Optimising the production of PLGA nanoparticles by combining design of experiment and machine learning

Poly(lactic-co-glycolic acid) (PLGA) nanoparticles are widely used in drug delivery for their safety and ability to provide controlled release. However, producing nanoparticles with the right characteristics, such as size and surface charge, requires careful optimisation, which can be both time-consuming and expensive.

This study compared traditional Design of Experiments (DOE) with machine learning (ML) approaches to optimise PLGA nanoparticle formulations produced by nanoprecipitation. Various ML models were evaluated, with Extreme Gradient Boosting (XGBoost) delivering the most accurate predictions for nanoparticle size. Overall, ML outperformed DOE in predicting particle size and identifying the most influential formulation parameters, even when using relatively small datasets. Both approaches struggled to predict zeta potential reliably, but ML provided greater insight into parameter impact. These findings demonstrate the value of combining ML with DOE, particularly in data-limited environments. This integrated approach offers a more efficient path to nanoparticle optimisation, helping streamline experimental design and accelerating development in both research and industrial contexts.

Explore more here.