Abstract 3709: Ezekiel: a simulated digital twin for virtual diagnostic biomarker screening

Background: Extracellular vesicles (EVs) are released from all stages of tumors and are highly abundant in blood, making them an attractive analyte for cancer screening diagnostics. However, background noise from non-tumor cells presents a significant challenge to cancer-specific signal detection. We have developed a novel assay platform for measuring combinations of surface biomarkers co-localized on EVs that are highly specific for tumor-associated EVs and a computational platform to simulate individual patient EV profiles for the purpose of nominating candidate biomarkers for use in this assay. Discovery Pipeline: Because the EV-based assay is a multi-marker test, combinatorial expansion of biomarkers presents a significant challenge to experimental validation. To discover biomarker candidates, we simulate performance using a virtual assay of simulated patients using a novel digital twin platform called Ezekiel. In these simulated patients, tumor gene expression profiles from multi-center cancer genomic consortia (e.g. TCGA) are combined with background gene expression from healthy tissue using tumor size and EV abundance estimates to generate a virtual analyte pool. Initial candidate biomarkers are selected based on a carefully curated set of quantitative and qualitative features, and pairs and triplets of biomarkers are ranked by signal-to-noise ratio in a simulated assay to refine a library of targets for screening. Validation: Biomarkers and biomarker combinations identified by this computational approach are validated experimentally. Individual biomarkers and immunoaffinity capture reagents are first validated using Western blotting and an abbreviated version of the EV-based assay in pooled plasma. Pairs or triplets of biomarkers are then evaluated in the full assay to select top candidates for development. We have confirmed co-localization of these biomarkers using Single Molecule Localization Microscopy (SMLM) and tumor gene co-expression in matched patient tissue and plasma. Results: We present a novel computationally driven discovery workflow for a single representative cancer indication. Starting from a pool of 4,782 biomarkers, we virtually screened 11,431,372 combinations resulting in 435 candidates, down selected to 145 biomarkers following manual review. 24 biomarkers were selected based on availability of high-quality antibodies, from which 13 biomarkers were identified that show activity with our assay platform. Representative orthogonal validation data is presented demonstrating the computationally-nominated biomarker’s success in enabling discrimination of cancer samples from healthy controls. Citation Format: Aaron Chevalier, Maciej Pacula, Gabrielle N. Barcaskey, MacKenzie S. King, Lauren Cuoco, Brittany Grimes, Lukasz Kornilowicz, Katherine S. Yang, Daniel Salem, Anthony Couvillon, Toumy Guettouche, Troy B. Hawkins. Ezekiel: a simulated digital twin for virtual diagnostic biomarker screening [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 3709.