Bojana Stanić1*, Nemanja Milošević2, Nataša Sukur2 and Nebojša Andrić1
1 Department of Biology and Ecology, Faculty of Sciences, University of Novi Sad, Novi Sad, Serbia
2 Department of Mathematics and Informatics, Faculty of Sciences, University of Novi Sad, Novi Sad, Serbia
bojana.stanic [at] dbe.uns.ac.rs
Abstract
Over the past few decades, toxicology has made a sharp turn from an observational science focused on analyzing chemical-induced endpoints to a data-rich discipline. The amount of new data stemming from the literature, high-throughput screening (HTS) assays, omics, and other technologies is rapidly accumulating, creating a fruitful ground for the application of artificial intelligence (AI). Through machine learning (ML), deep learning, large language models, and natural language processing techniques, AI can effectively navigate this complex data landscape, deciphering patterns, elucidating toxicity mechanisms, and enhancing risk prediction.
Here, some of the applications of ML in toxicology will be presented. The ML models were used to unravel the intricate mechanisms underlying chemical-induced female infertility. The adverse outcome pathway (AOP), a theoretical concept describing biological events leading to adverse effects, was used as a backbone in developing the ML models for female infertility. Utilizing eighteen HTS bioassays, these models tracked key biological processes outlined in AOP7 – receptor binding as a molecular initiating event, and gene expression and steroid production as key events – leading to adverse outcomes. These ML models efficiently simulated and predicted perturbations in each event within toxicity pathways for novel chemicals, revealing a group of chemicals that can affect all events in the AOP, thus forming a linear molecular pathway that can lead to female reproductive disorders. The ML models were also used to assess the potency of novel chemicals in binding to the progesterone receptor and to discriminate between agonists and antagonists of this receptor.
These examples underscore the vast potential of AI in toxicology research, offering a multitude of avenues for exploration. AI has the potential to transform toxicology into a more predictive, mechanism-based, and evidence-integrated scientific discipline to better safeguard human and environmental health from chemical hazards.
Keywords: machine learning, environmental chemicals, computational toxicology
Acknowledgement: This work was supported by the Provincial Secretariat for Higher Education and Scientific Research of the Autonomous Province of Vojvodina (grant number 142-451-3533).