Â鶹ÒùÔº

The majority of RNAs in each of our cells don't code for any of the thousands of proteins that make up our bodies. Instead, noncoding RNAs have critical roles in many biological processes—such as gene expression—making them ideal targets for a variety of ailments, including cancers. Despite that, the first RNA-targeting therapeutic in the market was launched only five years ago, and the vast majority of all Food and Drug Administration–approved drugs target proteins.

Targeting ncRNAs with drugs presents a much larger therapeutic opportunity than targeting proteins, but the currently available drug discovery tools are slow and computationally intensive.

At Vanderbilt University, Assistant Professor of Molecular Â鶹ÒùÔºiology and Biophysics and core member of the Center for Applied AI in Protein Dynamics Carlos Oliver is working with collaborators to unlock the untapped potential of ncRNAs, as they are a promising family of targets for the development of novel small-molecule therapeutics. Oliver's work was in Nature Communications and was completed in collaboration with researchers at McGill University in Montreal, Canada, and the École des Mines de Paris in Paris, France.

We sat down with Oliver to discuss targeting RNAs in medicine, the problems with current technologies, RNAmigos2, and the implications of the team's findings.

What issue/problem does your research address?

Our research addresses the challenge of discovering small-molecule drugs that target RNA, a promising yet underexplored frontier in medicine. While most drugs target proteins, only a tiny fraction of RNA has been harnessed beyond the small portion that codes for proteins, despite RNA's critical role in diseases like cancer and viral infections.

Traditional drug screening methods, like molecular docking, are too slow and computationally intensive to efficiently explore RNA's vast potential, limiting progress in this field.

What was unique about your approach to the research?

We developed RNAmigos2, a deep-learning tool that accelerates RNA-targeted drug screening by 10,000 times compared to traditional docking, which is physics-based and computationally expensive. RNAmigos2 uses a novel combination of coarse-grained 3D RNA modeling, synthetic docking data for training, and RNA-specific self-supervision to overcome the scarcity of RNA-ligand data. Unlike prior methods, it was rigorously validated on in vitro data it had never encountered during training, proving its real-world reliability.

What were your top three findings?

1. RNAmigos2 ranks active compounds in the top 2.8% of candidates across diverse RNA targets in the Protein Data Bank [the single repository of information about the 3D structures of proteins, nucleic acids, and complex assemblies], matching or surpassing docking accuracy while running in seconds instead of hours.
2. Our tool works synergistically with more fine-grained docking software by rapidly suggesting promising compounds for further validation. With a hybrid approach, we improve the discovery efficiency in all testing RNAs.
3. In a blind test, RNAmigos2 screened 20,000 compounds in minutes against unseen RNA riboswitches from an in vitro microarray. It achieved a 2.93-fold enrichment of active molecules at the top 1% and improved the diversity of hits over those obtained by docking.

What do you hope will be achieved with the research results in the short term?

We hope RNAmigos2, , will transform early-stage RNA drug discovery. By enabling researchers to rapidly screen large compound libraries and pinpoint promising candidates, it could streamline lab testing and accelerate the development of RNA-targeted therapies for clinical trials within the next few years.

What are your highest translational aspirations that might result from this research?

Our long-term vision is to unlock RNA as a primary drug target, revolutionizing treatment for complex diseases. We aspire to see RNAmigos2 contribute to new therapies for conditions such as cancer, genetic disorders, and viral infections in which RNA's regulatory roles are pivotal—potentially expanding the druggable space and advancing precision medicine over the next decade.

Who or what made the difference in your research?

Our team merged RNA biology, deep learning, and cheminformatics expertise, and this diverse collaboration was the backbone of our success. Public resources like the Protein Data Bank and the ChEMBL database, alongside tools like rDock, provided critical data and benchmarks. Incremental steps, such as refining the 2.5D graph representation of RNA structures and tweaking neural network designs, steadily built our breakthrough.

Where is this research taking you next?

As the state-of-the-art in structure-based RNA drug discovery, RNAmigos2 sets the stage for further innovation. I will be working to integrate it with binding site–prediction tools to identify RNA targets at the genome level and to build partnerships with experimental groups for deeper validation. With our open-source code, datasets, and model weights, global researchers can enhance its accuracy, apply it to new RNA families, and drive the next wave of RNA therapeutics.