Abstract 6365: A novel methylation-based classifier to identify cancer signal of origin using blood based testing

Abstract Background: Accurate prediction of the cancer signal of origin (CSO) is crucial for blood-based cancer detection in order to guide subsequent diagnostic workups and select the appropriate therapeutic regimes. Here, we present feasibility data from a novel, methylation-based classifier which is able to differentiate the histological origin of 12 cancer types from a single blood-based screening assay. Method: We developed an algorithm on Guardant Infinity®, a next-generation sequencing platform evaluating cell free DNA from plasma samples, to predict the CSO of 12 cancer types (bladder, breast, colorectal, esophageal, kidney, liver, lung, ovarian, pancreatic, prostate, stomach, and uterine/endometrial). This algorithm uses two steps to predict CSO based on the DNA methylation signature of plasma cell-free molecules from >2000 cancer-specific differentially methylated genomic regions. First, a regression model (multi-cancer classifier) was trained using 701 cancer samples across 17 distinct cancer types and 10, 413 non-cancer samples to identify cancer samples with sufficient tumor-derived methylation signal. Next, a methylation-based CSO classifier was used to determine the CSO of samples called positive by the multi-cancer classifier. The CSO classifier was trained on 13, 856 cancer samples from 12 supported cancer types and 4, 142 non-cancer samples, and assessed on 1, 464 advanced cancer samples that were called cancer positive by the cancer detection classifier in the hold out set. Results: Among the 1464 detected cancer samples, 88.2% (1291 / 1464) were correctly classified by the CSO classifier with the top prediction, and 93.6% (1370 / 1464) were correctly classified with the top two predictions. To estimate the real-world performance of this algorithm, the accuracy of the CSO classifier for different cancer types were weighted by the cancer incidence rate in the US population and then aggregated. These adjusted accuracies of the top CSO prediction and top 2 CSO predictions are 88.5% and 93.0%, respectively. Conclusion: Herein, we demonstrate the capability of a novel, methylation-based algorithm to predict CSO with high sensitivity and accuracy from a plasma-based sequencing platform. Further clinical validation in early cancer detection and disease monitoring is needed. Citation Format: Elmira Forouzmand, Yupeng He, Rachel Gittelman, Alan Selewa, Meromit Singer, Jill Tsai, Theresa Hoang, Jack Tung, Victoria M. Raymond, Amirali Talasaz, Darya Chudova. A novel methylation-based classifier to identify cancer signal of origin using blood based testing [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 6365.