Abstract 3681: A tumor-naive method for estimating tumor fraction using 5-hydroxymethylation cytosine in cell-free DNA

Abstract Accurate measurement of the tumor fraction (TF) in circulating cell-free DNA (cfDNA) allows estimation of tumor burden non-invasively, potentially useful in guiding treatment decisions, monitoring treatment response, and detecting minimal residual disease. 5-hydroxymethylation of cytosine (5hmC) marks activation of cancer drivers and downstream gene targets, and has been shown to positively correlate with gene expression and gene regulation. Here we trained a machine learning (ML) model to predict TF in a tumor-naive way from 5hmC-enriched sequencing data together with low-pass whole genome sequencing data. In cross-validation the ML model showed excellent performance, showing high correlation (R > 0.8) and low composite absolute error (CAE < 5%) compared to copy-number variation (CNV) based estimates. The high correlation was confirmed in an independent test set which also showed R > 0.8, with only a modest increase in CAE (<10%) using targeted point mutation-based TF estimates as an external reference comparator. Using in silico cancer mixtures that created contrived cfDNA samples targeting concentrations below 10%, our 5hmC-based TF prediction method achieved detection sensitivity of approximately 90% with a specificity of 97%. Finally, we observed that 5hmC-based predicted TF was significantly associated with overall survival (HR = 1.04, p < 0.001) in an independent prostate patient cfDNA cohort when controlling for other covariates. In summary, we find that the 5hmC-based TF estimate is a reliable metric of tumor burden irrespective of tumor mutational status. The 5hmC-based TF estimate may be used either alone or in conjunction with mutation and CNV-based estimates to provide more accurate measures of survival risk and therapy response during treatment. Citation Format: Ceyda Coruh, Onur Sakarya, Yuhong Ning, Yuan Xue, Michael Cipriano, Kyle Hazen, Alexander W. Wyatt, David Quigley, Felix Feng, Martin Sjostrom, Gulfem Guler, Wayne Volkmuth, Samuel Levy. A tumor-naive method for estimating tumor fraction using 5-hydroxymethylation cytosine in cell-free DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3681.