HybridRanker: Integrating network structure and disease knowledge to prioritize cancer candidate genes

Abstract

One of the notable fields in studying the genetics of cancer is disease gene identification which affects disease treatment and drug discovery. Many researches have been done in this field. Genome-wide association studies (GWAS) are one of them that focus on the identification of diseases-susceptible loci on chromosomes. Recently, computational approaches, known as gene prioritization methods have been used to identify candidate disease genes. Gene prioritization methods integrate several data sources to discover and prioritize the most probable candidate disease genes. In this paper, we propose a prioritization method, called HybridRanker which is a network-based technique and it also uses experimental data to identify candidate cancer genes. We apply our proposed method on colorectal cancer data. It is notable to say that in HybridRanker, for considering both local and global network information of a protein-protein interaction network, different algorithms such as shortest-path, random walk with restart and network propagation are exploited. By using these algorithms, initial scores are given to genes within the network. Furthermore, by looking through diseases with similar symptoms and also comorbid diseases and by extracting their causing genes, the gene scores are recalculated. We also use gene-phenotype relations for an additional scoring of the candidate genes. Our method is validated and compared with other prioritization methods in leave one-out cross-validation and the comparison results show the better performance of the HybridRanker.