Transcription factor regulatory networks underlie major features of cellular identity and complex function such as pluripotency, development and differentiation. Li and colleagues develop a graph neural network to predict transcription factor regulatory networks based on single-cell ATAC-seq data.