From spatial maps to treatment decisions: a roadmap for integrating explainable AI with multi-omics to guide precision immunotherapy.

Immunotherapy has fundamentally transformed oncology, yet durable clinical responses remain confined to a minority of patients, a therapeutic impasse that arises from the spatial complexity of the tumor microenvironment. Single-cell and spatial multi-omics now resolve this architecture at unprecedented resolution, while deep learning extracts prognostic signals from histopathology images. Yet these advances have not closed the translational gap, largely because black-box models, fragmented validation, and a lack of prospective interventional evidence prevent clinical adoption. Here, we argue that the strategic convergence of spatial multi-omics, explainable artificial intelligence (XAI), and mechanism-guided clinical trial design will provide the definitive translational bridge between tissue architecture and therapeutic decision-making. We outline a roadmap in which interpretable spatial biomarkers, derived from concept-based XAI and counterfactual reasoning, are locked as assays, prospectively validated in biomarker-stratified trials, and evaluated under real-world conditions through federated learning. We highlight the necessity of community-wide benchmarking, mandatory sharing of code and spatial data, and the deliberate reporting of negative results to discipline the field. To operationalize this integration, we introduce the Spatial Immune Engagement Index (SIEI), a distance-weighted metric that quantifies the proximity of CD8