Ionizable lipids are a pivotal component for lipid nanoparticles (LNPs) to optimize mRNA expression to fulfill the broad demands of various mRNA therapeutics. Traditionally, the screening of ionizable lipids needs extensive synthetic labor and stringent in vivo efficacy evaluation, which was often time-consuming, costly, and characterized by a lower success rate. Hence, we pioneered a machine-learning framework named Lipid with Artificial Intelligence (LipidAI) to evaluate the novel ionizable lipids rapidly. In this framework, the Methyl Tail Augmentation (MTA) strategy was first developed to triple the data by precisely adjusting the methyl groups on lipid tail chains. This ground-breaking approach compensated for data paucity in ionizable lipids libraries from the previous research and boosts model accuracy. Subsequently, the Ensemble Stacking Learning (ESL) algorithm was exploited to integrate multiple learning algorithms to surpass the predictive accuracy of a single algorithm used in former studies. Finally, we found that the predicted results of LipidAI were highly consistent with the actual data according to the in vivo expression of Luc-mRNA. Overall, this study highlights the remarkable potential of LipidAI in the rapid screening of ionizable lipids, adeptly avoiding the inherent drawbacks of traditional ionizable lipids development and thereby boosting the progress of LNP-based mRNA nano-drugs.

Ionizable lipids  /  Lipid nanoparticles  /  Machine learning  /  mRNA expression  /  In vivo  /  Expression prediction  /  Delivery  /  mRNA vaccines