“A child receiving a single shot containing mRNA vaccines that protect against multiple diseases, all delivered with one lipid nanoparticle”—this is how Professor Drew Weissman, the 2023 Nobel laureate in Physiology or Medicine [
1], described the potential of messenger RNA (mRNA) therapy in an interview with Forbes [
2]. In 2024, the Nobel Prize was again awarded to RNA researchers, this time to Victor Ambros and Gary Ruvkun for the discovery of microRNA and its role in post-transcriptional gene regulation [
3], further underscoring the transformative capacity of RNA therapeutics in the 21st century healthcare. In recent years, RNA drugs have spearheaded advances in drug development for metabolic diseases, oncology, and preventive vaccines. Renowned journals such as
Nature [
4] and
Science [
5] have extensively reported on related innovations and approaches. The success of RNA drugs is highly foreseeable (as shown in
Fig. 1(a)), because they offer distinct advantages over traditional drug development, including remarkably higher success rates (e.g., Alnylam Pharmaceuticals claims that the cumulative transition rate of RNA interference (RNAi) drugs from clinical phase 1 to phase 3 reaches 64.4% [
6], while those of traditional drugs are only 5%-7%), shorter discovery timelines (usually months, whereas traditional drugs require years), and lower costs. Despite these promising advantages, experimental techniques such as clustered regularly interspaced short palindromic repeats (CRISPR) [
7] and computational methods like RNA sequencing [
8] have remained inadequate to meet the needs for speed and diversity in the development of RNA drugs, highlighting a pressing need to adopt innovative technologies to advance this field.
Artificial intelligence (AI) is demonstrating its potential for driving the future of RNA drug development, owing to its ability to leverage parallel computing and learn complex patterns from large-scale data. These capabilities enable AI-driven approaches to address the limitations of speed and diversity inherent in experimental and computational methodologies, which will not only improve drug development efficiency but also unlock new opportunities for identifying innovative drug candidates [
9] and introduce online simulation for drug validation. Successful applications in bioinformatics [
10], demonstrated by the high performance and accuracy of models such as AlphaFold [
11] and ESMFold [
12], further underscore AI’s transformative impact. A wealth of milestone results illustrate that AI primarily leverages data-driven, learning-strategy-driven, and deep-learning-driven approaches to drive substantial advancements in drug design, function prediction, and other critical domains. First, the
data-driven approach forms the foundation by leveraging large-scale datasets and rule mining techniques to extract meaningful patterns and relationships between RNA molecules and their structures or biological functions, thereby facilitating the optimization of molecular structures into ideal conformations [
13]. Second, the
learning-strategy-driven approach employs various explicit intervention techniques, such as causal inference [
14] and reinforcement learning [
15], to further optimize decision-making processes. This allows for the identification and selection of specific drugs based on human requirements, advancing beyond data analysis into active adaptation. Third, in the
deep-learning-driven approach, which represents a higher level of complexity and automation, breakthroughs such as large language models (e.g., Chat Generative Pre-trained Transformer (ChatGPT) [
16]) have revolutionized the analysis of long RNA sequences [
17], whereas generative models like diffusion models support the
de novo design of functional RNAs [
18].
AI-driven methodologies empower researchers to systematically explore novel RNA structures, identify promising drug candidates, and expedite the drug-discovery pipeline. Recent advancements in generative AI have enabled the design of RNA vaccines, with one example showing that a GEMORNA-derived mRNA elicited a significantly stronger antibody response than the BNT162b2 vaccine, despite being administered at the same dose level [
19]. However, existing technologies are predominantly task-specific and fail to offer a holistic understanding of drug molecular interactions. These limitations hinder the broader adoption of AI in drug development, particularly in creating online, interactive systems that can rapidly design effective treatments while addressing the specific needs of individual patients. Overcoming these challenges will necessitate a deeper exploration of how AI can be seamlessly embedded within RNA drug-development workflows.
In the future, a possible streamlined workflow for AI-driven RNA drug development (as shown in
Fig. 1(b)) will rely on an interactive, software-based system that will overcome the limitations of single-task models and the narrow perspective of molecular analysis in current approaches. The workflow features two key feedback loops: an internal loop (indicated by black-bordered arrows) focused on platform-based design to enhance AI model performance; and an external loop (indicated by borderless arrows) that integrates real-world data to continually refine drug development. The internal loop begins with a comprehensive digitization of RNA data, which is then used to design personalized drug candidates, followed by drug assessments, automated drug synthesis, and biological experiments for preliminary clinical validation. The selected drug candidates are matched with appropriate delivery systems and then placed into an online simulation, allowing for the early observation of delivery dynamics, drug action, and degradation processes within the human body, while also enabling real-time collaboration and feedback integration. The entire workflow will be integrated into an interactive platform, allowing researchers to seamlessly collaborate with AI systems, improving efficiency and adaptability. Based on the characteristics of this future workflow, several challenging research topics have been identified for the near term:
(1)
High-resolution comprehensive visualization. Current molecular analysis struggles to capture RNA’s complete state due to structural separation, limiting our understanding of interactions. By integrating high-resolution three-dimensional (3D) visualization technologies such as virtual reality and AI-driven models like Mamba [
20], researchers can simulate real-time RNA-target interactions and gain detailed insights into RNA structures and dynamic changes. Through these advancements, a comprehensive digital visualization model of RNA can be established, offering a more complete and interactive representation of RNA structures and their behavior in biological processes.
(2) Personalized RNA drug discovery. The need for personalized RNA drugs comes from tailoring treatments to individual genetic profiles for improved efficacy and fewer side effects. Generative models using condition-based sampling and diffusion techniques can adapt RNA sequences by integrating patient-specific data. Subsequently, reinforcement learning and adaptive models can help refine these molecules by targeting specific mutations while ensuring stability and minimizing off-target effects, thereby creating highly personalized therapies.
(3)
Editable RNA generation platform. An interactive RNA generation platform leverages advanced AI models, such as ChatGPT [
16], to facilitate real-time human-AI collaboration in RNA design. By integrating these capabilities, the platform allows researchers to generate, refine, and edit RNA sequences based on criteria such as stability and target specificity, ultimately offering greater convenience and improved precision in RNA drug development.
The notable economic and social benefits of AI-driven RNA drug development stem from comprehensive digitization, personalized approaches, real-time interactivity, and ultimate integration into a unified software platform [
21]. AI-driven automation reduces labor-intensive tasks, enabling faster and more accurate RNA-target identification; this results in cost savings and the expedited testing of RNA therapies, significantly shortening the time needed to bring new drugs to market. As the platform scales industrially, it ensures consistent drug quality and greater cost efficiency through optimized, repeatable processes. The cumulative effect is a more sustainable and economical development model with widespread benefits—including but not limited to reduced healthcare costs and expanded access to personalized therapies.
In this opinion paper, a potential workflow for future AI-driven RNA drug development was envisioned, where an integrated platform enables real-time data sharing, provides cloud-based editing for customized drug design, and facilitates the online simulation of drug action processes. We hope this perspective will help shape the thinking of the AI-driven RNA drug-development paradigm.
CRediT authorship contribution statement
Yilin Yan: Writing - original draft, Conceptualization. Tianyu Wu: Writing - review & editing. Honglin Li: Writing - review & editing, Supervision. Yang Tang: Writing - review & editing, Supervision. Feng Qian: Supervision.
Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by the National Key Research and Development Program of China (2022YFB3305900), the Science and Technology Innovation Action Plan Computational Biology Program (24JS2830400), the Fundamental Research Funds for the Central Universities (222202517006), and the Programme of Introducing Talents of Discipline to Universities (the 111 Project) (B17017).
PDF
(879KB)
