Relating Disease-Gene Interaction Network With Disease-Associated ncRNAs

ncRNAs (non-coding RNAs) are increasingly recognized as critical biomarkers for many diseases. Computational predictions of the most promising disease-associated ncRNAs for biomedical screening are therefore of great importance. The biggest challenge in developing computational approaches for such inference is the integration of data features because of the rich variety of ncRNAs in genome. Moreover, current methods might suffer from the cold-start problem. In order to address the problems, a resource-allocation-based method was presented in this paper to predict disease-associated ncRNAs by incorporating only experimentally supported disease-gene interactions. Potential ncRNAs for a disease of interest were then ranked according to final allocated resource values. When applied to collected datasets for leave-one-out cross-validation on 537 diseases, our approach improved previous methods and demonstrated excellent prediction performance. Case studies on three important diseases suggested most of the top predicted results were validated by independent sources, which showed the usefulness of our method in real situations. Comprehensive prediction results of disease-associated ncRNAs were finally released for future biomedical identification.