In Vitro liposome release profile prediction using explainable machine learning approaches

Formulation features and test settings influence liposomal in vitro release (IVR) profiles, yet it is challenging to examine these multivariable associations across varied literature data. We created an explainable computational workflow in this proof-of-concept study for classifying liposomal release phenotypes and identify formulation/assay features linked to slow and fast release. Benchmarking kinetic models, simulating Weibull-parameterized release curves on a shared 0–168 h grid, clustering profiles using PCA and k-means, and training supervised classifiers on formulation and IVR descriptors were all done using a publicly available Accelerated IVR dataset. Slow, moderate, and fast kinetic phenotypes were found using PCA-k-means; 98.4% of the variance was explained by the first two principal components. XGBoost demonstrated the best cross-validated performance across evaluated models for the extreme slow-versus-fast subgroup (n = 59); however, class-wise recall and precision indicate preliminary rather than conclusive prediction performance. The most informative descriptors identified by feature selection and SHAP interpretation were media pH, drug loading, weighted lipid transition temperature, and media temperature. These results indicated that risk-based IVR technique development and hypothesis creation for liposomal formulations can be supported by explainable ML; nevertheless, prior to translational or regulatory usage, prospective validation on independently generated datasets is required.