AI-lab discovers brominated lipids that boost mRNA delivery to human cells

Researchers at the University of Toronto have developed LUMI-lab, an AI-enabled self-driving laboratory that has identified brominated lipid tails as a potent and previously unrecognised enhancer of messenger RNA delivery, with potential implications for next-generation lipid nanoparticle therapeutics

Artificial intelligence (AI) integrated with advanced robotics has begun to reshape molecular discovery, and a self-driving laboratory developed at the University of Toronto has now demonstrated how such systems can uncover unexpected chemical solutions to long-standing biomedical challenges.

Researchers at the Leslie Dan Faculty of Pharmacy, University of Toronto, Canada, have reported that their automated platform – known as LUMI-lab – has identified brominated lipid tails as a powerful enhancer of messenger RNA (mRNA) delivery into human cells. The finding is notable because brominated lipids had not previously been linked to improved mRNA transfection efficiency.

LUMI-lab, which stands for Large-scale Unsupervised Modeling followed by Iterative experiments, has combined large-scale molecular pretraining with active learning and robotic synthesis to create a closed-loop experimental system. The platform integrates a molecular foundation model with automated robotic systems capable of synthesise and test lipid nanoparticles at scale.

mRNA therapeutics have become one of the fastest-growing drug modalities, yet their clinical success depends heavily on lipid nanoparticles to enable safe and efficient delivery to specific tissues. To date, only three lipid nanoparticle formulations have received approval from the US Food and Drug Administration. The limited diversity of clinically validated lipid materials has constrained the expansion of mRNA technologies beyond vaccines and into broader therapeutic areas.

Against this backdrop, the Toronto team sought to expand the chemical design landscape. The scarcity of historical datasets in emerging fields such as mRNA delivery has posed a major obstacle to effective machine learning. To address this constraint, the researchers pretrained their foundation model on more than 28 million molecular structures. This approach enabled LUMI-lab to learn general chemical patterns and structural features before it moved to more specialised optimisation tasks.

Within an active learning framework, the system iteratively proposed candidate lipid structures, directed robotic synthesis and testing, and then refined its predictions in light of experimental results. Across ten active learning cycles, LUMI-lab synthesised and tested more than 1,700 lipid nanoparticles.

“Across ten active-learning cycles, LUMI-lab synthesised and tested more than 1,700 novel lipid nanoparticles, uncovering brominated-tail ionizable lipids that deliver mRNA into human lung cells more efficiently than approved benchmarks,” said Dr. Bowen Li, GSK chair in pharmaceutics and drug delivery at the Leslie Dan Faculty of Pharmacy and affiliate scientist at the Princess Margaret Cancer Centre, which forms part of University Health Network in Toronto.

“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,” Li added.

The emergence of brominated lipid tails proved striking given that brominated compounds accounted for only eight per cent of the chemical library explored by the system, they ended up representing more than half of the top-performing candidates. In preclinical models, several of these lipids surpassed the performance of the lipid component used in the COVID-19 mRNA vaccine developed by Moderna. At the same time, safety assessments indicated profiles comparable to established clinical lipid materials.

The implication is not merely incremental improvement. By autonomously identifying bromination as a beneficial chemical feature, LUMI-lab has demonstrated that AI systems can move beyond confirmation of established hypotheses to propose genuinely unexpected design principles. Such capacity could prove decisive in areas where human intuition has yet to map the full extent of viable chemical space.

The team has now begun to expand the platform to optimise several clinically relevant properties simultaneously.

“Next, we’re expanding LUMI-lab to optimise 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 novel lipid materials and open up a much larger, evidence-driven chemical space for mRNA therapeutics,’ he added.

Self-driving laboratories have attracted attention in recent years as a way to accelerate materials science and drug discovery. The LUMI-lab results suggest that, when combined with large-scale molecular pretraining and iterative experimentation, such systems can move from automation of routine tasks to genuine chemical insight.

For further reading please visit: 10.1016/j.cell.2026.01.012