TORONTO -- Integrating AI with advanced robotics to create self-driving labs (SDL) is a promising approach to tackling molecular discovery. A new SDL system, called LUMI-lab, combines large-scale molecular pretraining, active learning, and robotics, and has discovered that brominated lipids, not previously linked to mRNA delivery, enhance the efficiency of getting mRNA inside human cells.
The study, led by researchers at the University of Toronto's Leslie Dan Faculty of Pharmacy, was published today in Cell.
LUMI-lab (Large-scale Unsupervised Modeling followed by Iterative experiments), supported by an AC Translation research grant from U of T's Acceleration Consortium, integrates a molecular foundation model with automated robotic systems. Unexpected by the research team, it discovered a new class of mRNA‑boosting lipids, brominated lipid tails, as a major enhancer to improve transfection efficiency.
"Across ten active-learning cycles, LUMI-lab synthesized and tested more than 1,700 new lipid nanoparticles, uncovering brominated-tail ionizable lipids that deliver mRNA into human lung cells more efficiently than approved benchmarks," said Bowen Li, GSK Chair in Pharmaceutics and Drug Delivery, Leslie Dan Faculty of Pharmacy, University of Toronto, and affiliate scientist at the Princess Margaret Cancer Centre, University Health Network. "The key advance of this AI-driven system is that it independently identified bromination as an important, meaningful design feature without prior hypothesis or researchers telling it to look for it first."
While mRNA therapeutics are among the fastest-growing drug modalities, they currently rely on lipid nanoparticles (LNPs) for safe delivery to targeted areas of the human body, and to date, only three LNPs have received FDA approval. Platforms such as LUMI-lab are expanding the design landscape by accelerating the discovery of next-generation LNPs needed to unlock new therapeutic applications.
Additionally, SDL models for drug discovery depend on large, high-quality datasets to perform well. In emerging fields such as mRNA therapeutic development and delivery, the scarcity of historical data remains a major obstacle. To address data scarcity in this emerging field, the team selected a foundation-based model and pretrained LUMI on more than 28 million molecular structures, enabling it to learn general chemical patterns and structure before moving on to more specific tasks.
"When integrated into an active learning framework, the model can be continuously optimized in a closed-loop workflow, further enhancing its predictive accuracy," said Li, who also holds the Canada Research Chair in RNA Vaccines and Therapeutics.
Tested in preclinical models, some of the newly discovered lipids outperformed the lipid used in Moderna's COVID‑19 mRNA vaccine. While brominated lipids made up only 8 per cent of the chemical compound library used by LUMI-lab, they accounted for over half of the top-performing candidates. Brominated lipids also showed safety profiles similar to benchmark clinical lipids, supporting their potential for future therapeutic development.
"Next, we're expanding LUMI-lab to optimize multiple clinically relevant properties at once, not just delivery potency but also safety, tolerability, and tissue selectivity," said Li. "By closing the loop between AI predictions and automated experiments, we aim to shorten the design cycle for new lipid materials and open up a much larger, evidence-driven chemical space for mRNA therapeutics."
Funding:
Funded by Acceleration Consortium, GSK Canada, the Leslie Dan Faculty of Pharmacy, Canadian Institutes of Health Research, The Princess Margaret Cancer Foundation, the Connaught Fund, the J.P. Bickell Foundation, the Canadian Institutes of Health Research, the Natural Sciences and Engineering Research Council of Canada, the Canada Foundation for Innovation, Government of Canada, John R. Evans Leaders Fund, National Institutes of Health.
