Machine learning enhances biomarker discovery: From multi- omics to functional genomics.

Importance: Biomarkers are critical for precision medicine, supporting disease diagnosis, prognosis, personalized treatments, and monitoring. Traditional biomarker discovery methods, which often focus on single genes or proteins, face several challenges, including limited reproducibility, a limited ability to integrate multiple data streams, high false-positive rates, and inadequate predictive accuracy. Machine learning and deep learning methods, and large language models, paired with advancements in omics technologies, address these limitations by analyzing large, complex multi-omics datasets to identify more reliable and clinically useful biomarkers. Observations: Machine learning and deep learning have proven effective in biomarker discovery by integrating diverse and high-volume data types, such as genomics, transcriptomics, proteomics, metabolomics, imaging, and clinical records. These approaches successfully identify diagnostic, prognostic, and predictive biomarkers across fields, such as oncology, infectious diseases, neurological disorders, and autoimmune diseases. Newer methodological developments include approaches to identify functional biomarkers, notably biosynthetic gene clusters, crucial for discovering antibiotics and anticancer drugs. Key artificial intelligence (AI) techniques include neural networks, transformers, large language models, and feature selection methods, which are finding more and more application to omics data and in clinical settings. However, challenges remain regarding data quality, biological complexity, model interpretability, validation, and generalization. Regulatory and ethical considerations also impact clinical adoption, emphasizing the importance of validated, trustworthy, and explainable AI methods. Conclusions and Relevance: Machine learning, deep learning, and AI agent-based approaches significantly enhance biomarker discovery, providing valuable biological insights and advancing precision medicine. Future research should focus on directly linking genomic data to functional outcomes, particularly with biosynthetic gene clusters and non- coding RNAs. Rigorous validation, model interpretability, and regulatory compliance are essential for clinical implementation. These advancements promise to improve personalized treatment strategies and patient outcomes.