Abstract LB254: Sharpening the signal: Enhancing liquid biopsy specificity through intra-individual methylation analysis

Abstract Early cancer detection via liquid biopsy is challenged by the need for ultra-high specificity to minimize false positives in a low-prevalence population (∼1.4%). DNA methylation is a promising biomarker for detecting cancer across stages and tissue types, but physiological variability—due to factors like age and inflammation—can reduce biomarker specificity and impact clinical positive predictive value (PPV). Circulating tumor-derived DNA (ctDNA) exists at low frequencies (<0.1%), and it is difficult to distinguish cancer-related methylation patterns from background somatic DNA. Harbinger Health has developed Harbinger Hx, a targeted DNA methylation assay, leveraging machine learning (ML) for early cancer detection. As the majority of cell free DNA (cfDNA) originates from white blood cells (WBCs), we hypothesized that a paired intra-individual analysis (IIA)—comparing cfDNA to matched WBC-derived genomic DNA (gDNA)—could identify ctDNA signal from background somatic noise. IIA was performed with Harbinger Hx on three patient cohorts of increasing size (N36, N335 and N1280) selected from the CORE-HH clinical study (NCT05435066). The N1280 patient cohort contains 943 patients with no reported cancer diagnosis, and 337 with a confirmed cancer diagnosis from 9 cancer types. Briefly, paired cfDNA/gDNA was extracted for each patient. The DNA was then bisulfite treated, converted into indexed libraries, enriched for cancer specific regions of interest and sequenced to ≥100X unique median target coverage. A cfDNA-only ML cancer classifier was built (MLX), as well as a patient specific intra-individual classifier (IIX), by subtracting the fraction of tumor content (TC) signal observed in gDNA from the fraction of TC signal detected in cfDNA. A 10-fold cross validation (CV) approach was used to establish specificity thresholds for both MLX and IIX. We also assessed read-level IIA-filtering to identify and remove reads with shared phased methylation patterns in both patient samples with the goal of enriching for ctDNA reads. We observed a high correlation between cfDNA TC and the number of fragments remaining after IIA-filtering, suggesting that IIA can isolate TC signal. At a target specificity of 98.5%, we observed 68.2% and 66.4% sensitivity for MLX and IIX respectively. When the two tests were combined (i.e. only IIX-pass samples were evaluated by MLX), we observed 63.7% sensitivity at an achieved specificity of 99.5%. The combined test reduces the false positive rate without adjusting the specificity threshold for each individual test and increases confidence in performance generalization. We estimate that 2-3X more samples would be required to report 99.5% specificity for an individual test at the same level of confidence. Therefore, IIA represents a novel approach to reduce variability due to methylation noise, offering a strategy to achieve higher overall specificities and PPV. Citation Format: Miguel Williams, Sarah Falotico, Caitlin Gilley, Mayur Gurnani, Emily Neaga, Kristy Schlueter-Kuck, Shiva Farashahi, Jackson A. Killian, Jocelyn Charlton, Kieran Chacko, Tony Shuber. Sharpening the signal: Enhancing liquid biopsy specificity through intra-individual methylation analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_2):Abstract nr LB254.