Predicting chemical bioaccumulation in wildlife using machine learning

Environmental stress in the form of chemical pollution is now being recognised as an environmental emergency due to the potential impacts on public and animal health. This emergency has been recognised by the United Nations Environment Programme and is part of a triple threat, alongside climate change and biodiversity loss.

The main challenge here is that many knowledge gaps concerning chemicals in the environment remain despite decades of research.

One of these gaps is related to the potential accumulation of chemicals in exposed organisms (i.e. bioaccumulation). We know that chemicals are taken up by wildlife in the environment and evidence has demonstrated that bioaccumulation can vary depending on several factors including the type of chemical and species. Moreover, experimentally determining bioaccumulation has significant costs and requires high animal usage that opposes current policy aimed at reducing animal testing.

To overcome the issues mentioned above, this project will develop novel machine learning approaches that will enable us to better understand how chemicals bioaccumulate and then predict the bioaccumulation potential for; (a) chemicals across different species and (b) chemicals that have not been previously studied.

This predictive approach represents one of the only feasible ways to assess chemical risk in the environment due to the number of chemicals available in the global market and could significantly reduce the need for animal testing during environmental risk assessments as required by national and international policy.

By developing and validating these machine learning tools to reliably predict bioaccumulation we will enable better protection of the environment from the impact of chemical pollution.

Publications

Thomas H. Miller, Matteo D. Gallidabino, James I. MacRae, Christer Hogstrand, Nicolas R. Bury, Leon P. Barron, Jason R. Snape, and Stewart F. Owen. (2018). Machine Learning for Environmental Toxicology: A Call for Integration and Innovation. Environ. Sci. Technol. 52(22), 12953–12955. 

Thomas H. Miller, Matteo D. Gallidabino, James I. MacRae, Stewart F. Owen, Nicolas R. Bury and Leon P. Barron. (2019). Prediction of bioconcentration factors in fish and invertebrates using machine learning. Science of The Total Environment. 648, 80-89. 

Meet the Principal Investigator(s) for the project

Dr Thomas Miller - As an interdisciplinary scientist with a background in biology and analytical chemistry, my research interests are focussed on the impact of chemicals in the environment and the interaction this chemical stress has with other environmental stressors. My expertise lies in small molecule mass spectrometry to determine chemicals found in the environment (especially in wildlife) and to determine biomarkers and pathways associated with adverse effects in exposed organisms. I am also interested in the integration of artificial intelligence within environmental toxicology to support and solve different environmental challenges.  From the start of my PhD at King's College London my research was originally focussed on the uptake, biotransofrmation and elimination of pharmaceuticals in a freshwater invertebrate (Gammarus pulex) commonly found in UK rivers. I developed and validated machine learning models to predict these proccesses to support and potentially replace bioaccumulation testing during environmental risk assessments. I then moved into a postdoctoral position where I focussed on understanding the impact of pharmaceuticals by assessing behavioural disruption in these organisms. I developed and applied metabolomic workflows to gain a mechanistic understanding of animal behaviour and to link cause-effect relationships for different drug exposures. Here at Brunel, I will be working in three main areas concerned with chemical pollution. First is concerned with the determination of chemicals (and mixtures) using exposomics to characterise the chemical space in the environment, with a focus on internalised residues in animals. Second, improving mechanistic understanding of cause-effect relationships using metabolomics and lipidomics to determine biochemical changes that are phenotypically anchored. Finally, development and application of AI to support envrionmental risk assessment, replace animal testing and improve interpretation of complex datasets to better understand animal health. 

Related Research Group(s)

Pollution Research and Policy - Predictive approaches in toxicology, including combined chemical exposures and development of new frameworks for non-animal approaches for predicting toxicity; Endocrine disruptor research with an emphasis on mechanisms of disease and test method development; Pollution monitoring, clean-up technologies and chemical analytics.

Project last modified 16/11/2023