Evaluating the impact of lyophilization process parameters on mRNA encapsulated lipid nanoparticles using machine learning

Abstract

mRNA-encapsulated lipid nanoparticles (mRNA-LNPs) play a pivotal role in the mitigation of COVID-19. However, these vaccines are frequently accompanied by critical challenges such as a relatively short shelf life and the necessity for frozen storage. The development of lyophilized formulations has garnered significant interest in addressing these limitations, in turn enhancing the stability of mRNA-LNPs. Although lyophilization is a commonly employed manufacturing technique, systematic investigations of the relationship between its potential critical process parameters (pCPPs) and critical quality attributes (CQAs) of mRNA-LNPs remain scarce. In this study, we evaluated the effects of selected pCPPs (freezing rate, annealing temperature, and primary drying temperature) on various CQAs, including particle size, polydispersity index, mRNA and lipid content, encapsulation efficiency, RNA integrity, and in vitro activity. Among these attributes, only particle size exhibited a statistically significant variation across the investigated process parameters. Cryo-electron microscopy revealed that this observation was a consequence of bleb-like structures induced by lyophilization. Feature importance analysis based on a gradient boosting model identified freezing rate and annealing temperature as the most influential variables. The model further predicted that particle size was particularly sensitive to a freezing rate of approximately −0.3 °C/min, underscoring the necessity of precise control of freezing conditions to maintain product stability. These findings demonstrate that machine learning-based modelling can facilitate the quantitative evaluation of the effects of pCPPs on CQAs using a minimal number of experiments, thereby enabling a more comprehensive understanding of the mRNA-LNP manufacturing process and supporting the rational design of robust production strategies.