Supervised Deep Learning Methods for Human pre-miRNA Identification

Precursor microRNAs (pre-miRNAs), generated from pri-miRNAs inside the nucleus are processed by different enzymes and present their powerful influence in gene silencing. Human microRNA precursor (pre-miRNA) can block the synthesis of malfunctioning proteins that provoke many human diseases, e.g. cancer, vascular diseases. The identification of human pre-miRNAs can be very useful in the treatment of these diseases. Till now, different machine learning computational methods were introduced and implemented whose performance heavily depends on the hand-crafted selected features, usually prepared by users or domain experts manually. However, some deep learning computational methods were recently introduced as well and they provided better performances than the machine learning methods. In this paper- we tried to introduce better models of convolutional neural networks (CNN) and recurrent neural networks (RNN) to classify human pre-miRNAs from the RNA sequences. Since it is the specialty of deep neural networks, the complex features of RNA sequences were extracted automatically by our CNN and RNN models. Multiple filters having different sizes with max-pooling layers and long short-term memory (LSTM) layers with different numbers of units were used in our CNN and RNN models respectively. We used "One-hot" encoding for vectorization and "zero-padding" to preprocess the data. We implemented k-fold cross-validation(CV) on our CNN model. We used the grid search technique to find the optimal hyperparameters to tune our models. We evaluated our models on an independent test dataset and the results were quite satisfactory in identifying human pre-miRNA. Source codes are available at - https://github.com/asrezoun/humanpre- miRNA