High-throughput microfluidics and machine learning-assisted screening of lipid nanoparticle formulations for siRNA delivery.

Lipid nanoparticles (LNPs) are among the most advanced non-viral carriers for RNA delivery; however, the optimization of LNP formulations remains challenging due to the vast and multi-dimensional formulation space. Here, we established a high-throughput microfluidic and machine learning-assisted workflow for stepwise LNP formulation prioritization for siRNA delivery. A library of 864 candidate formulations was generated, and 203 randomly selected empty LNPs were experimentally characterized. Machine learning models were then used to predict particle size and PDI across the full formulation library, supporting the selection of 25 candidates for siRNA loading and biological validation. The results showed that the size have positive relationship with lipids concentration, and the PEG ratio have negative relationship with size, and positive with PDI. After siRNA loading, all selected formulations-maintained particle sizes below 150 nm. In vitro gene silencing results showed that SM-102- and CKK-E12-based formulations achieved stronger S100P knockdown, with relative S100P expression levels reduced to 0.0744 and 0.0611, respectively. In vivo biodistribution analysis further showed that CKK-E12-based LNPs exhibited higher lung signals compared with the other three ionizable lipid groups. These results indicate that favorable physicochemical properties are necessary but not sufficient to ensure optimal biological performance, highlighting the importance of biological validation after model-assisted screening. Overall, this workflow provides a practical strategy for integrating high-throughput formulation preparation, model-assisted prioritization, and biological validation for siRNA-LNP development.