Deciphering the complex: Methodological overview of statistical models to derive OMICS‐based biomarkers

Abstract

Recent technological advances in molecular biology have given rise to numerous large-scale datasets whose analysis imposes serious methodological challenges mainly relating to the size and complex structure of the data. Considerable experience in analyzing such data has been gained over the past decade, mainly in genetics, from the Genome-Wide Association Study era, and more recently in transcriptomics and metabolomics. Building upon the corresponding literature, we provide here a nontechnical overview of well-established methods used to analyze OMICS data within three main types of regression-based approaches: univariate models including multiple testing correction strategies, dimension reduction techniques, and variable selection models. Our methodological description focuses on methods for which ready-to-use implementations are available. We describe the main underlying assumptions, the main features, and advantages and limitations of each of the models. This descriptive summary constitutes a useful tool for driving methodological choices while analyzing OMICS data, especially in environmental epidemiology, where the emergence of the exposome concept clearly calls for unified methods to analyze marginally and jointly complex exposure and OMICS datasets.

Phthalates, perfluoroalkyl acids, metals and organochlorines and reproductive function: a multipollutant assessment in Greenlandic, Polish and Ukrainian men

“…A caveat of sPLS is that penalisation methods optimise prediction and will tend to select one of two highly correlated exposures with nearly equal coefficient sizes 19. Competing multipollutant modelling approaches exist22 (eg, tree-based models, elastic net penalised regression); however, simulation and validation assessments, specifically for data structures relevant for environmental epidemiology (often low dimensional), are lacking.…”

“…To address potential confounding, we employed a two-stage regression approach22: first, each outcome and each exposure were separately regressed on potential confounders, and second, sPLS regression models were fit inputting the residuals. We a priori selected the set of potential confounders,9 including study population and serum cotinine level, and variably age, body mass index (BMI), abstinence period and time of blood sampling, depending on the outcome (as specified in table 1).…”

The exposome concept: a challenge and a potential driver for environmental health research

“…This can lead to a high rate of false positives. Several statistical methods have been proposed to cope with this issue [20] and some of them have been applied [18,21]. However, systematic simulation studies are needed to characterise the efficiency of these approaches in the context of the exposome.…”

Pseudo-sample trajectories for variable interaction detection in Dissimilarity Partial Least Squares

“…Partly this is a logical consequence of the fact that most models are based on linear techniques such as Partial Least Squares (PLS) [1]. However, the presence of interaction between biological or chemical variables could easily be envisaged.…”