Prestige centrality-based functional outlier detection in gene expression analysis.

Abstract

Traditional gene expression analysis techniques capture an average gene expression state across sample replicates. However, the average signal across replicates will not capture activated gene networks in different states across replicates. For example, if a particular gene expression network is activated within a subset or all sample replicates, yet the activation state across the sample replicates differs by the specific genes activated in each replicate, the activation of this network will be washed out by averaging across replicates. This situation is likely to occur in single cell gene expression experiments or in noisy experimental settings where a small sub-population of cells contributes to the gene expression signature of interest. In this light, we developed a novel network-based approach which considers gene expression within each replicate across its entire gene expression profile, and identifies outliers across replicates. The power of this method is demonstrated by its ability to enrich for distant metastasis related genes derived from noisy expression data of CD44+CD24-/low tumor initiating cells.