MiRNN: An Improved Prediction Model of MicroRNA Precursors Using Gated Recurrent Units
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.
KeywordsMiRNN MicroRNA prediction Deep learning Bidirectional RNN GRUs End-to-end model
- 19.Park, S., Min, S., Choi, H., et al.: deepMiRGene: deep neural network based precursor microRNA prediction. arXiv preprint arXiv:1605.00017 (2016)
- 20.S, G.J.: miRBase. http://www.mirbase.org/
- 23.Fujita, P.A., Rhead, B., Zweig, A.S., et al.: The UCSC genome browser database: update 2011. Nucleic Acids Res. 39(suppl_1), D876–D882 (2010)Google Scholar
- 24.Hinton, G.E., Srivastava, N., Krizhevsky, A., et al.: Improving neural networks by preventing co-adaptation of feature detectors. arXiv preprint arXiv:1207.0580 (2012)