Prioritizing Disease Genes with Label Propagation on a Heterogeneous Network
Evidences from recent studies suggest that disease-causative genes can be identified more accurately from the modular structures in a heterogeneous network that integrates a disease phenotype similarity subnetwork, a gene-gene interaction subnetwork and a phenotype-gene association subnetwork. However, it is a challenging machine learning problem to explore a heterogeneous network comprising several subnetworks since each subnetwork contains its own cluster structures that need to be explored independently. We introduce a general regularization framework and an intuitive and efficient algorithm called MINProp for propagating information between an arbitrary number of subnetworks in a heterogeneous network. Our algorithm performs label propagation on each individual subnetwork with the current label information derived from all the subnetworks, and repeats this step until convergence to the global optimal solution of the convex objective function in the regularization framework.
In simulations, we show that MINProp can significantly improve the ranking task by removing the biases introduced by the discrepancy among the subnetworks. We then tested MINProp for disease gene prioritization on a large-scale heterogeneous network containing 8919 genes and 5080 OMIM phenotypes. MINProp achieved competitive or better overall gene ranking performance than CIPHER and random walk with restart, two best-performing methods for disease gene prioritization, in both leave-one-out cross-validation and the case study of discovering new disease phenotype-gene associations added to OMIM after May 2007. We also validated that MINProp can specifically improve the ranking of those disease genes in the dense modules of the gene-gene interaction network. Furthermore, MINProp revealed interesting global modular structure of human disease phenotype-gene associations and new associations that are only reported recently.