Abstract 1139: Massive scale phenotypic screening using generative chemogenomics repurposes safe drugs and discovers new drugs for intercepting early-stage lung adenocarcinoma progression

Abstract Background: Drug screening incorporating molecular profiling techniques is a powerful way to link drug discovery with precision medicine. At AACR 2024, an Australian group showed that profiling patient-derived in vitro models with approved drugs was a viable strategy. We showed that in vitro profiling data coupled with patient profiling found the Aurora kinase inhibitor AMG900 as an interceptor of tumor invasion and progression in early-stage lung adenocarcinoma (esLUAD). These studies show that useful, clinically stable, chemotype-phenotype associations exist in in vitro drug-treated molecular profiles. A generative AI platform, Archetype™, built to harness these insights, shows how to scale beyond existing data and the major physical and cost barriers involved in generating data. Methods: The Archetype™ platform comprises two linked computer models. The first is a generative AI that generates synthetic gene expression changes for new chemicals at different concentrations. The second is a genomic model, built from real patients’ esLUAD tumor samples, and learns an expression-based progression score that predicts survival outcomes. The models work together by generating a synthetic expression profile for each test compound which is used to compute a concentration-dependent progression score. We virtually screened billions of unique small molecules, including 110 million molecules from PubChem, to find significant pharmacodynamic relationships between the structure and the progression score, implying an underlying mechanism. Molecules that gave either higher or lower area under the concentration-progression curve than AMG-900 were selected for in vitro proliferation, invasion and migration testing in A549 cells and further GEMM testing. Results: Previous insights about AMG900 were recovered by the system even though AMG900 was not part of the training data. 980 PubChem compounds with AUC lower than AMG900 were found. 5 drugs with existing mouse PK data were chosen for validation from the top 20. 2 are in clinical development. In vitro proliferation, migration and invasion assays confirmed the in silico results. Compounds with no previously known application to lung cancer, no structural similarity to any of the previous 5 and lower AUC than AMG900 were identified. In vitro testing confirmed that the AI correctly identified novel A549 invasion inhibitors. GEMM model testing of the best repurposed inhibitor and mouse toxicity studies of the novel invasion inhibitors are ongoing. Conclusions: AI-driven phenotypic screening can bridge the gap between billions of drug-like molecules and patient-derived prognostic clinical signatures. By effectively removing many physical and cost constraints, we identified new active molecules through a comprehensive in silico search of the medicinal chemistry space. Citation Format: Abhilasha Sinha, John Hawley, Nanzeeba Faraby, Tanya Cashorali, Thomas A. Neyarapally, William Oh, Paul D. McDonagh, Charles A. Powell. Massive scale phenotypic screening using generative chemogenomics repurposes safe drugs and discovers new drugs for intercepting early-stage lung adenocarcinoma progression [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 1139.