Integration of Machine Learning With PBPK and QSAR Modeling Approaches to Facilitate Drug Discovery and Development

This review examines the application of machine learning (ML) in physiologically based pharmacokinetic (PBPK) modeling through improved prediction of input parameters, particularly via quantitative structure-activity relationship (QSAR) models, for absorption, distribution, metabolism, and excretion (ADME) properties across drug development phases. Traditional PBPK models, while mechanistically sound, face limitations in applicability domain when compound-specific physicochemical and biochemical parameters (e.g., partition coefficients, metabolic clearance rates, plasma protein binding) cannot be reliably determined or estimated for novel chemicals. ML-enhanced QSAR models address these input parameter limitations by predicting required parameters directly from chemical structure, thereby extending PBPK applicability to diverse chemical space without altering the mechanistic model framework itself. Recent advances in ML-powered QSAR methodology generate higher-quality predictions of partition coefficients, clearance rates, and binding parameters. When these ML-predicted parameters are integrated as inputs into conventional PBPK frameworks, the resulting simulations often demonstrate significantly improved predictive accuracy compared to PBPK models using conventional regression-based QSAR inputs. This workflow-wherein ML enhances QSAR prediction quality, which subsequently improves PBPK simulation reliability through better input parameters-facilitates critical drug discovery and development decisions, from early candidate selection to clinical translation and safety assessment, potentially reducing pharmacokinetic-related drug candidate failures. However, challenges remain, including limited chemical diversity in QSAR training data sets, model interpretability concerns, and evolving regulatory considerations for ML-derived parameters in PBPK submissions. Future developments point toward seamless integration of ML-enhanced QSAR predictions as validated inputs for mechanistic PBPK models through standardized validation approaches and collaboration among academia, industry, and regulatory bodies.