AI-driven genotype-phenotype modeling: a framework integrating multi-modal single-cell genomics and reverse vaccinology for de novo design of multi-epitope cancer vaccines.

Cancer vaccines have emerged as a promising strategy for personalized cancer immunotherapy; however, their development has traditionally relied on bulk sequencing approaches that average molecular information across millions of cells, thereby obscuring the extensive intratumoral heterogeneity that drives disease progression, therapeutic resistance, and immune escape. Recent advances in multi-modal single-cell genomics have transformed the ability to characterize tumors at unprecedented resolution, enabling the identification of distinct cellular populations, clonal evolutionary trajectories, and complex tumor-immune interactions. In parallel, artificial intelligence (AI) has rapidly expanded the capabilities of reverse vaccinology by facilitating large-scale analysis of genomic and immunological datasets for neoantigen discovery and vaccine design. This review aims to present a unique conceptual framework for future personalized cancer immunotherapies, rather than simply integrating the already established approaches. The framework is built on two levels: (1) filtering of false-positive targets using multi-modal single cell data and removing antigen loss clones; and (2) feeding the resulting rigorously filtered data into advanced structural and generative AI models to inform de novo design of multi-epitope vaccines. Particular emphasis is placed on the application of deep learning, graph neural networks, transformer architectures, and generative AI models for data preprocessing, clonal evolution analysis, immune microenvironment characterization, neoantigen prioritization, and peptide-major histocompatibility complex (MHC) interaction prediction. Furthermore, we discuss the development of integrated computational pipelines capable of translating high-resolution multi-modal single-cell data into personalized multi-epitope cancer vaccines. Finally, we highlight the major translational challenges, including model interpretability, tumor plasticity, manufacturing constraints, and clinical implementation. By integrating multi-modal single-cell genomics with advanced AI methodologies, reverse vaccinology is poised to accelerate the development of highly targeted, adaptive, and durable cancer vaccines, offering a promising roadmap for the future of personalized cancer immunotherapy.