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.”
