Maximizing the Robustness of Fragmentomics-based Cancer Biomarkers
Advances in circulating tumor DNA (ctDNA) analysis are rapidly expanding the insights that can be obtained about a patient’s cancer from a blood draw. A new frontier of ctDNA analysis is the field of fragmentomics. Fragmentomics uses computational approaches to identify patterns of ctDNA fragmentation that correlate with local epigenetic features. Recently reported fragmentomics methods classify cancer types based on ctDNA fragment size, end locations, end nucleotide sequence, or abundance at regulatory elements. While promising, these methods were developed on small sample sets, often with confounding between case/controls and batches, raising the risk of classifier overfitting. This proposal will systematically benchmark published fragmentomics algorithms, identify the most robust and generalizable features of DNA fragmentation in cancer, and use this information to develop a fragmentomics-based method for inferring estrogen receptor (ER) status from breast cancer ctDNA. The results of this project will provide critical, lacking knowledge about the relative performance and information content of these methods. This knowledge will accelerate the translation of fragmentomic assays into clinically useful biomarkers. Further, the method we propose for assessing ER status form blood could provide a minimally invasive means of guiding therapy selection for patients with breast cancer.