Earth mover's distance for differential analysis of heterogeneous genomics data

Abstract

The identification of mRNA transcripts with expression levels associated with clinical and/or biological entities is a major goal of biomedical research. Several methods have been developed to identify genes expressed differentially between biological or clinical classes of interest. These methods use statistical approaches and are based on the assumption that between classes variations are high and within class variation is low. However, many problems in biology and biomedicine contain samples that show high levels of within class heterogeneity. This makes the identification of differentially expressed genes very challenging. To address this challenge we developed a differential expression analysis method based on a signal processing approach that uses the Earth Mover's Distance to measure the overall difference between the distributions of a gene's expression in two classes of samples and uses permutations to estimate the false discovery rate for each gene. Applying this method to simulated and real biological data, we show that this method outperforms the conventional differential expression analysis methods.