Machine Learning Models to Enhance Predictability of IVIVC in Early Drug Development

Abstract

In vitro–in vivo correlation (IVIVC) serves as a critical predictive tool in pharmaceutical research, linking in vitro drug release profiles to in vivo pharmacokinetic behavior. Despite its regulatory and developmental importance, traditional IVIVC models often struggle to capture the complex, nonlinear relationships inherent in drug absorption and metabolism, particularly for novel and extended-release formulations. Recent advances in machine learning (ML) offer promising solutions to enhance the predictability and robustness of IVIVC, leveraging large, multidimensional datasets and sophisticated pattern recognition capabilities. This article explores the integration of various ML models, including Random Forest, Support Vector Regression, Gradient Boosting, and Neural Networks, into IVIVC modeling frameworks during early drug development. We present a comparative analysis of model performances against conventional statistical methods, using real and simulated datasets, with a focus on predictive accuracy, model validation, and regulatory considerations. The findings demonstrate that ML-based approaches substantially improve IVIVC predictability, offering new avenues for optimizing formulation strategies and reducing the reliance on extensive in vivo testing. The study underscores the potential of artificial intelligence as a transformative tool in pharmaceutical sciences, advocating for its broader adoption in preclinical and early clinical research phases.