35 Novel statistical methods to unlock the clinical potential of liquid biopsy sampling

Abstract

OBJECTIVES/GOALS: Decoding the origins of cell-free DNA (cfDNA) released from dying cells in a liquid biopsy sample (e.g. blood) can provide insight into the dynamic, organism-wide changes reflective of health and disease state. Making cfDNA an ideal target for genomic monitoring of disease-related changes. METHODS/STUDY POPULATION: Methylome-wide sequencing (WGBS) data present unique statistical challenges. To this end, we developed a novel statistical method using an Expectation-Maximization algorithm to decode the cellular origins of cfDNA fragments in liquid biopsies. Our flexible, probabilistic method leverages the co-regulation of neighboring CpG sites on the individual sequencing read to facilitate tissue of origin analysis, as opposed to prior methods that focus on the methylation rate of a single CpG site. We assess the performance of our model in various simulated settings and apply our model to an important clinical example in which we are able to detect early off-target tissue damage from radiation therapy via minimally invasive blood draws. RESULTS/ANTICIPATED RESULTS: We found our model more effective at capturing the range of biologically plausible methylation patterns on cfDNA read fragments compared to prior models that use single CpG sites. We also show our model is robust to high levels of noise inherent with WGBS data. We demonstrate the accuracy of cell-type proportion estimation on in-silico mixed cfDNA samples from real WGBS data. Finally, we use our model in a clinical application. We detect significant (p < 0.05) increases in cellular contributions from lung and cardiac tissue in breast cancer patients (n=15) undergoing radiation therapy compared to baseline. We also detect novel signals of radiation induced toxicity to the liver in right-sided breast cancer patients (n=8) receiving radiation treatment compared to matched left-sided breast cancer patients (n=7). DISCUSSION/SIGNIFICANCE: Here we address an unmet need in developing novel statistical methodologies that can handle the unique complexities of methylated cfDNA obtained from liquid biopsy samples. We also demonstrate the far-ranging clinical utility of serial liquid biopsy sampling to complement and advance standards of clinical care in oncology and other pathologies.