AI Platform Accelerates the Discovery of Novel mRNA Delivery Materials

Integrating artificial intelligence (AI) with advanced robotics to create self-driving labs (SDLs) is a promising approach to solving challenges in molecular discovery. A new SDL system named LUMI-lab combines large-scale molecular pre-training, active learning, and robotics to discover that brominated lipids—previously unrelated to mRNA delivery—can enhance the efficiency of mRNA entry into human cells. This study, led by researchers at the University of Toronto’s Leslie Dan Faculty of Pharmacy, was published in the journal Cell.

Supported by an AC Translational Research Grant from the University of Toronto’s Acceleration Consortium, LUMI-lab integrates a molecular foundation model with an automated robotic system. To the research team’s surprise, it identified a new class of mRNA-enhancing lipids—brominated lipid tails—as primary enhancers for increasing transfection efficiency.

“Through ten active learning cycles, LUMI-lab synthesized and tested over 1,700 novel lipid nanoparticles (LNPs), discovering brominated ionizable lipids that deliver mRNA into human lung cells with higher efficiency than approved benchmarks,” said Bowen Li, the GSK Chair in Pharmaceutics and Drug Delivery at the University of Toronto’s Leslie Dan Faculty of Pharmacy and an affiliate scientist at the Princess Margaret Cancer Centre, University Health Network. “The key advancement of this AI-driven system is that it independently identified bromination as a significant and meaningful feature without a prior hypothesis and without researchers telling it to look for it first.”

While mRNA therapies are one of the fastest-growing drug modalities, they currently rely on lipid nanoparticles for safe delivery to target areas in the human body. To date, only three LNPs have received FDA approval. Platforms like LUMI-lab are expanding the design space by accelerating the discovery of next-generation LNPs needed to unlock new therapeutic applications.

Furthermore, SDL models for drug discovery typically rely on large, high-quality datasets to perform well. In emerging fields like mRNA therapy development and delivery, the scarcity of historical data remains a major obstacle. To address this data shortage, the team opted for a foundation-based model, pre-training LUMI-lab on over 28 million molecular structures to allow it to learn general chemical patterns and structures before undertaking more specific tasks.

Figure 1. LUMI-lab is a powerful, data-efficient platform for autonomous discovery and optimization of molecules. (Xu, Y., et al., 2026)

“When integrated into an active learning framework, the model can be continuously optimized in a closed-loop workflow, further improving its predictive accuracy,” said Li, who also serves as the Canada Research Chair in RNA Vaccines and Therapeutics.

When tested in preclinical models, several newly discovered lipids outperformed the lipids used in Moderna’s COVID-19 mRNA vaccine. Although brominated lipids accounted for only 8% of the compound library used by LUMI-lab, they represented more than half of the top-performing candidates. Brominated lipids also demonstrated a safety profile similar to clinical benchmarks, supporting their potential for future therapeutic development.

“Next, we are expanding LUMI-lab to simultaneously optimize multiple clinically relevant attributes—not just delivery potency, but also safety, tolerability, and tissue selectivity,” Li said. “Through closed-loop AI predictions and automated experimentation, our goal is to shorten the design cycles for novel lipid materials and open up a larger, evidence-driven chemical space for mRNA therapies.”