Network-based global inference of human disease genes

Xuebing Wu¹, Rui Jiang¹, Michael Q Zhang^1,2 & Shao Li¹

Article highlights

Synopsis

The identification of genes responsible for human diseases is of great importance for both understanding human disease pathogenesis and improving clinical practice. Traditional gene-mapping approaches, such as linkage analysis and association studies (Botstein and Risch, 2003), though succeeded in identifying causative genes for many Mendelian diseases, have less power in identifying genes for complex and common diseases such as autism, inflammatory bowel disease, diabetes, coronary heart diseases, various cancers, and many others. In addition, these methods generally yield a large genomic region containing tens or even hundreds of candidate genes that need to be further analyzed, resulting in a task that is often expensive and laborious. Therefore, one of the greatest challenges in human genetics is to computationally prioritize candidate genes from the results of gene-mapping studies (even across the whole genome), to assist biologists in identifying causative genes.

Evidences from many sources suggest that similar phenotypes are caused by functionally related genes, referring to as the modular nature of human genetic diseases (Oti and Brunner, 2007). Many studies show that the causative genes for the same or similar diseases will generally reside in the same biological module, either a protein complex (Lage et al, 2007), a pathway (Wood et al, 2007), or a subnetwork of protein interactions (Lim et al, 2006). With this understanding, we propose a regression model that integrates human protein–protein interaction network and disease phenotype similarities with known gene–phenotype associations to infer novel gene–phenotype associations. Our method, named CIPHER (Correlating protein Interaction network and PHEnotype network to pRedict disease genes), can robustly uncover causative genes with high accuracy for many phenotypes, outperforming most of other disease gene-finding methods currently available.

CIPHER is applicable to phenotypes without any known causative genes by taking advantage of information from phenotypically similar diseases. Further, CIPHER is effective in prioritizing candidate genes from the whole genome instead of genetic loci, which enables us to perform genome-wide scans of causative genes for most of the recorded human phenotypes. In contrast, many existing methods rely on a list of known causative genes for the query phenotype or are limited to phenotypes with associated loci, which currently comprise less than half of the recorded human phenotypes (McKusick, 2007). The regression model of CIPHER can be further extended to the nonlinear situation to explore gene cooperative behavor in complex diseases. As application examples, three putative pairs of cooperative genes in breast cancer are identified.

A draft genetic landscape of human diseases is predicted by CIPHER. The inferred genome-wide molecular basis for 1126 phenotypes with at least one known causative genes in our data reveals the modular organization of human genotype–phenotype relationships (Figure 3). The modular disease landscape shows that a set of functionally related genes are implicated in a set of genetically overlapped phenotypes, suggesting interesting and meaningful connections between gene functions and specific disease categories. We further depict a much more comprehensive landscape for human diseases, including more than 14 000 human proteins and more than 5000 phenotypes. All these data are publicly available online. We hope the predicted genetic landscape will facilitate the discovery of disease genes in the future.

MORE ARTICLES LIKE THIS

These links to content published by NPG are automatically generated.