| Mar 20, 2026 |
An AI model that screens ionizable lipids by 3D conformation identified a candidate 14.8 times more efficient than current clinical lipids, enabling spleen-targeted mRNA vaccines.
|
|
(Nanowerk News) A team at the National Center for Nanoscience and Technology (NCNST), part of the Chinese Academy of Sciences, has built an artificial intelligence model that designs ionizable lipids based on their three-dimensional shape rather than their flat chemical formula. The approach overcomes two persistent problems in messenger RNA (mRNA) drug delivery: inefficient uptake by cells and the inability to direct treatments to specific organs. The work was published in Nature Biomedical Engineering (“Artificial intelligence-guided design of LNPs for in vivo targeted mRNA delivery via analysis of the spatial conformation of ionizable lipids”).
|
Key Findings
- An AI-identified lipid called P1 delivered mRNA 14.8 times more efficiently than the clinically approved lipid ALC-0315.
- Lipid P1’s unique three-dimensional conformation allows it to bind immunoglobulin M (IgM) in blood, enabling selective delivery to the spleen.
- An mRNA cancer vaccine built on this spleen-targeted platform triggered strong immune responses and shrank established tumors in a mouse melanoma model.
|
|
Conventional lipid nanoparticle (LNP) design relies almost entirely on the two-dimensional chemical structures of ionizable lipids, the molecules that help package and protect mRNA inside nanoparticles. This approach overlooks the way those lipids actually fold and flex under physiological conditions, missing spatial features that influence how well the particles interact with biological systems.
|
|
The NCNST team made three-dimensional spatial conformation a central input to its AI framework. The researchers first assembled a library of ionizable lipids and ran molecular dynamics simulations to capture each molecule’s dynamic shape in a realistic environment. They then converted the resulting 3D conformational data into two-dimensional density maps, a format suitable for training an AI screening model.
|
 |
| Schemetic illustration for the development of an AI model for LNP optimization. (Image: SU Linjia et al.) (click on image to enlarge)
|
|
The trained model flagged a lipid designated P1 as a top candidate. In head-to-head tests, P1 outperformed ALC-0315, the ionizable lipid used in approved COVID-19 mRNA vaccines, by a factor of 14.8 in mRNA delivery efficiency. The performance gain traced back to P1’s distinctive spatial geometry, which enables selective binding to immunoglobulin M (IgM) circulating in the bloodstream. That interaction steers the nanoparticle toward the spleen.
|
|
To demonstrate the practical value of spleen-targeted delivery, the team formulated an mRNA cancer vaccine using P1-based LNPs and tested it in mice bearing melanoma tumors. The vaccine activated robust T-cell responses and stimulated high levels of antigen-specific antibodies, achieving simultaneous humoral and cellular immunity. Treated animals showed marked tumor regression and developed long-term immune protection against the cancer.
|
|
The study also uncovers the molecular mechanism by which ionizable lipids facilitate lysosomal escape, the step in which mRNA cargo exits the cell’s recycling compartments and reaches the cytoplasm where it can be translated into protein. Understanding this bottleneck at the molecular level could inform the design of more efficient delivery vehicles for a range of mRNA-based therapies.
|
|
By linking a lipid’s spatial conformation directly to its biological performance, the AI-driven approach provides a more systematic alternative to empirical screening of two-dimensional chemical diagrams. The same framework could, in principle, be applied to design LNPs tuned for other target organs, though the authors do not speculate on specific applications beyond spleen-directed cancer immunotherapy.
|
