SaferWorldbyDesign: Integrating AI into Chemical Safety Assessment: Opportunities, Challenges and the Path Forward

The long-term investment in new approach methodologies (NAMs) has resulted in an emerging consensus of how to use information from in silico, in vitro and in vivo sources to assess the safety of chemicals. In the last two decades much of the resources have been dedicated to the development of in vitro methods. These investments were successful and consequently a great number of in vitro methods are now available. These methods, even when used in combination are, not yet accepted to replace regulatory repeated dose testing. They are, however, key elements for early screening and avoiding the development of sustainable new chemistry. To indicate this, I have used the expression green chemistry and green (in vitro) toxicology go hand in hand. The challenge now is to select the appropriate in vitro methods, because broad in vitro screening has become too expensive (both in terms of time and financially). The solution to this challenge could be the implementation of in silico and artificial intelligence methodologies. 

The rapid development of machine learning and AI in the last two years promises new unrivaled opportunities as well as significant challenges. To assess the state of the art of the application of AI in chemical safety assessment and to develop guidance on how to make best use of data, ECETOC (European Centre for Ecotoxicology and TOxicology of Chemicals, Brussels) organized a workshop to leverage the collective expertise of scientists from academia, regulators and industry. The major recommendations from this workshop are associated with the data and building trust in AI for its use in a regulatory context.  

Data availability can be increased by reducing barriers for data sharing or even providing incentives for doing so. This may be achieved the “honest broker” concept, which has been used for data sharing within the pharmaceutical and agricultural industries. It is essential that data are provided in a machine-readable format and with sufficient meta data. Data quality can be addressed by a scoring system and automated if sufficient meta data are available. FAIR criteria and the storage / maintenance of databases should be an integral and mandatory part of all (inter)nationally funded projects. 

Building trust in the use of AI requires clear data source identification. As the AI applied should be based on original experimental data, the use of generative AI should be avoided as this technology would produce none original data which may result in bias such as overfitting. It also requires an understanding where AI can be used and how (applicability domains). AI outcomes should be assessed by experts to ensure that the results sense. The use of AI for regulatory purposes would require an active role of international organizations ensuring organizational imparity (e.g. OECD).

Finally, the in-house use of in silico technologies for the prioritization and informed decision making of further (in vitro) testing has a far lower threshold then regulatory replacement of any required test. Therefore, advancing the use of such applications, as well as communication of success and failure will help identify application domains and to build trust for subsequent regulatory use.  

Speaker: Ben van Ravenzwaay (ECETOC)

He received his degree as Doctor of Environmental Sciences/Toxicology at University of Wageningen in 1988, following the preparation of doctoral thesis at the German Cancer Research Centre (DKFZ) Heidelberg (Germany); 1979 – 1985 Study of Environmental Hygiene/Toxicology, University of Wageningen, Netherlands. 

He worked for 34 years at BASF SE, Ludwigshafen, Germany. The first years as head of the laboratory for toxicokinetics and metabolism. He then moved to regulatory toxicology and became group leader for regulatory toxicology for plant protection products. The last 20 years at BASF SE he worked as Senior Vice President of the Department for Experimental Toxicology and Ecology, which included as of 2019 BASF Metabolome Solutions GmbH, Berlin, Germany. 

He is an associate professor for Reproduction Toxicity of the University of Wageningen since 2012 and had a teaching assignment at the University of Kaiserslautern for industrial toxicology from 2006 until 2021. 

He is a member of the Scientific Committee of the European Centre for Ecotoxicology and Toxicology (ECETOC) since 2000, and chairman of the SC between 2013 and 2023. Moreover, he is member of editorial boards of “Archives of Toxicology”, “Chemical Biological Interactions” and “Toxicology Letters.  He was member of the board of trustees of Health and Environment Science Institute (HESI) from 2012 – 2018.  He is a Member of the German Society for Pharmacology and Toxicology, a European registered toxicologist and SOT-Member. He is an author of more than 250 peer reviewed papers. Since January 2022 he is an independent consultant for environmental sciences, Altrip, Germany