Machine learning predictions of drug release from isocyanate-derived aerogels

This work utilized machine learning (ML) algorithms to predict and validate the in vitro drug release kinetics of a short worm-like nanostructured isocyanate-derived aerogel: the first time ML has been employed to study the drug delivery properties of this important class of materials. The algorithms were first trained with sixteen datasets, each containing eight release data points, before using them to predict the release profiles of the unknown. The predicted data was validated via the random sampling and cross-validation techniques. In both instances, the established models were used to predict the release kinetics of four aerogel nanostructures with known experimental release profiles. A good correlation between the experimental and predicted release profiles was observed, with gradient boosting being the best-performing algorithm (R² > 0.9). Furthermore, the ranking of the importance of each input feature for drug release from the aerogels aligns with previous studies, validating the rationale behind the modeling. Morphology, quantified by the K-index (contact angle/porosity), and the macropore-to-mesopore ratios were found to be the most influential factors, after time, in determining drug release profiles. The findings from this study suggest that ML can serve as a valuable tool for predicting the drug release kinetics of aerogels, thereby saving time and cost involved in conducting laborious drug delivery experiments. We envisage that this study will provide a foundation for future related computational works and reduce the trial-and-error experimental approach to solving scientific problems.