Prediction of lipid nanoparticles for mRNA vaccines by the machine learning algorithm

Area of Science:

• Biotechnology and Pharmaceutical Sciences
• Computational Chemistry and Molecular Modeling
• Vaccine Development and Delivery Systems

Background:

• Lipid nanoparticles (LNPs) are crucial for mRNA vaccine delivery.
• Traditional LNP optimization is time-consuming and costly due to extensive experimental screening of ionizable lipids.
• Accelerating LNP development is essential for rapid vaccine production.

Purpose of the Study:

• To apply computational methods, specifically machine learning, to accelerate the optimization of LNP formulations for mRNA vaccines.
• To develop a predictive model for LNP formulations based on experimental data.
• To identify critical ionizable lipid substructures influencing LNP performance.

Main Methods:

• Collected 325 data samples of mRNA vaccine LNP formulations and their corresponding IgG titers.
• Developed a prediction model using the lightGBM machine learning algorithm, achieving R² > 0.87.
• Validated model predictions through animal experiments and investigated molecular mechanisms using molecular dynamic modeling.

Main Results:

• The lightGBM model demonstrated high performance in predicting LNP formulation efficiency.
• Identified critical ionizable lipid substructures, consistent with existing literature.
• Experimental results confirmed model predictions, showing higher efficiency for LNP with DLin-MC3-DMA (MC3) compared to SM-102 at a 6:1 N/P ratio.

Conclusions:

• A novel machine learning predictive model for LNP-based mRNA vaccines was successfully developed and experimentally validated.
• The integrated approach of machine learning and molecular modeling provides a powerful tool for LNP formulation.
• This predictive model can significantly accelerate the virtual screening of LNP formulations for future mRNA vaccine development.