PNNL

Abstract

Comprehensive analyses of multiple biological components including nucleic acids, proteins, metabolites, and lipids (i.e., “multiomics”) provide unique insights into complex biological processes. Combining insights from these components through multiomics data integration enhances the depth and nuance of biological understanding available from these measurements. DiXerent data integration methods exist that provide insights at the sample- or feature-level, with the optimal method being dependent on the analysis objectives of a particular study. Among these methods, those that link components based on biological prior knowledge—pathway analyses—represent the most direct way of translating molecular-level observations into meaningful biological insights. However, pathway integration of small molecule metabolites and lipids faces significant barriers to full utilization in these analyses. Challenges include the fast turnover and complex interactions of small molecules compared to biological macromolecules, low metabolite annotation rates, isomerism among lipids, and a lack of lipid representation in existing pathway knowledge bases. While these issues stem from a variety of causes, improvements to multiomics pathway integration—including better incorporation of lipids into pathway knowledge bases and increased adoption of artificial intelligence approaches—can greatly enhance the utility of small molecule and especially lipidomics data. Here we describe and showcase integration of a multiomics dataset across diverse classes of algorithms to investigate how to use small molecules in biological hypothesis generation and provide a perspective on how to improve these approaches going forward.