Explainable AI for multi-omics in precision medicine: a systematic review

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Background: The convergence of multi-omics technologies and artificial intelligence (AI) has opened new frontiers in precision medicine; however, the complexity and opacity of advanced AI models remain a major barrier to clinical adoption. This systematic review aims to critically evaluate explainable AI (XAI) strategies for multi-omics integration and their role in bridging the translational gap between computational innovation and clinical utility. Methods: A systematic literature search was conducted across PubMed/MEDLINE, Scopus, and Web of Science databases for studies published between 2020 and 2025, following PRISMA 2020 guidelines. Studies addressing multi-omics integration using explainable or interpretable AI methods in precision medicine were included. Data extraction and narrative synthesis were performed due to methodological heterogeneity. Results: A total of 116 studies were included in the final analysis. Computational approaches ranged from classical machine learning and deep learning to graph-based and transformer architectures. XAI techniques, including SHAP (SHapley Additive exPlanations), attention mechanisms, and saliency maps, enabled interpretable predictions across gene, pathway, and network levels. Applications were most prominent in cancer subtyping, biomarker discovery, drug response prediction, and prognosis modeling. Despite promising performance, key challenges persist, including data heterogeneity, high dimensionality, batch effects, overfitting, limited reproducibility, and insufficient clinical validation. Discussion: XAI enhances transparency, trust, and biological interpretability in multi-omics models, facilitating their integration into clinical workflows. Emerging directions such as federated learning, causal AI, foundation models, digital twins, and human-in-the-loop systems offer potential solutions to current limitations. Standardized evaluation frameworks and robust clinical validation are essential to advance real-world implementation. This review provides a comprehensive roadmap for developing reliable and clinically actionable XAI-driven multi-omics systems in precision medicine.