A Novel Approach of Multilevel Clustering of Adverse Outcome Pathway Instances for the Prediction of Adverse Outcome

Description

Translating Adverse Outcome Pathways (AOPs) into mathematical models is in the center of quantitative prediction of adverse outcome (AO). This poster introduces a novel approach that AOP instances are used in multilevel clustering. In this work, an AOP instance is defined by a certain sequential combination of key events and AO. The impact of biochemical properties of chemicals are modeled to be homogenous among the chemicals who exhibit the same AOP instances. Moreover, chemicals are modeled to show differential impacts from biochemical properties if they follow different AOP instances. AOP instance allows flexible reconstruction of key event relationship network. Then the data is clustered by AOP instances under a multilevel structure. A Bayesian hierarchical model is employed to integrate uncertainty assessment, specifically the multivariate covariance. A previous AOP case study of developmental neurological toxicity from Spinu et al. (2022) is used for demonstration. Results show that this novel multilevel approach has improvements over the original model in terms of predictive performance.

Period	2024 Feb 5
Held at	RISK-HUNT3R, Netherlands
Degree of Recognition	International