Abstract 2386: Preclinical model atlas to enable clinical translatability for precision oncology

Abstract Preclinical models, such as 2D cultures, 3D organoids, and patient-derived xenograft (PDX) models, play a crucial role in drug development pipelines. However, their respective suitability for various research questions remains unclear. For example, RSL3-induced ferroptosis in 2D models is abolished in 3D models and in vivo. Similarly, some compounds targeting KRAS pathways are potent in 2D but not effective in vivo. Comprehensive understanding of the strengths and limitations of these models is paramount for drawing meaningful scientific conclusions. Identifying the models best suited to represent patient populations and forecast treatment response with precision is essential in enabling reverse translational research for hypothesis testing and target validation. Building upon our previous work which focused on 133 3D culture models from DEPMAP, we have expanded our efforts to encompass a larger, more comprehensive array of preclinical models. This includes DEPMAP models (207 3D and 2099 2D), 48 paired 2D-3D organoids from AbbVie, as well as around 800 internally generated and 4582 externally acquired PDX models, giving a grand total of roughly 7800 preclinical models with genomics and transcriptomics data. We employed CellTindeRX, a model-free and likelihood-based tool, to enable precision oncology by forecasting effective treatments for individuals. Briefly, we identify the tumor model most representative of a given human sample by their transcriptomics similarity, then we use the matching tumor model to further link the human sample to corresponding drug sensitivity measurements. Our approach preserves the transcriptomic signals from immune and stroma components of patient tumors, so that model comparisons consider both intrinsic and extrinsic aspects of the diseases. In general, we observe agreement for most human cancer types in TCGA between pre-clinical models and their corresponding patient diseases; however, some indications, such as stomach adenocarcinoma (STAD), show poor agreement, underscoring the biological gaps between preclinical models and patient samples. We further align preclinical models to patient samples from real world data and clinical trials to demonstrate a synthetic drug screen framework where we forecast patients’ treatment response using PRISM drug sensitivity profiles from matched cell lines. Using this screening platform, TOP1i is identified as the top efficacious drug class for small cell lung cancer (SCLC). This observation is aligned with clinical treatment guidelines for this disease.These findings offer valuable guidance for selecting preclinical models that faithfully represent the patient population for reverse translational efforts, such as compound efficacy testing or MOA/target validation, thereby enhancing the accuracy and relevance of our drug development efforts. Citation Format: Anlu Chen, Thanh Bui, Weilong Zhao, Shaobo Qin, Vivian Prindle, Kimberly Sher, Adam Richardson, Joshua Plotnik, Aloma D'Souza, Marybeth Pysz, Josue A. Samayoa, Sarah Kongpachith, Xi Zhao. Preclinical model atlas to enable clinical translatability for precision oncology [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 2386.