Improving insights from metabolomic functional analysis combining multivariate tools.

BACKGROUND

Metabolomics is a scientific field that relies on the comprehensive analysis of metabolites to provide direct insights into functional processes in biological systems. Metabolomic data provides valuable insights into the functional processes of biological systems, often analyzed through univariate and multivariate approaches, and well as with functional or pathway analysis using different methods such as mummichog. Yet, the integration of results from these sources to aid the interpretation of their biological significance remains challenging. This represents a significant bottleneck limiting the applicability of multivariate analysis of metabolomic data, despite its potential for providing deep biological insights.

RESULTS

In this work we propose two straightforward methods to facilitate the interpretation of results from multivariate analysis and functional metabolic analysis using: i) p-values from multivariate tests as input in functional analysis, and ii) cluster-CV to assess the impact on the predictive performance of a multivariate model at the pathway level. Four simulated data sets were analyzed including a data set with no class separation, and three data sets with a statistically significant discrimination between classes by including either univariate, multivariate, or both types of discriminant effects. The data sets were analyzed using univariate tests and OPLS-DA. Furthermore, p-values for each feature estimated by univariate analysis and OPLS-DA were used as input for functional analysis in mummichog. Cluster-CV was then used to assess the effect of detected metabolic pathways on the class separation observed by OPLS-DA.

SIGNIFICANCE

Through simulated data, we show how these approaches enhance the interpretation of biological effects driving multivariate models and support the identification of altered pathways not detected by univariate analysis. By providing a deeper understanding of metabolic phenotypes, these methods might improve the biological insights derived from statistical and functional analysis of future or previous studies.