PRISM: A structure-guided computational approach for identifying RNA-targeting small molecules by integrating 3D...

Targeting RNA with small molecules represents a promising frontier for therapeutic intervention, offering new opportunities for "undruggable" targets. However, accurately predicting the interaction between small molecules and RNA remains challenging, primarily due to the highly dynamic 3D conformations of RNA which are often overlooked by methods relying solely on linear sequences. While recent advancements demonstrate that RSIs can be predicted using secondary structures to bypass 3D folding inaccuracies, it is essential to note that such models are still fundamentally trained on labels derived from experimental 3D structures in the Protein Data Bank (PDB). Here, we present PRISM, a structure-guided computational framework designed to predict RNA-small molecule binding by integrating four complementary chemical and biological representations: the predicted 3D atomic structure of the RNA, its nucleotide sequence, the 2D topological graph of the ligand, and its chemical composition (SMILES). To capture the complex spatial arrangement of RNA binding pockets, PRISM utilizes a geometry-aware modeling approach that respects the 3D spatial orientation of atoms. This structural information is combined with the ligand's chemical topology and sequence data through a dynamic weighting mechanism. This mechanism mimics a rational drug design process by automatically determining whether 3D structural constraints or sequence motifs are more critical for a specific interaction. Extensive evaluations across diverse benchmarks and a large-scale High-Throughput Screenin dataset demonstrate PRISM's robustness. Notably, PRISM maintains high predictive performance even when relying on predicted 3D structures, effectively distinguishing rare active hits from vast non-binding decoys in realistic, imbalanced screening scenarios. By effectively leveraging predicted 3D conformational data alongside chemical topology, PRISM provides an interpretable and powerful tool to support structure-based drug discovery (SBDD) for novel RNA therapeutics.