Machine learning-driven optimization of mRNA-lipid nanoparticle vaccine quality with XGBoost/Bayesian method and ensemble model approaches

To enhance the efficiency of vaccine manufacturing, this study focuses on optimizing the microfluidic conditions and lipid mix ratios of messenger RNA-lipid nanoparticles (mRNA-LNP). Different mRNA-LNP formulations (n = 24) were developed using an I-optimal design, where machine learning tools (XGBoost/Bayesian optimization and self-validated ensemble (SVEM)) were used to optimize the process and predict lipid mix ratio. The investigation included material attributes, their respective ratios, and process attributes. The critical responses like particle size (PS), polydispersity index (PDI), Zeta potential, pKa, heat trend cycle, encapsulation efficiency (EE), recovery ratio, and encapsulated mRNA were evaluated. Overall prediction of SVEM (>97%) was comparably better than that of XGBoost/Bayesian optimization (>94%). Moreover, in actual experimental outcomes, SVEM prediction is close to the actual data as confirmed by the experimental PS (94–96 nm) is close to the predicted one (95–97 nm). The other parameters including PDI and EE were also close to the actual experimental data. [Display omitted] •mRNA-LNP formulations are optimized and outcomes are predicted.•Process conditions are optimized using Bayesian optimization with >94% accuracy.•Prediction of lipid mixture ratio for targeted outcome using SVEM with >97% accuracy.•Predictive model developed with small-size experiments.