Integrating artificial intelligence and physiologically based pharmacokinetic modeling to predict in vitro and in vivo fate of amorphous solid dispersions.

Amorphous solid dispersions (ASDs) have emerged as a pivotal strategy in enhancing the dissolution profiles of poorly water-soluble drugs. Although the apparent dissolution rate (both molecular and colloidal drugs) within ASDs has been determined in past experiments, only molecular drugs in the dissolution can be effectively absorbed in vivo. This study aims to develop a prediction model for the in vitro (molecular dissolution) and in vivo (systemic absorption) fate of ASD formulations by integrating artificial intelligence and physiologically based pharmacokinetic (AI/PBPK) modeling. 238 ASD formulations, including 22 small-molecule drugs and 1693 drug molecular dissolution time points, were collected for molecular dissolution prediction. The Tabular Prior Data Fitted Networks (TabPFN) model exhibited robust predictive performance across six machine learning frameworks, achieving a low root mean squared error (RMSE) of 0.116 ± 0.020 and a high determination coefficient (R) of 0.905 ± 0.028. Then, the release kinetic model bridges the in vitro and in vivo dissolution of ASDs. The PBPK models of three on-market ASDs (Noxafil®, Prograf®, Gris-PEG®) were developed and validated with the absolute average fold error (AAFE) of less than 2. The simulated systemic absorption exhibits high similarity to the results obtained using the classical deconvolution method, with a similarity factor (f) exceeding 50. An additional 37 marketed ASD formulations were evaluated using our framework, with most of the Cmax and AUC predictions falling within two-fold error ranges, demonstrating the high accuracy and robustness of our ASD AI/PBPK model. This research not only provides a practical and rapid computational tool for the rational design and evaluation of ASDs but also significantly advances our understanding of the interplay between molecular dissolution and systemic absorption, thus contributing to the optimization of drug delivery systems.