Multivariate analysis of metabolomic data to identify biological pathways modified by a clinical intervention.

Abstract

Introduction: High throughput metabolomic assays offer a huge opportunity to quantify the cellular processes underlying disease and intervention pathways. However, the multi-dimensional inter-relatedness between these processes coupled with the complex noisy measurement environment create a need for generation of new methods that move beyond simple pairwise associations.

Objectives: To evaluate the utility of a simple, multivariate, relational comparison method called CLARITY in the context of metabolomics data.

Methods: Nuclear magnetic resonance (NMR)-derived metabolomics data collected from the same individuals (N = 125) before and after a weight management intervention in By-Band-Sleeve (BBS), a clinical trial of metabolic and bariatric surgery, were used. First a traditional (univariate) linear mixed model approach was taken to identify metabolites that were changed post-intervention. The CLARITY method was then used to generate a covariance-based relational anomaly score with a view to increasing classification performance of the underlying cause of changes to the levels of and covariances between metabolites.

Results: CLARITY enabled further characterisation of metabolites identified in univariate linear regression analyses, differentiating those that exhibited covariance changes from those with mean changes only. An additional cluster of metabolites were identified as undergoing a relational change that would not have been detected using traditional methods.

Conclusions: Gathering insights about biological pathways from large-scale metabolomics data has the potential to inform the future design of modelling and laboratory experiments aimed at capturing the underlying biological processes relevant to disease.