Abstract 3717: A digital patient & simulated clinical trial of cytarabine treatment

Abstract Historically, computational models have primarily relied on single-omic datasets, particularly transcriptomic data, to predict cancer sensitivity to drugs. These models often identified single gene alterations as biomarkers of drug response. While informative, the predictive potential of these biomarker panels remains limited. Moreover, these same panels have failed to predict potential toxicities associated with anti-cancer drugs. This underscores the urgent need to construct more robust biomarker panels capable of not only predicting patient response but also anticipating potential toxic effects. To address the complexity of cancer biology, we leveraged Champions’ extensively annotated bank of over 1500 Patient-Derived Xenograft Models (PDX). to create the deepest multi-omic dataset, incorporating proteomics, transcriptomics, genomics, and phenotypic data. Simultaneously, we assembled a comprehensive multi-omic normal tissue dataset, encompassing samples from major organs. In this work, we present an integrated approach to construct a digital patient capable of predicting both sensitivity and toxicity concerns by leveraging this rich multi-omics dataset. Our strategy involved pharmaco-pheno multi-omics integrations. Specifically, we conducted an ex-vivo cytotoxicity assay to establish sensitivity and resistance status to cytarabine using a panel of primary Acute Myeloid Leukemia (AML) samples. These samples had been previously characterized through proteomics, transcriptomics, and immunophenotypic analyses. The resulting cytotoxicity data, along with all omics and phenotypic information, were input into an N-based machine learning platform that leveraged partial least square Differential analysis. The output yielded a panel of multi-omics biomarkers predictive of cytarabine response. When prospectively applied to a new set of AML samples, our biomarker panel successfully predicted sensitivity in all but one sample. Intriguingly, the model achieved a 57% objective response rate. Furthermore, when we extended the model to normal tissue data, data suggested potential toxicity concerns for bone marrow, lung, heart, and peripheral nervous system. These findings correlate with observations in the clinic, therefore highlighting the potential use of digital patient strategies to simulate clinical trials. Citation Format: Gilad Silberberg, Stefano Cairo, Mara Gilardi, Brandon Walling, Marianna Zipeto, Michael Ritchie. A digital patient & simulated clinical trial of cytarabine treatment [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 3717.