Phoenics: a novel statistical approach for longitudinal metabolomic pathway analysis

• DOI: 10.1186/s12859-025-06118-z
• Publication Date: 2025-04-16T06:56:02Z
• Authors: Camille Guilmineau; Marie Tremblay-Fr...; Nathalie Vialaneix; Rémi Servien
• Keywords: Interpretability; Metabolic pathway; Biological data

Abstract

Background Metabolomics describes the metabolic profile of an organism at a given time by the concentrations of its constituent metabolites. When studied over time, metabolite concentrations can help understand the dynamical evolution of a biological process. However, metabolites are involved into sequences of chemical reactions, called metabolic pathways, related to a given biological function. Accounting for these pathways into statistical methods for metabolomic data is thus a relevant way to directly express results in terms of biological functions and to increase their interpretability.

Methods We propose a new method, phoenics, to perform differential analysis for longitudinal metabolomic data at the pathway level. In short, phoenics proceeds in two steps: First, the matrix of metabolite quantifications is transformed by a dimension reduction approach accounting for pathway information. Then, a mixed linear model is fitted on the transformed data.

Results This method was applied to semi-synthetic NMR data and two real NMR datasets assessing the effects of antibiotics and irritable bowel syndrome on feces. Results showed that phoenics properly controls the Type I error rate and has a better ability to detect differential metabolic pathways and to extract new impacted biological functions than alternative methods. The method is implemented in the package phoenics available on CRAN