Levonorgestrel loaded biodegradable microparticles for injectable contraception: Preparation, characterization and modelling of drug release

The integration of mechanistic modeling and machine learning facilitates the understanding and engineering of drug release from controlled release systems. Here, we present hybrid models to predict the effect of drug loading on levonorgestrel release from spray-dried poly(L-lactic acid) microparticles. We developed three Monte Carlo methods that differ in the consideration of polymer's degradability and crystallinity, to simulate drug release from the matrices using the Python programming language. To build each method, we utilized data from the characterization of the particles, such as the actual drug content (ranges from 6% to 52%), size (Dv(50) similar to 5 mu m), and polymer crystallinity (ranges from 0% to 15%). We trained each method using drug release data from particles of 4 batches and derived appropriate machine learning models through regression analysis. Results indicate the contribution of drug diffusion and polymer degradation to drug release for particles of lower drug content (<20 %w/w). At higher drug loadings, particles encountered a combination of burst and diffusional release. We validated the predictive powers of the machine learning models by testing them against experimental data. This paper specifically highlights the power of hybrid modeling to engineer drug release for long-term contraception.

Automated summary

The integration of mechanistic modeling and machine learning can predict the drug release from controlled systems. In this study, hybrid models were developed to predict the effect of drug loading on levonorgestrel release from poly(L-lactic acid) microparticles. The Monte Carlo methods were used to simulate drug release, and machine learning models were derived through regression analysis. The results showed the contribution of drug diffusion and polymer degradation to drug release, and the different release mechanisms under different drug loadings.