Machine Learning\nElucidates Design Features of Plasmid\nDeoxyribonucleic Acid Lipid Nanoparticles for Cell Type-Preferential\nTransfection

To broaden the accessibility of cell and gene therapies,\nit is\nessential to develop and optimize nonviral, cell type-preferential\ngene carriers such as lipid nanoparticles (LNPs). While high-throughput\nscreening (HTS) approaches have proven effective in accelerating LNP\ndiscovery, they are often costly, labor-intensive, and do not consistently\nyield actionable design rules that direct screening efforts toward\nthe most relevant chemical and formulation parameters. In this study,\nwe employed a machine learning (ML) workflow, utilizing well-curated\nplasmid DNA LNP transfection data sets across six cell types, to extract\ncompositional and chemical insights from HTS studies. Our approach\nachieved prediction errors averaging between 5 and 10%, depending\non the cell type. By applying SHapley Additive exPlanations to our\nML models, we uncovered key composition–function relationships\nthat govern cell type-preferential LNP transfection efficiency. Notably,\nwe identified consistent LNP composition parameters that enhance <i>in vitro</i> transfection efficiency across diverse cell types,\nincluding a helper lipid molar percentage of charged lipids between\n9 and 50% and the inclusion of cationic/zwitterionic helper lipids.\nAdditionally, several parameters were found to modulate cell type-preferentiality,\nsuch as the total molar percentage of ionizable and helper lipids,\nN/P ratio, PEGylated lipid molar percentage of uncharged lipids, and\nhydrophobicity of the helper lipid. This study leverages HTS of compositionally\ndiverse LNP libraries combined with ML analysis to elucidate the interactions\nbetween lipid components in LNP formulations, providing insights that\ncontribute to the design of LNP compositions tailored for cell type-preferential\ntransfection.