生理学信息深度学习：下一代PBPK建模的多尺度框架

Physiologically Based Pharmacokinetic (PBPK) modeling serves as a cornerstone of model-informed drug development (MIDD), offering a mechanistic framework for predicting drug absorption, distribution, metabolism, and excretion (ADME). Despite its advantages, the widespread adoption of PBPK modeling faces challenges such as high computational costs for large-scale simulations, difficulties in parameter identification for complex biological systems, and uncertainties in interspecies extrapolation. A unified Scientific Machine Learning (SciML) framework is proposed to bridge the gap between mechanistic rigor and data-driven flexibility. Key contributions include: (1) Foundation PBPK Transformers that frame pharmacokinetic forecasting as a sequence modeling task; (2) Physiologically Constrained Deep Learning models that enhance physiological consistency; and (3) Multi-scale modeling strategies that improve model applicability across different biological levels.