The Use of Machine Learning and Artificial Intelligence in Toxicology and Risk Assessment
Tuesday, June 9th, 2020
10:00 AM US ET
Presenter: Timothy E H Allen, Research Scientist, Willis Group, University of Cambridge
Artificial intelligence (AI) and machine learning (ML) algorithms are gaining a lot of attention in toxicology. These algorithms are advantageous as they can identify new patterns in data and make predictions in a way and on a scale that human scientists cannot. What are they and how do they work? How do they learn and help make decisions? And where do they fit into the science of toxicology?
Some of these questions and some of the myths surrounding these research areas will be addressed in this talk. This will include introducing some of the ideas in AI and ML and explaining how some algorithms, including random forests and neural networks, work. This will be followed with a focus on the areas of toxicology in which we are trying to apply these ideas – specifically in the area of predictive toxicology. Computational models have been constructed based on structural alerts, random forests and neural networks to predict pharmacologically important human molecular initiating events, the initial interactions molecules make with biomolecules or biosystems that can lead to adverse outcomes. Attempts are also being made to overcome the disadvantages of these algorithms – particularly how they are seen as “black boxes” with little understanding of their internal working – by combining their predictions and comparing how different chemicals are assessed by the algorithms.
AI and ML approaches undoubtedly have a major role to play in the future of toxicology – but a greater understanding of the algorithms, how they work and why specific predictions are made are areas that need to be considered to see greater adoption of these valuable tools. Approaches such as those presented here allow us to answer some of these questions and can support the use of their powerful predictivity in safety science.