Foundation AI Models for Predictive Immuno-Oncology: Personalizing Immunotherapy Through Multimodal Learning

Abstract

Immunotherapy has revolutionized modern oncology by enabling durable clinical responses across multiple malignancies through activation of the host immune system. However, only a subset of patients experiences sustained therapeutic benefit because of the remarkable complexity and heterogeneity of tumor–immune interactions. Accurate prediction of immunotherapy response remains one of the greatest challenges in precision oncology. Recent advances in foundation artificial intelligence (AI) models have introduced a new paradigm for predictive immuno-oncology by enabling large-scale multimodal representation learning across radiological imaging, digital pathology, genomics, transcriptomics, proteomics, immunomics, spatial biology, laboratory biomarkers, electronic health records, and longitudinal clinical data. Unlike conventional task-specific machine learning algorithms, foundation models learn generalized biomedical representations through self-supervised pretraining that can be adapted across numerous downstream oncology applications. Transformer architectures, multimodal large language models, graph neural networks, retrieval-augmented generation, generative AI, and multimodal foundation models have demonstrated remarkable potential for predicting immunotherapy response, characterizing the tumor microenvironment, identifying novel biomarkers, optimizing treatment selection, and supporting adaptive clinical decision-making. These intelligent systems further facilitate integration of radiomics, computational pathology, spatial multi-omics, digital twins, and real-world clinical evidence into comprehensive precision immunotherapy workflows. Despite substantial progress, important challenges remain regarding data harmonization, explainability, computational scalability, interoperability, cybersecurity, regulatory validation, and equitable clinical implementation. This review provides a comprehensive overview of foundation AI models in predictive immunooncology, highlighting computational principles, current clinical applications, implementation challenges, and future perspectives for personalized immunotherapy.[1]