From inherited susceptibility to tumor state: toward a stage-aware architecture for genomic risk stratification in...

Prostate cancer risk is currently represented by two largely separate genomic layers. Germline polygenic scores estimate inherited susceptibility before diagnosis, whereas tumor genomic and transcriptomic classifiers estimate biological aggressiveness and treatment response after tissue acquisition. Artificial intelligence can connect these layers, but simply concatenating high-dimensional features is unlikely to produce transportable clinical tools. This Mini Review examines how biologically informed machine learning, single-cell and spatial atlases, whole-omics response models, and validated tissue classifiers are reshaping prostate cancer risk stratification. We argue that the most useful architecture is a stage-aware system: ancestry-calibrated germline risk first modifies screening intensity; tissue or liquid-biopsy models then resolve tumor state; and longitudinal molecular measurements update risk under treatment. Evidence is strongest for polygenic enrichment of screening populations and for locked transcriptomic classifiers tested in randomized-trial cohorts. By contrast, many multi-omics models remain retrospective, dataset-dependent, and weakly calibrated across ancestry, disease stage, and specimen type. Progress requires explicit clinical decision points, external and prospective validation, uncertainty reporting, ancestry-aware calibration, and comparisons against simpler clinicopathological models. A unified framework should therefore treat artificial intelligence not as a single prognostic score, but as an auditable sequence of genomic updates that links inherited susceptibility to evolving tumor biology.