Trials and tribulations of ‘omics data analysis: assessing quality of SIMCA-based multivariate models using examples from pulmonary medicine

Respiratory diseases are multifactorial heterogeneous diseases that have proved recalcitrant to understanding using focused molecular techniques. This trend has led to the rise of 'omics approaches (e.g., transcriptomics, proteomics) and subsequent acquisition of large-scale datasets consisting of multiple variables. In 'omics technology-based investigations, discrepancies between the number of variables analyzed (e.g., mRNA, proteins, metabolites) and the number of study subjects constitutes a major statistical challenge. The application of traditional univariate statistical methods (e.g., t-test) to these "short-and-wide" datasets may result in high numbers of false positives, while the predominant approach of p-value correction to account for these high false positive rates (e.g., FDR, Bonferroni) are associated with significant losses in statistical power. In other words, the benefit in decreased false positives must be counterbalanced with a concomitant loss in true positives. As an alternative, multivariate statistical analysis (MVA) is increasingly being employed to cope with 'omics-based data structures. When properly applied, MVA approaches can be powerful tools for integration and interpretation of complex 'omics-based datasets towards the goal of identifying biomarkers and/or subphenotypes. However, MVA methods are also prone to over-interpretation and misuse. A common software used in biomedical research to perform MVA-based analyses is the SIMCA package, which includes multiple MVA methods. In this opinion piece, we propose guidelines for minimum reporting standards for a SIMCA-based workflow, in terms of data preprocessing (e.g., normalization, scaling) and model statistics (number of components, R2, Q2, and CV-ANOVA p-value). Examples of these applications in recent COPD and asthma studies are provided. It is expected that readers will gain an increased understanding of the power and utility of MVA methods for applications in biomedical research.