Statistical strategies for relating metabolomics and proteomics data: a real case study in nutrition research area

The current investigations were carried out in the context of a nutritional case study aiming at assessing the postnatal impact of maternal dietary protein restriction during pregnancy and lactation on rat offspring plasma metabolome and hypothalamic proteome. Although data generated by different “Omics” technologies are usually considered and analyzed separately, their interrelation may offer a valuable opportunity for assessing the emerging ‘integrated biology’ concept. The overall strategy of analysis first investigated data pretreatment and variable selection for each dataset. Then, three multivariate analyses were applied to investigate the links between the abundance of metabolites and the expression of proteins collected on the same samples. Unfold principal component analysis and regularized canonical correlation analysis did not take into account the presence of groups of individuals related to the intervention study. On the contrary, the predictive MultiBlock Partial Least Squares method used this information. Regularized canonical correlation analysis appeared as a relevant approach to investigate of the relationships between the two datasets. However, in order to highlight the molecular compounds, proteins and metabolites, associated in interacting or common metabolic pathways for the experimental groups, MultiBlock partial least squares was the most appropriate method in the present nutritional case study.