Integrating multi-omics with machine learning: from genomic variant functional annotation to pharmacogenomics and biomarker discovery

Integrating heterogeneous multi-omics data with machine learning (ML) provides a powerful framework for interpreting genomic variants and advancing pharmacogenomic research in precision medicine. High-throughput technologies have generated vast datasets spanning genomics, transcriptomics, epigenomics, and proteomics, yet integrating these high-dimensional data into clinically meaningful insights remains a major computational challenge. Machine learning and deep learning models, including Graph Neural Networks (GNNs), Transformers, and Variational Autoencoders (VAEs), offer promising solutions. This narrative review provides a state-of-the-art overview of ML-driven multi-omics integration across early, intermediate, and late fusion paradigms. We highlight how these approaches may support functional variant interpretation, including the reclassification of Variants of Uncertain Significance (VUS), and emerging pharmacogenomic applications. Key areas include predictive biomarkers for drug efficacy, therapeutic resistance in precision oncology, and personalized dosing in rare and complex diseases. Finally, we discuss challenges related to data heterogeneity, clinical interpretability, and regulatory approval, as well as the requirements for translating emerging computational approaches into clinically validated pharmacogenomic practice and precision pharmacotherapy.