AI-driven multi-omics biomarkers in precision breast cancer therapeutics: from discovery to clinical translation

Breast cancer is biologically heterogeneous, yet conventional clinical decisions rely on static, single-marker proxies (ER, PR, HER2, and Ki-67) that capture only a narrow slice of tumor biology. This mini-review argues that AI-driven multi-omics biomarkers offer a valuable shift toward systems-level, model-derived probabilities integrating genomic, transcriptomic, proteomic, metabolomic, and imaging information to explain and predict disease behavior more faithfully. We synthesize recent advances across three domains: (i) computational strategies for multi-omics integration, from classical penalized regression and ensemble learning to deep multimodal architectures and foundation-model embeddings; (ii) therapeutically relevant biomarker classes across tissue, liquid biopsy, and imaging modalities; and (iii) emerging pathways for embedding these biomarkers into clinical decision support. We highlight validation, interpretability, and robust handling of missing modalities as prerequisites for clinical trust, and outline how biomarker-driven trials, standardized reporting (e.g., TRIPOD + AI), and privacy-preserving data consortia can move AI-driven multi-omics biomarkers toward clinical and regulatory maturity.