miRNA-dis: microRNA precursor identification based on distance structure status pairs

Abstract

MicroRNA precursor identification is an important task in bioinformatics. Support Vector Machine (SVM) is one of the most effective machine learning methods used in this field. The performance of SVM-based methods depends on the vector representations of RNAs. However, the discriminative power of the existing feature vectors is limited, and many methods lack an interpretable model for analysis of characteristic sequence features. Prior studies have demonstrated that sequence or structure order effects were relevant for discrimination, but little work has explored how to use this kind of information for human pre-microRNA identification. In this study, in order to incorporate the structure-order information into the prediction, a method called “miRNA-dis” was proposed, in which the feature vector was constructed by the occurrence frequency of the “distance structure status pair” or just the “distance-pair”. Rigorous cross-validations on a much larger and more stringent newly constructed benchmark dataset showed that the miRNA-dis outperformed some state-of-the-art predictors in this area. Remarkably, miRNA-dis trained with human data can correctly predict 87.02% of the 4022 pre-miRNAs from 11 different species ranging from animals, plants and viruses. miRNA-dis would be a useful high throughput tool for large-scale analysis of microRNA precursors. In addition, the learnt model can be easily analyzed in terms of discriminative features, and some interesting patterns were discovered, which could reflect the characteristics of microRNAs. A user-friendly web-server of miRNA-dis was constructed, which is freely accessible to the public at the web-site on http://bioinformatics.hitsz.edu.cn/miRNA-dis/.