Supplementary_table.xlsx

Extrachromosomal circular DNA (ecDNA) is a predominant driver of oncogenic copy-number amplifications across diverse tumor types, however, there is currently no clinically validated non-invasive method for its detection. In this study, we present a cfDNA fragmentation profiling approach that leverages the increased abundance of short DNA fragments within amplified genomic regions as a biomarker for gene amplification events. We utilized metrics, such as fragment size entropy, motif diversity score, and weighted fragment size as features to develop gene-specific machine learning classifiers and validated their performance using somatic panel sequencing data from liquid-biopsy samples of cancer patients. This approach circumvents dependence on conventional read-depth analyses while achieving robust performance. Collectively, our results establish cfDNA fragmentation profiling as a powerful, reliable, and minimally invasive method for detecting amplifications.