Prediction for the Domain of RNA with the Support Vector Machine ∗

The three-domain system is a biological classification of RNA. In bioinformatics, predicting the domain of RNA is helpful in the research of DNA and protein. By reviewing the related literature, we notice that many researches are conducted for domain prediction with only the primary structure. However, compared with the primary structure, the secondary structure of an RNA contains more discriminative information. Therefore, we propose an SVM-based prediction algorithm that considers both the features of primary and secondary structures. In our experiment, we adopt 1606 RNA sequences from RNase P, 5S ribosomal RNA and snoRNA databases. The experimental results show that our algorithm achieves 96.39%, 95.70%, and 95.46% accuracies by combining three softwares of secondary structure prediction, pknotsRG, NUPACK, and RNAstructure, respectively. Thus, our method is a new effective approach for predicting the domain of an RNA sequence.