AI Is Reshaping RNA-Targeted Drug Discovery, From Small Molecules to Oligonucleotides
核心洞察
Artificial intelligence and machine learning are increasingly embedded across RNA drug discovery, spanning structure prediction, target prioritization, molecular design, safety assessment, and experimental planning.
RNA-targeted small molecules and oligonucleotide therapeutics rely on distinct AI approaches: structure- and dynamics-driven methods versus sequence- and data-driven methods, respectively.
The approval of risdiplam established the feasibility of selectively targeting RNA with drug-like small molecules, while 2025 saw FDA approvals of the siRNA drugs fitusiran and plozasiran and the ASO donidalorsen.
Artificial intelligence is becoming an indispensable tool in RNA-targeted drug discovery, helping researchers navigate the multidimensional optimization problems that have long made RNA a difficult therapeutic target. A review published in Frontiers in Chemical Biology examines how AI and machine learning methods—including deep learning, foundation models, multimodal learning, and emerging agentic systems—are increasingly embedded in iterative workflows spanning structure modeling, target prioritization, molecular design, safety assessment, and experimental planning across both RNA-targeting small molecules and oligonucleotide therapeutics.
The clinical momentum behind RNA-based medicines has grown substantially. The approval of risdiplam, a small-molecule splicing modifier for spinal muscular atrophy (搜索), established the feasibility of selectively targeting RNA-mediated processes with drug-like molecules. Building on this, the oral splicing modulators branaplam and votoplam (PTC518), which lower huntingtin (搜索) (HTT) via pseudoexon inclusion, reached phase 2 trials for Huntington's disease (搜索); votoplam met its primary endpoint, whereas branaplam's VIBRANT-HD study was halted for neuropathy. In parallel, 2025 marked notable progress for oligonucleotide drugs, with FDA approval of three agents: the siRNA therapeutics fitusiran and plozasiran, and the antisense oligonucleotide (ASO) donidalorsen.
RNA Structure Prediction in the AI Era
RNA structure is central to RNA-targeted drug discovery because conformation governs biological function and how a therapeutic engages its target. For small molecules, the three-dimensional fold defines the shape and accessibility of ligandable pockets, which often emerge from internal loops, bulges, junctions, and other higher-order structural motifs. For ASOs and siRNAs, structure matters differently: because they act through Watson–Crick hybridization, efficacy depends mainly on target-site accessibility, with stable local secondary structure potentially occluding complementary regions and reducing potency.
Early AI applications focused on RNA secondary structure prediction. Deep learning methods such as SPOT-RNA, MXfold2, and UFold reframed folding as a data-driven pattern-recognition problem, improving prediction of base-pairing patterns including pseudoknots and non-canonical interactions. MXfold2 integrates learned folding scores with Turner nearest-neighbor thermodynamic parameters, improving robustness on novel families. More recent work has shifted toward RNA representation learning, exemplified by RNA-FM, a self-supervised RNA language model that generates transferable representations for secondary-structure prediction, contact modeling, and related tasks.
Three-dimensional prediction has progressed rapidly and is especially consequential for pocket definition. RoseTTAFoldNA extended deep learning prediction to protein–nucleic acid complexes, while RhoFold+ combined RNA language-model representations with a dedicated structure module for single-chain 3D prediction. AlphaFold3 broadened modeling to biomolecular complexes containing proteins, nucleic acids, ions, modified residues, and small molecules. However, important limitations remain. In the blinded, community-wide CASP15 assessment, RNA prediction methods showed substantial variation in accuracy across target classes and structural complexity, and independent analyses note that RNA prediction has not reached the reliability achieved for proteins, particularly for novel families and difficult folds.
AI for RNA-Targeting Small Molecules
Computational methods for predicting RNA–small molecule interactions divide into structure-based and ligand-based approaches. Structure-based methods include Rsite2, which uses secondary-structure-derived distance patterns to identify functional sites, and RBind, which models RNA tertiary structure as a nucleotide-interaction network. More recent machine learning methods build on this: Rlsite combines pre-trained RNA representations with graph attention networks, fpocketR provides RNA-specific pocket detection, and DRLiPS extends the framework from binding-site prediction to druggability assessment. RNAmigos2 uses graph-neural-network representations to predict RNA–ligand compatibility and accelerate virtual screening.
A common limitation across these methods is their reliance on experimentally determined or modeled RNA structures. Ligand-based approaches, by contrast, exploit similarity between compounds and do not require RNA structural information, but depend on the availability of known active compounds. Curated benchmark resources—including R-BIND, ROBIN, and Inforna—remain modest and biased toward well-studied targets, so generalization to novel RNA targets remains to be explored.
An emerging direction predicts RNA–small molecule interactions directly from sequence and ligand information. SMRTnet integrates RNA sequence, secondary structure, and compound representations to predict interactions, with prospective validation across disease targets. SMARTBind couples an RNA language model with contrastive learning to identify small-molecule binders and their binding sites from sequence alone, reducing dependence on tertiary structure.
A recurring theme is that induced fit is a core design problem: the binding-competent pocket is frequently shaped by the ligand rather than pre-formed in the free RNA. Ensemble-aware methods such as SHAMAN integrate molecular simulations, enhanced sampling, and small-molecule probes to identify ligandable sites across conformational ensembles. AI now assists at several points, including learned scoring functions such as AnnapuRNA, diffusion-based models such as DiffDock built on SE(3)-equivariant graph neural networks, flexible-docking methods such as DynamicBind, and end-to-end predictors such as AlphaFold3.
AI for Oligonucleotide Design
Unlike small molecules, oligonucleotides act through programmable base pairing, but potency and safety depend on more than complementarity—accessibility, strand architecture, chemical modification, nuclease stability, immune activation, off-target hybridization, and delivery are interdependent variables well suited to data-driven design. Chemical modification has been central to clinical maturation, with bridged nucleic acids improving duplex stability and affinity, while phosphorothioate linkages and stereodefined chemistry contributed to nuclease resistance and pharmacological optimization.
For siRNAs, early empirical rules provided the first practical framework for improving silencing efficacy by capturing features such as thermodynamic asymmetry, guide-strand selection, position-dependent nucleotide preferences, GC content, and target-site properties. As activity datasets accumulated, BIOPREDsi showed that experimentally measured sequence–activity relationships could be learned with neural networks and applied to genome-scale library design. More recent models treat efficacy as a sequence- and context-dependent interaction problem: OligoFormer integrates thermodynamic parameters, RNA-FM embeddings, and Transformer representations, while siRNADiscovery uses a graph neural network to combine empirical rules with non-empirical siRNA and mRNA features.
ASO design poses a distinct, modality-specific problem. Because ASOs are single-stranded, optimal design depends strongly on mechanism of action: gapmer ASOs require architectures supporting RNase H-mediated cleavage, whereas splice-switching and steric-blocking ASOs act by occupying regulatory regions and altering RNA processing or RNA–protein interactions. Machine learning tools are beginning to address these requirements: eSkip-Finder uses curated exon-skipping datasets to predict sequences likely to promote exon skipping, while ASOptimizer applies deep learning to optimize both sequence and chemical modification patterns for chemically modified, gene-regulating ASOs.
Safety-aware modeling is especially important for ASOs, because toxicity can arise from off-target hybridization, protein binding, immune activation, tissue exposure, and mechanism-specific effects. Early random-forest models partly predicted hepatotoxic potential from sequence and modification pattern, while motif-based studies linked specific sequence motifs to LNA-gapmer hepatotoxicity. Mechanistic analyses showed that high-affinity gapmers can trigger RNase H1-dependent off-target reduction of long pre-mRNA transcripts. This work emphasizes that ASO toxicity emerges from synergistic effects among nucleotide sequence, chemical modifications, and the positional context of those modifications, rather than from any single factor.
A Translational Gap in RNA Drug Development
The review's findings align with a broader call for integrated translational research in the RNA therapeutics field. A related Frontiers Research Topic notes that mRNA, siRNA, and ASO drugs have moved from molecular biology tools to approved medicines, with more than a dozen RNA-based drugs across these classes reaching regulatory approval since 2018. These modalities share a defining pharmacological challenge: unlike small molecules, their behavior in the body is governed by a complex interplay of formulation design, intracellular trafficking, innate immune sensing, and tissue-targeting mechanisms that traditional drug development frameworks were not built to handle.
Despite this rapid progress, the translational literature remains fragmented, with formulation scientists, pharmacokineticists, immunologists, clinical pharmacologists, and RNA chemical biologists rarely publishing within a shared framework. The Research Topic aims to build an integrated, translational reference collection that treats mRNA, siRNA, ASO therapeutics, and RNA-targeted small molecules as drugs first—examining how they are formulated for stability and delivery, how they distribute across tissues and act intracellularly, how safety and immunogenicity are characterized, and what lessons from approved RNA therapeutics, including risdiplam, can accelerate the next generation of medicines.