AI could reduce the need for animal testing

New software technology could be more effective at detecting whether pollutants will be harmful to aquatic life than testing directly on animals, according to a new paper published in Environmental Science & Technology.

The project team composed by researchers from King’s College London, the Universities of Northumbria and Suffolk, the Francis Crick Institute and AstraZeneca say indeed that machine learning – a field of artificial intelligence – could be used to predict the impact of pollutants on aquatic wildlife.

Machine learning uses a series of algorithms to model complex data relationships, which can be applied to predict and show how chemical substances accumulate in fish and invertebrates. The researchers say that this predictive modelling is so effective that it could reduce the need for direct tests in animals to measure pollution effects on them. If more people have the skills needed to carry out this type of testing, it could benefit environmental testing across the board.

Pollution from contaminants are a huge concern across the globe, not only for the environment but also for public health. Governments are beginning to address this problem by only allowing chemicals to reach the market once they have been fully tested and the potential impact on living organisms in our waters has been assessed.

Dr Thomas Miller, from the research team at King’s said: “Until now, the only viable way to really understand the impact of chemicals in the aquatic environment was to study live animals. But we’ve shown that there are alternative ways to do this.

“As part of an ongoing collaboration between academic and industry bodies, we have shown that machine learning, or artificial intelligence, can be used to model and predict chemical and biological processes that occur when animals are exposed to certain chemicals.”

Dr Matteo Gallidabino, Senior Lecturer in Forensic Science at Northumbria University, is one of the researchers involved in the project. He specialises in statistical modelling and machine learning techniques for forensic applications and helped the group to develop and test the mathematical models used in the software.

He explained: “Machine learning allows the rapid modelling of complex relationships in large datasets that are otherwise extremely challenging to manage and analyse. It is an incredibly flexible tool. Its use in environmental sciences, however, is relatively new and still under-investigated.”

Following on from this, in their Environmental Science & Technology paper, the team have called for these methods to be applied much more widely in the field of environmental toxicology and are urging the international community to follow their lead.

They say that there is a need for greater collaboration and more training for scientists in this field to learn these new skills and they call on governments and regulators to rise to the challenge and embrace machine learning.

Dr Stewart Owen from the pharmaceutical company, AstraZeneca, added: “Machine learning is increasingly being used to innovate and solve complex problems across all industries, from financial services to healthcare. However, we need to accelerate our understanding and application of these tools to better understand and respond to the environmental challenges we face as a society. In doing so, we can begin to push the boundaries of science today, to support us in meeting the greatest challenges of tomorrow more effectively.”