Abstract
Dying tumor cells shed DNA fragments into the circulation that are known as circulating tumor DNA (ctDNA). Liquid biopsy tests aim to detect cancer using known markers, including genetic alterations and epigenetic profiles of ctDNA. Despite various advantages, the major limitation remains the low fraction of tumor-originating DNA fragments in a high background of normal blood-cell originating fragments in the cell-free DNA (cfDNA) pool in plasma. Deep targeted sequencing of cfDNA allows for enrichment of fragments in known cancer marker-associated regions of the genome, thus increasing the chances of detecting the low fraction variant harboring fragments. Most targeted sequencing panels are designed to include known recurrent mutations or methylation markers of cancer. Here, we propose the integration of cancer-specific chromatin accessibility states into panel designs for liquid biopsy. Using machine learning approaches, we first identify accessible and inaccessible chromatin regions specific to each major human cancer type. We then introduce a score that quantifies local chromatin accessibility in tumor relative to blood cells and show that this metric can be useful for prioritizing marker regions with higher chances of being detected in cfDNA for inclusion in future panel designs.