Deep Learning Paradigm for Precision Lung Cancer Therapy with AI-Driven Genotype-Phenotype Mining and Patient-Derived Organoid Validation

The rapid and accurate prediction of anticancer drug responses is critical for enhancing treatment efficacy and improving clinical outcomes in cancer patients. However, the practical implementation of current predictive models is hampered by dual limitations: machine learning approaches reliant on cell line data often exhibit suboptimal accuracy, while patient-derived organoids (PDOs) platform typically lack the rapid turnaround required for timely clinical decision-making. Here, we present a deep learning framework to predict drug response in lung cancer patients by integrating patient genomic sequencing data with compound structural information, trained against phenotypic drug sensitivity profiles from lung cancer PDOs, to predict drug responses in lung cancer patients. Our model enables individualized prediction of antitumor activity across diverse chemical structures, demonstrating capabilities for predicting efficacy of both approved drugs and novel compounds, as well as facilitating drug repurposing. The framework achieved 81.6% prediction accuracy, which was experimentally validated using patient-derived organoid models. More importantly, evaluation in a clinical cohort of lung cancer patients confirmed the model's ability to accurately reflect actual treatment responses. This study represents the first successful integration of genotype, drug structure, and organoid phenotype within a unified computational framework, significantly enhancing the accuracy and biological interpretability of drug response predictions while providing a clinically applicable tool for precision oncology in lung cancer.