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

Area of Science:

• Biotechnology
• Pharmaceutical Sciences
• Materials Science

Background:

• Efficient vaccine manufacturing relies on precise control over nanoparticle formulation.
• Messenger RNA-lipid nanoparticles (mRNA-LNP) are crucial for vaccine delivery, necessitating process optimization.
• Current methods for optimizing mRNA-LNP production can be time-consuming and resource-intensive.

Purpose of the Study:

• To optimize microfluidic conditions and lipid mix ratios for enhanced mRNA-LNP production efficiency.
• To evaluate and compare machine learning models for predicting optimal mRNA-LNP formulations.
• To identify critical material and process attributes influencing mRNA-LNP characteristics.

Main Methods:

• Developed 24 different mRNA-LNP formulations using an I-optimal design.
• Employed machine learning tools, including XGBoost/Bayesian optimization and self-validated ensemble (SVEM), for process optimization and lipid mix ratio prediction.
• Assessed critical responses such as particle size (PS), polydispersity index (PDI), Zeta potential, encapsulation efficiency (EE), and recovery ratio.

Main Results:

• The SVEM model demonstrated superior prediction accuracy (>97%) compared to XGBoost/Bayesian optimization (>94%).
• Experimental validation confirmed SVEM's predictions, with actual particle sizes closely matching predicted values (e.g., 94-96 nm vs. 95-97 nm).
• Key parameters like PDI and EE also showed close agreement between SVEM predictions and experimental outcomes.

Conclusions:

• Machine learning, particularly the SVEM model, offers a highly accurate approach for optimizing mRNA-LNP production.
• Optimized microfluidic conditions and lipid ratios can significantly improve vaccine manufacturing efficiency.
• This data-driven strategy facilitates the development of high-quality mRNA-LNP formulations for vaccine applications.