ATAC-seq Guided Interpretable Machine Learning Reveals Cancer-Specific Chromatin Features in Cell-free DNA

Cell-free DNAs (cfDNAs) are DNA fragments found in blood. In healthy individuals, cfDNAs are primarily derived from immune cells, while in cancer patients, a significant fraction of cfDNAs originates from cancerous cells. These cancer-derived cfDNAs contain specific mutations, making cfDNA analysis a promising diagnostic biomarker. Recent studies have revealed that epigenetic information, such as DNA methylation and nucleosome positioning, is retained in cfDNAs, enhancing the accuracy of cell-of-origin predictions. This study aims to characterize the chromatin architecture preserved in cfDNAs by looking at nucleosomal DNA enrichment. Nucleosome fragments from both breast and pancreatic cancer patients are significantly enriched in open chromatin regions. A differential enrichment was observed between healthy donors and cancer patients at cell type-specific ATAC-seq peaks. Leveraging this pattern of open chromatin enrichment, we enhanced the prediction accuracy for identifying breast cancer-derived cfDNA through machine learning. Our analysis pipeline provides an interpretable machine learning platform that effectively detects cancer-specific nucleosome enrichment in cfDNAs.