Figure 3

Genes with consistent hyper-variability across cancer types. (A) The 100 genes that most consistently show hyper variability across cancer types. We first define a normal range of expression using normal samples across all tissue types expecting that normal samples from a few tissue types will deviate from this normal range due to the tissue specificity of some genes. Each cell in the matrix indicates the percentage of samples of each type in which expression is outside the normal range. We observed that for the majority of genes, the percentage of samples in each normal tissue type outside normal range is close to either 0% (most tissues) or 100% (the small number of tissues for which the gene is specific). We also observed that in cancer, percentages are consistently away from 0% or 100%, indicating high variability. (B) Principal components for normal samples in adrenal cortex, colon, endometrium, kidney, skin, stomach and vulva. Circles illustrate profiles of normal expression for each tissue type. (C) Principal components for cancer samples. Increased variability is present in cancer but not manifested as multiple tightly defined sub-groups for each cancer type. Instead, we observe lack of regulation in cancer around tightly regulated regions of normal expression in each tissue type. The anti-profile method is based on this observation: stochastic departure from tightly regulated normal expression in these genes is characteristic in cancer and can be used in predictive settings.