Walking on multiple disease-gene networks to prioritize candidate genes

Uncovering causal genes for human inherited diseases, as the primary step toward understanding the pathogenesis of these diseases, requires a combined analysis of genetic and genomic data. Although bioinformatics methods have been designed to prioritize candidate genes resulting from genetic linkage analysis or association studies, the coverage of both diseases and genes in existing methods is quite limited, thereby preventing the scan of causal genes for a significant proportion of diseases at the whole-genome level. To overcome this limitation, we propose a method named pgWalk to prioritize candidate genes by integrating multiple phenomic and genomic data. We derive three types of phenotype similarities among 7719 diseases and nine types of functional similarities among 20327 genes. Based on a pair of phenotype and gene similarities, we construct a disease-gene network and then simulate the process that a random walker wanders on such a heterogeneous network to quantify the strength of association between a candidate gene and a query disease. A weighted version of the Fisher's method with dependent correction is adopted to integrate 27 scores obtained in this way, and a final q-value is calibrated for prioritizing candidate genes. A series of validation experiments are conducted to demonstrate the superior performance of this approach. We further show the effectiveness of this method in exome sequencing studies of autism and epileptic encephalopathies. An online service and the standalone software of pgWalk can be found at http://bioinfo.au.tsinghua.edu.cn/jianglab/pgwalk.