The overview of DeepIRES architecture. The original data are screened and redundancy is removed to build the training dataset. The input sequences are converted into input matrices using one-hot encoding, from which useful features are extracted by residual blocks and bidirectional GRU. Then, the self-attention layer is used to focus on the most crucial features, while a dense layer with sigmoid activation function is used to give the prediction result. Ten-fold validation is applied to assess the performance of the model on the training dataset.