Prediction of RNA Modifications Based on Multi-interval Nucleotide Pair Mutual Information

RNA modifications are modified nucleotides with extra chemical groups, which are critical in a number of cellular and biological processes. It is essential to know the occurrence sites of these modifications for fully understanding the biological mechanisms. Here, we propose a method for identification of these sites with local and global information of RNA sequence. The sites of three most frequently observed modifications, i.e., m
⁶
A, m
¹
A, A-to-I, are predicted simultaneously. To build the predictor, features are extracted from the RNA sequences through bidirectional multi-interval nucleotide pair mutual information method at first. Then, the Elastic Net is used to reduce redundant and noisy information to remain crucial features. At last, the support vector machine (SVM) classifier is applied as the predictor with these optimal features. It has been shown via cross-validations that the accuracy achieved by the powerful new predictor are better than previous methods. It is expected that the proposed methods may be considered as a useful computing tool for RNA analysis.