Publications

A fundamental problem in human health is the inference of disease-causing genes, with important applications to diagnosis and treatment. Previous work in this direction relied on knowledge of multiple loci associated with the disease, or causal genes for similar diseases, which limited its applicability. Here we present a new approach to causal gene prediction that is based on integrating protein-protein interaction network data with gene expression data under a condition of interest. The latter are used to derive a set of disease-related genes which is assumed to be in close proximity in the network to the causal genes. Our method applies a set-cover-like heuristic to identify a small set of genes that best “cover” the disease-related genes. We perform comprehensive simulations to validate our method and test its robustness to noise. In addition, we validate our method on real gene expression data and on gene specific knockouts. Finally, we apply it to suggest possible genes that are involved in myasthenia gravis.