Deep Learning Approaches for the RNA-RNA Interaction Prediction Problem

RNA-RNA interactions (RRIs) play a fundamental role in numerous biological processes, such as gene regulation, viral replication, and RNA-based therapeutics. In recent years, deep learning approaches have emerged as powerful tools for predicting RRIs. This study provides a comparative analysis of deep learning techniques applied to the RNA-RNA interaction prediction problem. We evaluate representative architectures, including feedforward neural networks, recurrent models such as gated recurrent units (GRUs), transformer-based models, and pretrained biomedical language models. The comparison is conducted using benchmark RRI datasets, considering performance metrics such as accuracy, precision, recall, F1-score, AUC-ROC, and Matthews correlation coefficient. Additionally, we examine feature representation strategies, including sequence encoding and secondary structure features, and conduct ablation analyses to assess their contribution to prediction performance. The results demonstrate differences in predictive performance across the evaluated architectures and highlight the contribution of structural features to RRI prediction. These findings provide a comparative perspective on deep learning approaches for RNA-RNA interaction prediction and identify directions for further development of computational RNA interactomics.