Deep Learning Analysis of High‐Throughput Transcription Factor– DNA Binding Affinity Data: A Quantitative Comparison With Pairwise‐Additive Models

ABSTRACT Many transcription factors (TFs) regulate gene expression by binding to specific DNA sequences. Widely used models of TF–DNA binding, such as position weight matrices (PWMs) and position‐specific affinity matrices (PSAMs), assume that binding free energy can be represented as the sum of independent nucleotide contributions. However, there is ample evidence that non‐additive effects significantly influence TF binding. Here, we utilize data from a high‐throughput in vitro assay ( ivt FOODIE) to generate genome‐scale TF–DNA dissociation constants ( K d ), and use these data to systematically evaluate sequence‐to‐affinity models of increasing complexity. We demonstrate that pairwise additive models exhibit systematic deviations from the measured affinity landscapes. Models incorporating adjacent dinucleotide interactions or employing deep learning architectures achieve markedly improved agreement with experimental K d values. The degree of non‐pairwise‐additive sequence dependence varied substantially among TFs and appeared to vary with DNA‐binding domain type. In silico mutation screening predicts widespread, TF‐specific long‐range inter‐position dependencies, highlighting the role of energetic coupling across distant positions in target recognition. These results provide a quantitative understanding of non‐pairwise‐additive energetic effects across diverse TFs.