WTKO-CNN Model Decodes Chromatin Accessibility Differences

A new deep learning model, WTKO-CNN, distinguishes DNA sequences from wild-type and knockout conditions by analyzing ATAC-seq peaks. The convolutional neural network with attention mechanism achieves high predictive performance. Saliency maps identify influential nucleotide positions, enabling de novo motif discovery through k-mer extraction and clustering. Sequence logos and consensus motifs from CNN filters reveal biologically meaningful patterns, validated against known transcription factor binding sites using MEME, TOMTOM, and HOMER. The study identifies motifs associated with transcriptional control, offering insights into how chromatin regulators alter regulatory DNA element accessibility.