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

Accurate identification of small molecule binders to RNA is critical for chemical probes and therapeutics. Computational approaches offer a cost-effective strategy to identify small molecules targeting RNA but are often limited by poor predictive accuracy and high computational demands. Here, we introduce S mall M olecule A pproaches to R NA T argeting Bind er Discovery (SMARTBind), a structure-agnostic ligand discovery framework that combines an RNA large language model with contrastive learning and a ligand-specific decoy enhancement strategy. The RNA language model, pre-trained on millions of RNA sequences, together with the decoy enhancement strategy, addresses data scarcity and improves model generalizability. SMARTBind uses only RNA primary sequence to identify small molecule binders and their binding sites accurately. Across multiple benchmarks and case studies, SMARTBind outperforms existing data-driven and docking-based methods while significantly reducing computational cost. In a real-world application, SMARTBind identified novel small molecules targeting the precursor of oncogenic microRNA-21, validated by in vitro and cellular assays. These results highlight SMARTBind’s potential as a scalable, accurate, and structure-independent platform for RNA-targeted small molecule discovery.