Reviews: Deep Recurrent Neural Network-Based Identification of Precursor microRNAs

The paper presents an LSTM model with an attention mechanism for classifying whether an RNA molecule is a pre-microRNA from its sequence and secondary structure. Class weights are incorporated into log-loss to account for class imbalance in the datasets used. The proposed method is extensively evaluated against 5 other existing methods on 3 datasets, and is shown to outperform the existing methods in most cases. The paper then attempts to give some insight into the features that are important for achieving good performance. First, by showing that secondary structures are largely responsible, but sequence features give a small boost, and second, by interpreting the attention weights using an adapted version of class activation mapping (proposed in an earlier CVPR paper).