Fig. 1

Overview of the differential mutation framework. Our method evaluates each gene for differential mutation when comparing cancer and healthy cohorts. For a cancer type of interest, we first count, for each individual, the number of somatic mutations found in each gene. Similarly, we use the 1000 Genomes sequencing data to count, for each individual, how many variants appear in each gene (top left). For each individual, we rank normalize the genes so that each gene has a score between 0 and 1 that reflects the relative number of mutations or variations that fall within it, compared to other genes within that individual (top middle). Next, for each gene, we aggregate its mutation and variation scores across healthy and cancer cohorts separately, resulting in a set of normalized variation scores as well as a set of normalized mutation scores (top right). We use each of these sets to build a histogram estimating the density of mutation or variant normalized scores. Shown here are the smoothed densities for the three most mutated genes in breast cancer (bottom right). Finally, in order to uncover whether a gene has a mutational profile that is very different between natural and cancer cohorts, we compute the difference between the two distributions using a modification of the classical Earth Mover’s Distance, which we refer to as a unidirectional Earth Mover’s Difference (uEMD). Genes with large differences between the two distributions are predicted as cancer genes (bottom left). See “Methods” for details