of chromatin sensitivity and should help in understanding the binding behaviour of individual TFs as well as their potential contribution to gene regulation.