IM-MOTIC: An AI-Assisted Evidence-Driven Framework for Mechanism Mining, Target Exploration, and Validation in Environmental Chemical Toxicity

Introduction: The rapid emergence of environmental chemicals has increased the need for efficient approaches to characterize their mechanisms of toxicity. However, mechanistic evidence is often dispersed across toxicological databases, pathway annotations, experimental models, and published studies, making it difficult to identify biologically plausible and testable mechanisms. We developed Intelligent Mining-Mechanisms of Toxicity in Chemicals (IM-MOTIC) as an evidence-driven framework for mechanism inference, target exploration, and experimental validation.

Methods: IM-MOTIC integrates evidence from toxicity and chemical-disease databases (CTD), chemical bioactivity and target databases (ToxCast, ChEMBL), pathway and functional annotation resources (KEGG, GO, Reactome), and protein-protein interaction databases (STRING), supplemented by curated literature evidence. The framework includes five steps: defining the chemical or chemical group and toxicity endpoint; standardizing and integrating evidence; organizing weighted mechanism relationships; inferring candidate toxicity pathways; and prioritizing mechanisms using a Mechanism Confidence Score. Evidence weights are assigned according to evidence type, biological relevance, directness, species support, and literature reliability. IM-MOTIC was further used to extend prioritized mechanism networks, identify underexplored mechanism-associated targets, and guide molecular and cellular validation. Pesticide-associated neurotoxicity was used as a proof-of-concept case.

Results: IM-MOTIC generated evidence-linked mechanism maps connecting pesticides, molecular targets, biological pathways, neurotoxic phenotypes, and disease-related outcomes. In the pesticide neurotoxicity case, the framework prioritized mechanisms involving neuroinflammatory signaling, oxidative stress, mitochondrial and metabolic dysfunction, and epigenetic regulation. Network extension further highlighted molecular nodes and underexplored targets for experimental investigation. Subsequent validation supported the involvement of selected predicted targets and pathways in pesticide-induced neurotoxic responses, demonstrating the feasibility of using IM-MOTIC to infer, extend, prioritize, and validate mechanistic targets.

Conclusions: IM-MOTIC provides a practical framework for integrating heterogeneous toxicological evidence, prioritizing and extending candidate mechanisms, identifying potential molecular targets, and guiding experimental validation. This approach may support mechanism-oriented toxicology research and next-generation risk assessment of environmental chemicals.