Methodological Challenges in the Application of QSAR Models for Chemical Prioritization and Toxicity Assessment: A Case Study on Aryl Hydrocarbon Receptor Activity in Environmental Pollutant Mixtures

TL;DR

This study develops a QSAR model to prioritize chemicals based on their toxicity levels, helping assess environmental pollutants more effectively.

Contribution

A QSAR model that classifies compounds into multiple activity levels for aryl hydrocarbon receptor (AhR) activity is developed and validated.

Findings

The QSAR model achieved 77–87% weighted accuracy in predicting AhR activity levels.

The model successfully identified high-AhR activity compounds like benzonaphthothiophene and perylene.

Combining QSAR predictions with experimental data improved site-specific toxicity assessments.

Abstract

The complexity of
chemical mixtures in the environment
challenges
their in-depth risk assessment due to the diverse compounds in use
and the lack of experimental toxicity data. In silico models can be
used to fill data gaps for compounds with unknown toxic potency. QSAR
models typically distinguish only between active and inactive compounds,
providing no information about the levels of activity. In this study,
a quantitative structure–activity relationship (QSAR) model
that classifies compounds into multiple activity levels was developed
to address data gaps in the levels of aryl hydrocarbon receptor-mediated
(AhR) activity of compounds commonly detected in environmental samples.
Its practical applicability has been demonstrated on highly complex
mixtures of aquatic pollutants from the Joined Danube Survey to prioritize
the most relevant compounds for experimental assessment. The model’s
performance showed high sensitivity and specificity, with weighted
overall accuracy ranging from 77 to 87%. The combination of experimental
and QSAR predicted data was used to calculate site-specific AhR activity,
which was compared to the overall AhR activity detected by in vitro
bioassays. Experimental testing confirmed the ability of the QSAR
model to identify compounds with high AhR activity, including benzonaphthothiophene,
perylene, acridone, and triphenylene, and prioritize the most relevant
suspected effect drivers. Our model can predict toxic potency and
thus prioritize the potential bioactive compounds based on specific
activity levels. Our study shows that when QSAR models are used for
compound prioritization, several factors must be considered: cytotoxicity,
solubility, the high rate of false positives for low-toxicity compounds,
and the model’s applicability domain.

Figures