Summary

Researchers at Empa have developed an AI-powered physiologically based pharmacokinetic model (PBPK model) that simulates the distribution of nanomaterials in the organism. The system uses machine learning to predict how different nanoparticles are distributed in the body based on their properties (size, coating, surface charge). The model was trained on the basis of 18 animal studies and enables promising nanoparticle candidates to be tested virtually before costly and ethically problematic animal experiments are conducted.

People

Topics

  • Artificial Intelligence in Medicine
  • Nanotechnology and Nanomedicine
  • Reduction of Animal Testing
  • Safe and Sustainable by Design (SSbD)
  • Pharmacokinetic Modeling
  • Blood-Brain Barrier and Brain Tumor Treatment

Clarus Lead

Swiss animal protection law requires researchers to minimize animal experiments – a regulatory pressure that is accelerating the development of alternatives to animal models. Wu's AI mouse addresses this need directly: it enables the safety of new nanomaterials to be assessed before manufacturing and thus shortens the development process. This corresponds to the growing trend of combining technological innovation with ethical and regulatory requirements – an approach that has signaling effects for other industries beyond medicine.

Detailed Summary

Nanoparticles are considered promising carriers for active ingredients because their extreme smallness (approximately 500 times smaller than a human hair) allows them to cross the body's protective barriers such as the blood-brain barrier. This opens up new possibilities for treating brain tumors, where conventional chemotherapies fail at this barrier. However, the distribution of nanoparticles varies considerably depending on shape, material composition, and size – a circumstance that previously could only be investigated through elaborate, expensive, and ethically questionable animal studies.

Jimeng Wu integrated a multivariate linear regression model into her AI model, which enables the system to automatically adjust its parameters to the measurable properties of new nanoparticles. This differs fundamentally from traditional PBPK models, which must be recalibrated for each individual substance. The system then calculates the distribution in the simulated mouse body – currently for liver, kidneys, lungs, and spleen. The practical benefit is that researchers and industry partners can virtually test which types of particles are suitable for a specific therapeutic task before costly clinical studies are initiated.

However, Peter Wick points out a critical limitation: the training dataset currently comprises only 18 peer-reviewed publications with sufficient data quality. Many studies inadequately describe the properties of their nanoparticles, which limits the model's reliability. Wu's future work aims to train the model with additional data and transfer it to human PBPK models – a step that would also enable the investigation of sensitive target organs such as the brain.

Key Statements

  • AI-powered PBPK model enables virtual screening of nanoparticles without animal testing
  • Machine learning automatically adjusts model parameters to particle properties
  • Current training dataset is small (18 studies); validation and expansion necessary
  • Goal: Shortening development cycles for nanomedicines and reduction of animal testing
  • Bridge strategy for transfer to human models in development

Critical Questions

  1. Data Quality: How is it ensured that the 18 training studies are representative of the diversity of nanoparticle properties, and what error rates are realistic for predictions based on this small dataset?

  2. Validation: Have the AI model's predictions already been validated against new experimental animal studies to demonstrate reliability?

  3. Transferability: What biological differences between mouse and human could affect the model's predictive power when transferring to human PBPK models as planned?

  4. Conflicts of Interest: Which industry partners are involved in the development, and could commercial incentives influence the interpretation of model results?

  5. Regulatory Acceptance: Will regulatory authorities (EMA, FDA) accept predictions from this model as a replacement for preclinical animal studies, or will a hybrid approach remain necessary?

  6. Scalability: How resource-intensive is it to expand the model for new nanoparticle classes or other organisms?

Bibliography

Primary Source: Empa Press Release – AI Mouse Model Advances Medical Research (28.05.2026) https://www.empa.ch/web/s604/ki-mausmodell-bringt-medizinische-forschung-voran

Scientific Publication: Wu, J., Wick, P., Nowack, B. (2025): Data-Driven Prediction of Nanoparticle Biodistribution from Physicochemical Descriptors. ACS Nano, 19(29). https://doi.org/10.1021/acsnano.5c03040

Verification Status: ✓ 28.05.2026

This text was created with the support of an AI model. Editorial Responsibility: clarus.news | Fact-Check: 28.05.2026