AMOI: A Conceptual Framework for Deep Integration of Omics, ML, and AOPs─A Critical Review with Illustrative Case Analysis

This is the authors' abstract. We don't add key points for this paper.

Although new approach methodologies (NAMs), such as omics technologies, machine learning (ML) and adverse outcome pathways (AOPs), have advanced chemical risk assessment individually, their simple serial integration─omics → ML → AOP─cannot fully exploit the predictive and mechanistic potential of these approaches. Based on a systematic review, we propose a paradigm shift toward deep integration, termed AOP-guided, ML-driven multiomics integration (AMOI), in which ML algorithms leverage dose–response and temporal information from multiomics data, as well as the causal AOP structure, to iteratively quantify, refine and update quantitative AOPs (qAOPs). Through an illustrative case study on AhR activation and aromatase inhibition, we demonstrate that omics-based benchmark doses (BMDomics) can anchor qAOPs and that ML facilitates cross-layer multiomics integration and key event relationship (KER) quantification. However, the case also highlights critical limitations, particularly the unreliability of complex ML models under small sample constraints. We therefore provide practical guidelines for selecting an ML strategy across sample size regimes and outline a roadmap for community-driven data infrastructure. This work provides a conceptual framework within NAMs, outlining a path from conceptual exploration toward regulatory-ready next-generation risk assessment (NGRA).