Disease gene identification by walking on multilayer heterogeneous networks

Identifying disease genes from a set of candidate genes is one of the main objectives in bioinformatics. Most of existing random walk algorithms to identify disease genes preferentially visit highlyconnected genes. Moreover, these algorithms access only a single gene network or an aggregated network of various gene data, leading to bias and incompleteness. To address these issues, we propose a topologically biased random walk with restart (BRWR) algorithm applied to multilayer-heterogeneous networks for the identification of disease genes. The BRWR by tuning the biased parameters can explore different layers of functional and physical interactions between proteins and genes, and can be conducted on heterogeneous networks in which the walkers can traverse a network with different types of nodes and edges. Experimental results show that the BRWR algorithm to identify candidate disease genes outperforms existing ones. Finally, the BRWR algorithm on multilayer-heterogeneous networks is used to predict disease genes implicated in the undiagnosed neonatal progeroid syndrome. Overall, by a proper tuning of the biases in the walks, our algorithm on different interaction sources can effectively improve the performance of candidate disease gene identification.