Artificial intelligence as the new frontier in chemical risk assessment

Abstract

The rapid progress of AI impacts various areas of life, including toxicology, and promises a major role for AI in future risk assessments. Toxicology has shifted from a purely empirical science focused on observing chemical exposure outcomes to a data-rich field ripe for AI integration. AI methods are well-suited to handling and integrating large, diverse data volumes - a key challenge in modern toxicology. Additionally, AI enables Predictive Toxicology, as demonstrated by the automated read-across tool RASAR that achieved 87% balanced accuracy across nine OECD tests and 190,000 chemicals, outperforming animal test reproducibility. AI's ability to handle big data and provide probabilistic outputs facilitates probabilistic risk assessment. Rather than just replicating human skills at larger scales, AI should be viewed as a transformative technology. Despite potential challenges, like model black-boxing and dataset biases, explainable AI (xAI) is emerging to address these issues.

Keywords: big data; computational toxicology; machine learning; regulatory toxicology; scientific revolution.

Figure 1

ToxAIcology - AI-assisted toxicology. The figure depicts how new approaches are increasingly providing Big Data and have transformed toxicology to a data-rich discipline. The increase in algorithm and computational power (Big Computer) now allow making Big Sense from evidence retrieval to its integration to predictions and the support to their reporting.