Identification of structural determinants of acute toxicity towards Crassostrea virginica: Multivariate QSAR modeling and prioritization of diverse chemicals.

Abstract

Crassostrea virginica, a bivalve mollusc, is a critical bioindicator for monitoring the environmental quality of coastal and marine ecosystems. They support essential habitats for other marine species and are also valued for their significant dietary importance to humans. It is a widely used sentinel species in ecotoxicological studies. The objective of the present work is to predict the acute toxicity (pEC50) of diverse organic chemicals in this species by developing chemometric models. Here, we utilized a dataset of 226 unique chemicals to generate quantitative structure-activity relationship (QSAR) models. Furthermore, the implementation of the intelligent consensus prediction (ICP) approach improved prediction quality by overcoming the estimation bias of individual models for a particular test compound. The consensus model 2 (CM2) is regarded as the best model, with lower prediction error (MAE95%(test) = 0.381) and higher predictive performance (Q2F1 = 0.761, Q2F2 = 0.759). The SHapley Additive exPlanations (SHAP) analysis revealed that hydrophobicity is the most impactful property contributing towards intoxication. The crucial toxicophores and non-toxicophores identified from the models will help to gain a deep insight for environmental risk assessment. Screening the Pesticides Properties Database (PPDB) helped to prioritize the potential toxicants. The study provides a sustainable framework for the early identification of potential hazards to safeguard this sentinel species and, consequently, the marine ecosystems.