Animal-free testing with AI

EU legislation prohibits the use of laboratory animals in the development of cosmetics. But in other cases, animal testing is actually mandatory in the EU. For example, when drug development requires investigating whether chemicals are toxic to humans. "There are all kinds of reasons not to want laboratory animal use," said Marc Teunis, associate lector at the HU University of Applied Sciences Utrecht. "More and more people feel that it is not ethical to use animals for these tests. But there is another important consideration: animals do not react to substances in the same way as humans. Some organs, such as the intestines and the brain, really work differently. So you can't predict very well at all with these tests what a substance does in people."

"There is another problem: we have accumulated an incredible amount of knowledge worldwide about human reactions to chemicals. This accumulation of knowledge is increasing so rapidly that - unlike in the past - a researcher can no longer oversee all the relevant knowledge for his field. This also means that tests with already known results are repeated. Unnecessary testing, in other words. Moreover, because the overall picture is lost, we do not always know what knowledge is still missing to make rapid progress."

"We are working with the most prominent partners in the field of AI and toxicology research, on a tool that discovers patterns in the vast global collection of scientific research," Marc said. "That tool can predict the likelihood of a substance having a toxic reaction on humans. This uses language models, just like ChatGPT. But unlike ChatGPT, this tool relies solely on scientific knowledge. The tool can not only distil what we already know about known substances: it can also compare at the nano level the properties of new substances with properties of substances whose effects are already known. This allows it to predict risks even from substances not yet tested. That prediction then still needs to be verified."

Marc: "The tool can better predict correlations. This also allows it to indicate that there is a risk to a specific organ. And for specific organs, there are much more suitable testing methods than laboratory animals. For example, organoids: organs about 8 mm in size grown in the lab. My colleagues in the lab at the HU grow those by reprogramming ordinary human cells into stem cells, and from those stem cells they can make specific cells again through chemical modifications. A kidney cell, for example. They then grow such a kidney cell into an organoid."

Mirthe Klaassen, Life Sciences student: "I am working on this project because I am fascinated by the huge impact AI is going to have on medical research. Marc developed a customised research project for me that I could do instead of a ready-made minor. Few people worldwide know much about toxicology as well as AI. Marc is one of those people. So working with him on this project is a unique opportunity. Once this tool is ready, not only will we be able to make more accurate predictions, and better understand how people react to substances, but we can also reach important conclusions faster. If we develop drugs faster, we can help more people. It is very motivating to be able to contribute to that already as a student." 

• Research Group Innovative Testing in Life Sciences & Chemistry 
• Biology and Medical Laboratory Research (Life Sciences) Degree Programme (NL)
• National Institute of Public Health and the Environment (RIVM) 
• SURF
• John Hopkins University
• Vrije Universiteit Brussel

The research in this HU Story is made possible in part by:

• ONTOX (with a Horizon subsidy)
• VHP4Safety (with NWO subsidy)