DeepRNA-DTI: a deep learning approach for RNA-compound interaction prediction with binding site interpretability.

Area of Science:

• Computational chemistry
• Drug discovery
• Bioinformatics

Background:

• RNA-targeted therapeutics offer a new avenue for drug development.
• Predicting RNA-compound interactions is difficult due to data limitations and RNA complexity.

Purpose of the Study:

• To develop DeepRNA-DTI, a deep learning model for predicting RNA-compound interactions and binding sites.
• To provide mechanistic insights into RNA-compound recognition.

Main Methods:

• Utilized transfer learning with RNA-FM and Mole-BERT embeddings.
• Employed a multitask learning framework for interaction and binding site prediction.
• Trained on a comprehensive dataset from the Protein Data Bank and literature.

Main Results:

• DeepRNA-DTI outperforms existing methods in RNA-compound interaction prediction.
• Demonstrated robust generalization across diverse RNA subtypes.
• Successfully screened millions of compounds, identifying known binders and novel scaffolds for pre-miR-21.

Conclusions:

• DeepRNA-DTI enhances the identification of RNA-targeting compounds.
• The model offers valuable insights for RNA-directed drug discovery.
• Publicly available code and data facilitate further research.