A Deep Learning Approach for RNA-Compound Interaction Prediction with Binding Site Interpretability

Abstract

RNA-compound interaction prediction is crucial for expanding the therapeutic target space beyond proteins. However, existing models are limited by data scarcity and often lack interpretability. We present DeepRNA-DTI, the first sequence-based deep learning model for RNA-compound interaction prediction. Our model leverages pretrained embeddings from RNA-FM for RNA sequences and MoleBERT for compounds, capturing complex interaction patterns through attention mechanisms. DeepRNA-DTI jointly predicts drug-target interactions (DTI) and RNA binding sites, enhancing interpretability. Trained on datasets from the Protein Data Bank (PDB) and literature, DeepRNA-DTI demonstrates improved performance in RNA-compound interaction tasks compared to existing methods. Our approach offers valuable insights into binding sites and opens new avenues for RNA-targeted drug discovery.