Fig. 1: Process of building TFBS-context model and downstream tasks based on TFBU.

a Illustration of the TFBU. A TFBU is defined as a fragment of DNA sequence that consists of two parts: the core TFBS and the TFBS-context. b Illustration of selecting training samples for the TFBS-context model. Both positive and negative samples were from accessible genome regions with high motif PPM matching scores. The GC content distribution and the histone modification state were balanced between positive and negative samples. Then the core TFBS in these DNA fragments were masked to form TFBS-context datasets for the deep learning model. c Optimizing TFBS-contexts by genetic algorithm with the guidance of the TFBS-context model. The TFBS-contexts deep learning model is TF specific, and was separately trained for each TF. d Illustrations of evaluating the TFBU’s function. The TFBUs were inserted into the plasmid as enhancers to validate their enhancer activity. e Illustrations of tasks based on the concept of TFBU.