Small Molecule Approach to RNA Targeting Binder Discovery (SMARTBind) Using Deep Learning Without Structural Input.

Area of Science:

• Computational chemistry
• Molecular biology
• Drug discovery

Background:

• Identifying small molecule binders to RNA is crucial for developing chemical probes and therapeutics.
• Current computational methods for RNA-targeted small molecule discovery often suffer from low accuracy and high computational costs.

Purpose of the Study:

• To introduce SMARTBind, a novel structure-agnostic framework for RNA-targeted small molecule binder discovery.
• To enhance the accuracy and efficiency of computational RNA-targeted drug discovery.

Main Methods:

• SMARTBind integrates an RNA large language model, pre-trained on extensive RNA sequence data, with contrastive learning.
• A ligand-specific decoy enhancement strategy was employed to address data scarcity and improve model generalizability.
• The framework utilizes only the RNA primary sequence for identifying small molecule binders and their binding sites.

Main Results:

• SMARTBind demonstrated superior performance compared to existing data-driven and docking-based methods across multiple benchmarks.
• The platform significantly reduced computational costs associated with RNA-targeted small molecule discovery.
• Novel small molecules targeting the precursor of microRNA-21 were identified and validated through in vitro and cellular assays.

Conclusions:

• SMARTBind offers a scalable, accurate, and structure-independent platform for discovering RNA-targeted small molecules.
• The developed framework has significant potential for advancing chemical probe and therapeutic development.
• This approach addresses key limitations in current computational methods for RNA-ligand interactions.