Lipid nanoparticles (LNPs) enable RNA medicines, yet ionizable-lipid discovery remains largely empirical, particularly when in vitro activity must be translated into functional hepatic delivery in vivo. Here we establish an interpretable discovery framework that combines modular degradable lipid chemistry, high-throughput phenotyping, machine learning, virtual library expansion, and independent in vivo validation to define a learnable ionizable-lipid design space. Starting from a 520-member library, we train an XGBoost model on firefly luciferase (Fluc) expression in HeLa cells and use physicochemical descriptors and molecular fingerprints to prospectively prioritize candidates with higher predicted in vitro tr... More
Lipid nanoparticles (LNPs) enable RNA medicines, yet ionizable-lipid discovery remains largely empirical, particularly when in vitro activity must be translated into functional hepatic delivery in vivo. Here we establish an interpretable discovery framework that combines modular degradable lipid chemistry, high-throughput phenotyping, machine learning, virtual library expansion, and independent in vivo validation to define a learnable ionizable-lipid design space. Starting from a 520-member library, we train an XGBoost model on firefly luciferase (Fluc) expression in HeLa cells and use physicochemical descriptors and molecular fingerprints to prospectively prioritize candidates with higher predicted in vitro transfection activity across a 10,920-member virtual space. In vivo phenotyping subsequently identifies PA97MD101 as a lead ionizable lipid. In mice, PA97MD101 enhances hepatic mRNA expression and Cre-dependent recombination and supports Cas9/sgTtr-mediated TTR suppression. Its performance is associated with pH-dependent membrane activity and reduced lysosomal colocalization. These findings support further evaluation of PA97MD101 for functional hepatic RNA delivery.
