Abstract

Contemporary genetic epidemiology, and in particular genome-wide association or gene expression studies, are characterized by significant multiple hypothesis testing and require assessment of whether association exists for each ‘test’, their strength and the evidence in relation to the others. Despite the common practice of comparing P-values across different hypothesis tests in genetics, P-values must be interpreted alongside the sample size and experimental design used for their computation. Here, we derive and apply new methodology for differential gene expression analysis using the evidential statistical paradigm, demonstrating how to measure statistical evidence in the presence of covariates, model misspecification, and for composite hypotheses. Existing methodology uses both frequentist and Bayesian paradigms but have largely been derived assuming analysis is carried out in small samples. With increasing size and complexity of differential gene expression studies, new measures of statistical evidence that are sensible and can leverage large sample properties amid the analysis of big data are now available.