Identifying RNA-small Molecule Binding Sites Using Geometric Deep Learning with Language Models.

Area of Science:

• Computational biology
• Drug discovery
• Structural bioinformatics

Background:

• RNA molecules are key therapeutic targets, but identifying small molecule binders is challenging.
• Accurate computational methods for predicting RNA-small molecule interactions are needed.
• Large language models (LLMs) show promise for biological sequence analysis.

Purpose of the Study:

• To develop an accurate and efficient computational framework for predicting RNA-small molecule binding sites.
• To leverage advances in RNA-specific LLMs and geometric deep learning for this task.
• To improve the discovery of novel RNA-targeted therapeutics.

Main Methods:

• Developed RNABind, an embedding-informed geometric deep learning framework.
• Integrated RNA LLMs with geometric deep learning to encode RNA sequence and structure.
• Compiled the largest RNA-small molecule interaction dataset from multi-chain complexes.
• Evaluated eight pre-trained RNA LLMs on binding site prediction.

Main Results:

• RNABind significantly outperforms existing state-of-the-art methods for RNA-small molecule binding site prediction.
• The framework effectively utilizes both RNA sequence and structural information.
• Comprehensive evaluation demonstrated the robustness and accuracy of RNABind.

Conclusions:

• RNABind offers a powerful computational tool for predicting RNA-small molecule binding sites.
• This work facilitates future innovations in RNA-targeted drug discovery.
• The study highlights the potential of integrating LLMs and geometric deep learning in structural bioinformatics.