Related Experiment Video
Updated: Apr 11, 2026
In Silico Modeling Method for Computational Aquatic Toxicology of Endocrine Disruptors: A Software-Based Approach Using QSAR Toolbox
Published on: August 28, 2019
Hybrid experimental-QSAR-artificial intelligence framework for chronic ecological risk assessment of emerging
Vinicius Roveri1, Gilmar Aparecido Dos Santos2, Walber Toma2
1Universidade Metropolitana de Santos (UNIMES), Avenida Conselheiro Nébias, 536 - Encruzilhada, 11045-002, Santos, São Paulo, Brazil; Centro Interdisciplinar de Investigação Marinha e Ambiental (CIIMAR/CIMAR), Avenida General Norton de Matos S/N, 4450-208, Matosinhos, Portugal; Laboratório de Pesquisa em Produtos Naturais, Universidade Santa Cecília (UNISANTA), Rua Oswaldo Cruz, 266, C21, bloco C, 11045-907, Santos, São Paulo, Brazil.
Abstract:
Urbanized coastal zones are increasingly exposed to complex mixtures of pharmaceuticals and illicit drug markers (PhACs), yet ecological risk assessment (ERA) remains constrained by limited experimental chronic toxicity data, particularly for marine environments. Here, we applied an integrated Weight-of-Evidence (WoE) framework combining experimentally derived ecotoxicological endpoints with quantitative structure-activity relationship (QSAR)-based predictive toxicology tools (ECOSAR and VEGA) and an artificial-intelligence deep-learning model (TRIDENT) to assess chronic ecological risks of twelve PhACs in drainage channels discharging into Enseada Beach (Guarujá, southeastern Brazil). Experimental evidence identified acetaminophen (risk quotients: RQ = 25.4, fish) and caffeine (RQ = 1.52, algae) as high-risk substances, while diclofenac exhibited negligible risk (RQ < 0.01). First-tier QSAR screening with ECOSAR highlighted caffeine, losartan, and benzoylecgonine as priority pollutants, with RQs reaching 3.07 for losartan in fish. VEGA applicability-domain (ADI) analysis indicated that approximately 65-70% of compound-taxon combinations were predicted within a reliable chemical space (ADI ≥ 0.85), increasing confidence in chronic risk estimates for structurally complex and ionizable compounds, including cocaine and benzoylecgonine. TRIDENT predictions moderated extreme QSAR outputs while preserving taxon-specific sensitivity, confirming caffeine, acetaminophen, and losartan as high-risk compounds and classifying cocaine and benzoylecgonine as moderate risk in the final WoE synthesis. Algae emerged as the most sensitive taxonomic group, followed by crustaceans and fish. Overall, this study demonstrates the value of integrating QSAR tools and AI-based models within a WoE framework to improve ERA in data-limited tropical coastal environments, highlighting a transferable and transparent methodology with broad international relevance.
