Machine learning for metabolomics research in drug discovery

In a pharmaceutical context, metabolomics is an underexplored area of research. Nevertheless, its utility in clinical pathology, biomarker discovery, metabolic subtyping, and prognosis has transformed medicine. As this young domain evolves, its promise as an approach to drug discovery becomes more evident. It has established links between human phenotypes and quantitative biochemical parameters, enabling the construction of genome-scale metabolic networks. While the human metabolome is too vast for manual analysis, machine learning (ML) algorithms can efficiently recognize latent patterns in complex, large sets of metabolic data. ML-driven studies of the human metabolome and its constituents can inform efforts to reduce the quantity of resources spent at critical stages of the pipeline by facilitating target identification, mechanism of action elucidation, lead discovery, off-target effect evaluation, and in vivo response prediction. Metabolism-informed ML models generate insights that significantly advance efforts to reduce attrition rates and optimize drug efficacy. While applications of more advanced ML methods in studies of human metabolism are just beginning to form a body of literature, they have yielded promising results with implications for data-driven drug discovery.