Predicting miRNA-Disease Associations via a New MeSH Headings Representation of Diseases and eXtreme Gradient Boosting

Abstract

Taking into account the intrinsic high cost and time-consuming in traditional Vitro studies, a computational approach that can enable researchers to easily predict the potential miRNA-disease associations is imminently required. In this paper, we propose a computational method to predict potential associations between miRNAs and diseases via a new MeSH headings representation of diseases and eXtreme Gradient Boosting algorithm. Particularly, a novel MeSHHeading2vec method is first utilized to obtain a higher-quality MeSH heading representation of diseases, and then it is fused with miRNA functional similarity, disease semantic similarity and Gaussian interaction profile kernel similarity information to efficiently represent miRNA-disease pairs. Second, the deep auto-encoder neural network is adopted to extract the more representative feature subspace from the initial feature set. Finally, the eXtreme Gradient Boosting (XGBoost) algorithm is implemented for training and prediction. In the 5-fold cross-validation experiment, our method obtained average accuracy and AUC of 0.8668 and 0.9407, which performed better than many existing works.