Data analysis of MS-based clinical lipidomics studies with crossover design: A tutorial mini-review of statistical methods.

Clinical lipidomics using mass spectrometry (MS) is important to support discovery of biomarkers for diagnosis and understanding the pathophysiology of diseases. Frequently, lipidomics data from clinical studies have large variations among individuals because the human metabolome/lipidome is strongly influenced by genotype, daily activity, diet and gut flora. This inter-personal variability makes data analysis more complex and normally requires a large cohort for robust statistical analysis. Crossover designed experiments treat each subject as his or her own control, thereby reducing the between-subject variability, such that the effects of exposure/treatment are more likely to be identified when using a relatively small number of subjects. This design repeatedly samples an individual when crossing over from one treatment/exposure to another during the course of the study. The acquired datasets have a distinct data structure resulting from repeated longitudinal measurements. A variety of statistical methods are used in published crossover studies, but many appear to ignore the data structure inherent in the experimental design. An appropriate data analysis approach is critical to discovering robust clinical biomarkers. Hereby, we summarize the statistical methodologies suitable for clinical lipidomics studies using crossover design. To help understand and apply these methods to practical cases, we focused on the general concepts of statistical models in the context of analysis of metabolomics data without spending too much effort on mathematical details. Importantly, we aim to evaluate these methods and provide suggestions for data analysis and biomarker discovery. We applied the discussed methods on a MS-based lipidomics dataset from a double-blind random crossover designed clinical dietary intervention study. The strength and potential pitfalls of each method are briefly discussed and a suggestion for analytic workflow proposed.