Fig. 4From: An interpretable bimodal neural network characterizes the sequence and preexisting chromatin predictors of induced transcription factor bindingAscl1 cognate motif multiplicity and flanking nucleotides predict induced Ascl1 binding. a Motifs enriched at saliency hills (regions of the input sequences that are used by the sequence sub-network to make predictions). b Frequencies of CAGSTG k-mers at all Ascl1 binding sites. c BichromSEQ scores increase with increasing motif multiplicity at Ascl1 binding windows. d Embedding CAGSTG motifs in simulated sequences confirms that the BichromSEQ sub-network uses the number of motif occurrences as a score-driving predictor. e BichromSEQ scores assigned to CAGSTG k-mers vary widely based on nucleotides flanking the cognate k-merBack to article page