Building multivariate systems biology models

Systems biology methods using large-scale "omics" data sets face unique challenges: integrating and analyzing near limitless data space, while recognizing and removing systematic variation or noise. Herein we propose a complementary multivariate analysis workflow to both integrate "omics" data from disparate sources and analyze the results for specific and unique sample correlations. This workflow combines principal component analysis (PCA), orthogonal projections to latent structures discriminate analysis (OPLS-DA), orthogonal 2 projections to latent structures (O2PLS), and shared and unique structures (SUS) plots. The workflow is demonstrated using data from a study in which ApoE3Leiden mice were fed an atherogenic diet consisting of increasing cholesterol levels followed by therapeutic intervention (fenofibrate, rosuvastatin, and LXR activator T-0901317). The levels of structural lipids (lipidomics) and free fatty acids in liver were quantified via liquid chromatography-mass spectrometry (LC-MS). The complementary workflow identified diglycerides as key hepatic metabolites affected by dietary cholesterol and drug intervention. Modeling of the three therapeutics for mice fed a high-cholesterol diet further highlighted diglycerides as metabolites of interest in atherogenesis, suggesting a role in eliciting chronic liver inflammation. In particular, O2PLS-based SUS2 plots showed that treatment with T-0901317 or rosuvastatin returned the diglyceride profile in high-cholesterol-fed mice to that of control animals.