Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Physiologically Informed Deep Learning: A Multi-Scale Framework for Next-Generation PBPK Modeling

Shunqi Liu, Han Qiu, Tong Wang

TL;DR

A unified Scientific Machine Learning (SciML) framework that bridges mechanistic rigor and data-driven flexibility is proposed and experiments show that the framework reduces physiological violation rates from 2.00% to 0.50% under constraints while offering a path to faster simulation.

Abstract

Physiologically Based Pharmacokinetic (PBPK) modeling is a cornerstone of model-informed drug development (MIDD), providing a mechanistic framework to predict drug absorption, distribution, metabolism, and excretion (ADME). Despite its utility, adoption is hindered by high computational costs for large-scale simulations, difficulty in parameter identification for complex biological systems, and uncertainty in interspecies extrapolation. In this work, we propose a unified Scientific Machine Learning (SciML) framework that bridges mechanistic rigor and data-driven flexibility. We introduce three contributions: (1) Foundation PBPK Transformers, which treat pharmacokinetic forecasting as a sequence modeling task; (2) Physiologically Constrained Diffusion Models (PCDM), a generative approach that uses a physics-informed loss to synthesize biologically compliant virtual patient populations; and (3) Neural Allometry, a hybrid architecture combining Graph Neural Networks (GNNs) with Neural ODEs to learn continuous cross-species scaling laws. Experiments on synthetic datasets show that the framework reduces physiological violation rates from 2.00% to 0.50% under constraints while offering a path to faster simulation.

Physiologically Informed Deep Learning: A Multi-Scale Framework for Next-Generation PBPK Modeling

TL;DR

A unified Scientific Machine Learning (SciML) framework that bridges mechanistic rigor and data-driven flexibility is proposed and experiments show that the framework reduces physiological violation rates from 2.00% to 0.50% under constraints while offering a path to faster simulation.

Abstract

Physiologically Based Pharmacokinetic (PBPK) modeling is a cornerstone of model-informed drug development (MIDD), providing a mechanistic framework to predict drug absorption, distribution, metabolism, and excretion (ADME). Despite its utility, adoption is hindered by high computational costs for large-scale simulations, difficulty in parameter identification for complex biological systems, and uncertainty in interspecies extrapolation. In this work, we propose a unified Scientific Machine Learning (SciML) framework that bridges mechanistic rigor and data-driven flexibility. We introduce three contributions: (1) Foundation PBPK Transformers, which treat pharmacokinetic forecasting as a sequence modeling task; (2) Physiologically Constrained Diffusion Models (PCDM), a generative approach that uses a physics-informed loss to synthesize biologically compliant virtual patient populations; and (3) Neural Allometry, a hybrid architecture combining Graph Neural Networks (GNNs) with Neural ODEs to learn continuous cross-species scaling laws. Experiments on synthetic datasets show that the framework reduces physiological violation rates from 2.00% to 0.50% under constraints while offering a path to faster simulation.
Paper Structure (21 sections, 7 equations, 3 figures, 1 table)

This paper contains 21 sections, 7 equations, 3 figures, 1 table.

Table of Contents

  1. Introduction
  2. Related Work
  3. AI in Pharmacokinetics
  4. Scientific Machine Learning (SciML)
  5. Generative Models in Medicine
  6. Methodology
  7. Foundation PBPK Transformers
  8. Architecture
  9. Physiologically Constrained Diffusion Models (PCDM)
  10. Physics-Informed Loss Function
  11. Neural Allometry (Hybrid GNN + Neural ODE)
  12. Graph Neural Network Encoder
  13. Neural ODE Solver
  14. Experimental Setup
  15. Data Generation
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Comparison of Transformer Forecasting vs Ground Truth ODE. The Transformer model (blue) accurately predicts the drug concentration profile from limited initial observations.
  • Figure 2: Ablation Study Results. The introduction of Physics-Informed Loss (blue dots) significantly tightens the distribution compared to the unconstrained model (red dots), reducing physiological violations. The left panel shows the scatter plot of Liver Volume vs. Weight, while the right panel quantifies the violation rates.
  • Figure 3: Cross-Species Prediction. The model successfully predicts Human PK profiles (blue dashed) using only Rat and Dog training data, matching the ground truth (red solid).