RADAR: Representation Learning across Disease Information Networks for Similar Disease Detection

Discovering similar diseases can provide valuable clues for revealing their pathogenesis and predicting therapeutic drugs. Current methods for the measurement of disease similarity are mostly based on either semantic associations among diseases or functional associations between diseases and related genes. In either case, quantitative data are required. However this demand can not always be met. Moreover, many of these methods only use a single metric to evaluate disease similarity from an individual data source, which may lead to biased conclusions lacking consideration of other aspects. In this study, we proposed a novel framework, namely RADAR, for learning representations for diseases to measure their similarities. RADAR calculates disease similarity by different metrics fully based on the associations between diseases and other disease-related data, and constructs a multi-layer similarity network by integrating multiple disease similarity networks derived from multiple data sources in order to provide a comprehensive evaluation of disease similarities. A benchmark disease set and 90 random disease sets were used to assess the performance of RADAR. Experimental results demonstrated that RADAR is effective for detecting similar diseases.