MiRNN: An Improved Prediction Model of MicroRNA Precursors Using Gated Recurrent Units

Abstract

MicroRNAs (miRNAs) are small noncoding RNAs that derived from hairpin-forming miRNA precursors (pre-miRNAs) and regulating gene expression at the post-transcriptional level. Many sophisticated computational tools have been developed for miRNA prediction. However, all these existing approaches for predicting miRNA require large amounts of task-specific knowledge in the form of handcrafted features and data pre-processing. In this article, we introduce MiRNN (MiRNN is available at https://github.com/CadenC/MiRNN ), a novel computational predictor based on bidirectional gated recurrent units (GRUs). Our system is truly end-to-end, requiring no feature engineering or data preprocessing, thus making it applicable to a wide range of sequence classification tasks. Its main purpose is to omit the procedure of feature extraction and to provide accurate prediction by using the high-level features extracted from the bidirectional recurrent neural network. The experimental results show that MiRNN can produce state-of-the-art performance on pre-miRNA prediction task. The overall prediction accuracy of our model on miRBase data sets is 93.70%. In addition, we trained our model on various clade specific dataset and obtained increased accuracy.