Read-Across for Toxicological Data Gap Filling: State-of-the-Art, Challenges and Future Needs

Read-across is a well-established New Approach Methodology (NAM). Its approach and methods are well-developed and widely used, especially for data gap filling relating to in vivo tests. It is well supported by guidance and reporting templates, as well as computational tools. However, greater expertise and clarity are required to gain further acceptance. Challenges and needs focus on improving the acceptability of read-across predictions. These include: 1) understanding how uncertainties are quantified and how the overall (tolerable) uncertainty can be defined, 2) assessing similarity between target and source molecules, especially beyond strict 2D structural similarity, whilst avoiding activity cliffs, 3) key similarity challenges exist in metabolism, including the ability to confirm the principle pathways and biotransformations leading to toxicologically significant metabolites, as well as the rates of their formation, 4) the lack of suitable data to support read-across; thus, efforts to justify using non-standard data should continue to be developed, and 5) the use of artificial intelligence (AI) in read-across presenting its own set of challenges, which are, however, outweighed by the opportunities and gains. The role of AI in assessing similarity and, indeed, all aspects of read-across will grow, but at this time, how and at what speed it will grow is unknown. Read-across also has the potential to assist in the integration into non-animal chemical safety assessments and Next Generation Risk Assessment (NGRA).