Metabolic Pathway Analysis: Advantages and Pitfalls for the Functional Interpretation of Metabolomics and Lipidomics Data

Over the past decades, pathway analysis has become one of the most commonly used approaches for the functional interpretation of metabolomics data. Although the approach is widely used, it is not well standardized and the impact of different methodologies on the functional outcome is not well understood. Using four publicly available datasets, we investigated two main aspects of topological pathway analysis, namely the consideration of non-human native enzymatic reactions (e.g., from microbiota) and the interconnectivity of individual pathways. The exclusion of non-human native reactions led to detached and poorly represented reaction networks and to loss of information. The consideration of connectivity between pathways led to better emphasis of certain central metabolites in the network; however, it occasionally overemphasized the hub compounds. We proposed and examined a penalization scheme to diminish the effect of such compounds in the pathway evaluation. In order to compare and assess the results between different methodologies, we also performed over-representation analysis of the same datasets. We believe that our findings will raise awareness on both the capabilities and shortcomings of the currently used pathway analysis practices in metabolomics. Additionally, it will provide insights on various methodologies and strategies that should be considered for the analysis and interpretation of metabolomics data.

Automated summary

Pathway analysis is commonly used for functional interpretation of metabolomics data, but lacks standardization and understanding of its impact on functional outcome. In this study, we investigated the consideration of non-human native enzymatic reactions and interconnectivity of pathways in pathway analysis. Exclusion of non-human reactions resulted in loss of information, while considering pathway connectivity provided better emphasis on certain metabolites but occasionally overemphasized central compounds. We proposed a penalization scheme to mitigate the effect of such compounds. We also performed over-representation analysis to compare and assess different methodologies. Our findings raise awareness on the capabilities and shortcomings of current pathway analysis practices in metabolomics.