Network-based gene-disease prioritization using PROPHNET

Motivation and Objectives A major goal in biomedicine is to determine the underlying genetic causes of human diseases in order to better understand them and support their prevention and treatment. However, the genetic bases of many multifactorial diseases are still unclear, and high-throughput technologies typically report hundred or thousands of genes associated to a disease of interest. In this context is where gene-disease prioritization methods are of use. These computational methods make use of available data to obtain prioritized lists of genes (diseases) associated to a query set of diseases (genes). Prioritization is based on “guilt-byassociation” which states that biological entities that are associated or interacting are more likely to share function. This allows to infer new relationships from already known interactions. Many network-based prioritization methods have been proposed in the literature, performing well across different validation tests (Wang et al., 2011; Barabasi et al., 2011; Navlakha et al., 2010). We focus our study on two recent methods: rcNet (Hwang et al., 2011) and domainRBF (Zhang et al., 2011) since they outperform previous methods. Despite their good performance, these methods have clear limitations. First, they are strongly tailored to a specific domain of interest (gene-disease prioritization for rcNet and protein domain-disease prioritization for domainRBF, respectively). Hence, they cannot be applied to the prioritization of other biological entities of interest. Second, they do not allow to consider more than two types of networks for performing the prioritization (gene and disease networks in rcNet and domain and disease networks in domainRBF). However, we hypothesise that simultaneously integrating data from more than two complementary sources may improve the obtained results. For example, a gene-disease prioritization may benefit from known relationships between genes and diseases, but also from known interactions between drugs targeting certain genes to prevent or treat a specific disease. We present ProphNet, a generic method of prioritization that achieves a better performance by integrating and propagating information in an arbitrary number of heterogeneous data networks. Our method is generic since it allows to prioritize any biological entity of any kind with respect to some biological entities of another kind. Therefore, the user can customize the goal of the prioritization task (disease-gene, domaindisease, etc.) and the networks that are being taking into account for prophNet to achieve this goal. ProphNet is available as a web application at http://genome2.ugr.es/prophnet/. MATLAB source code, datasets and detailed experiments can also be downloaded at http://genome2.ugr. es/prophnet/prophnet.zip. In this talk we present prophNet and compare its results to those obtained by rcNet and domainRBF in two cases of study associated to gene-disease and domaindisease prioritization, respectively.