New research at the Leslie Dan Faculty of Pharmacy describes a proof of concept for a new artificial intelligence (AI) tool to design lipid nanoparticles (LNPs) that meet more than one objective. In a recent paper published in Nature Biotechnology, Assistant Professor Bowen Li and his colleagues described how the system could be applied to design a lipid nanoparticle that can target osteoarthritis in the knee while avoiding liver tissue, thus improving efficacy and safety of the therapy.
“Our lab focuses on incorporating AI into nanomedicine and mRNA delivery, and we are always thinking about how we can use AI to advance next-generation delivery systems and address some fundamental questions in the field,” says Li.
“We’ve previously used AI to accelerate the development and testing of nanoparticles, but now we need LNPs that can address more than one question at a time. This research demonstrates the feasibility of a multi-objective AI model – the first of its kind in the mRNA delivery and mRNA medicine fields.”
“We’ve previously used AI to accelerate the development and testing of nanoparticles, but now we need LNPs that can address more than one question at a time.”
LNPs are an essential part of the system that delivers mRNA to cells. They are designed to surround the mRNA and ensure that it is delivered to and can enter specific, targeted cells. And while they are highly effective at delivering mRNA to their target cells, they can also accumulate in the liver, increasing the risk of side effects.
Before the application of AI in this field, researchers would identify potential LNPs by synthesizing and screening large libraries of LNPs, looking for activity in the target cells. AI has sped up this process significantly, as systems can be trained on the structure and function of large libraries of lipids and predict which will be most effective for researchers to test experimentally.
Li’s group has been at the forefront of this research field. In 2024, Li and collaborators developed an AI platform called AGILE, an innovative system to predict LNP candidates for a target tissue. But it was designed to meet the single objective of maximizing delivery to a specific tissue. In practice, though, therapeutics usually have more than one objective, for example to reach the target tissue and not be taken up by liver cells.
Li team demonstrated feasibility of MOLEA in osteoarthritis
To address this challenge, Li’s group developed multi-objective LNP engineering with artificial intelligence (MOLEA), an AI tool that can design LNPs that both target specific tissues and avoid others to optimize specificity and efficacy. In the new Nature Biotechnology paper, they demonstrated that MOLEA could predict an effective and selective LNP to deliver gene editing tools to treat osteoarthritis.
They first pre-trained MOLEA with data from a virtual library of 15 million lipid structures, which provided foundational knowledge for the system. They then fine-tuned the system using data from 1,152 lipids experimentally tested on the specific cell types of interest. Based on this data, MOLEA predicted the most effective LNP from a library of lipids, and the team validated the prediction.
Following the four steps, MOLEA identified a candidate LNP that was highly effective at delivering mRNA gene editing tools to chondrocytes (cells in the cartilage of knee joints) in animal models. The LNP was also much more selective than the clinical benchmark, with a significant improvement in uptake in chondrocytes than in liver cells, thus improving the effectiveness and reducing the side effects of the treatment.
“In medicine, you often need to achieve a balance between two or more objectives. In this case, the multi-objective goal was to deliver mRNA to one specific tissue and minimize leakage to the other organ,” says Li. “We designed an LNP that not only achieves high performance in the knee joint, but also minimized its impact on the liver. This represents the first work in this field to develop an AI model that can address this multi-objective question.”
“We designed an LNP that not only achieves high performance in the knee joint, but also minimized its impact on the liver. This represents the first work in this field to develop an AI model that can address this multi-objective question.”
As the first demonstration of the platform, Li says that MOLEA could be used to design LNPs that meet additional objectives and can be tailored for any tissue or condition with an mRNA-based therapeutic.
“This work shows that using an AI model to address multi-objective questions in mRNA delivery or RNA medicine is feasible. We’ve demonstrated that there is a pathway, and now we can further refine the model to address different types of multi-objective questions,” says Li.
“Our goal is to expand the system to address even more objectives and complicated questions, and ultimately can be used to develop clinical applications.”
More News
Image
Dean's Message: Stepping up to address Ontario’s health needs
New provincial investment to U of T will support the growth of the PharmD program at the Leslie Dan Faculty of Pharmacy.
Read More
Image
Pharmacy alum and Second World War veteran Tom Lock among Chinese-Canadians honoured on Canada Post stamp
Tom Lock, one of the first Chinese-Canadian pharmacists in Toronto, supported Chinese community and downtown neighbourhood with pharmacy services and support.
Read More
Image
Faculty Research Retreat explores opportunities for collaboration, innovation and impact
From advancing primary care transformation to accelerating scientific discovery and strengthening patient partnerships, researchers at the University of Toronto's Leslie Dan Faculty of Pharmacy are exploring new opportunities to connect their expertise and amplify the impact of their work.
Read More