A multiobjective AI model for LNP engineering enhances tissue-selective mRNA delivery

Lipid nanoparticle (LNP) delivery of RNA therapeutics is constrained by poor tissue selectivity and off-target toxicity. Most high-throughput screening approaches have focused on single-target efficacy while overlooking off-target uptake. Here we report multiobjective LNP engineering with artificial intelligence (MOLEA), a system that integrates high-dimensional lipid representations, cell-type-resolved transfection data and multitask optimization to design ionizable lipids with both high potency and biological selectivity. MOLEA learns structure-function relationships across diverse cellular contexts to identify lipids that preferentially deliver mRNA to target tissue while minimizing hepatocyte transfection. Applying MOLEA to cartilage, we developed K9 LNPs, which achieve >90% transfection efficiency in mouse joint chondrocytes and a 13.5-fold increase in knee-to-liver selectivity compared to the clinical benchmark SM-102. We demonstrate chondrocyte-specific Mmp13 editing in osteoarthritis mouse models, leading to sustained cartilage protection and suppression of disease-associated immune and matrix remodeling. Our findings demonstrate how artificial-intelligence-guided multiobjective optimization can enable precision RNA delivery with potential applications to other tissues.