Chemometric modelling for eco-toxicological risk assessment of diverse chemicals towards mammalian species

Abstract

The extensive use of chemicals in daily life can cause environmental pollution, raising global concerns about the health and safety of mammals and other species in the food chain. Many chemicals, including pesticides, pharmaceuticals, etc., can get inside the body of an organism via inhalation, ingestion, or dermal contact, causing toxic effects and potentially leading to the death of species. This research aims to assess the toxicity of diverse chemicals toward mammalian species, considering the endpoint as median lethal dose (LD₅₀). A partial least squares (PLS)-based quantitative read-across structure-activity relationship (q-RASAR) model was established by merging the structural and physicochemical properties of quantitative structure-activity relationship (QSAR) with the read-across (RA)'s similarity and error-based measures. An ARKA-RASAR approach was additionally applied to assess the model performance. Despite of stronger internal validation of ARKA-RASAR model, the PLS-based q-RASAR model was selected as the final model because of its improved external predictivity (Q²_F1 = 0.616-0.690; Q²_F2 = 0.616-0.689, MAE_test = 0.331-0.303) along with its lower complexity and ease of interpretation. For possible toxic substances identification, the q-RASAR model was also utilized to screen the Pesticides Properties DataBase (PPDB) and ECOTOX repository, ensuring the practical application of the developed model. The model revealed specific fragments like the polar surface area, electronegativity, molecular complexity, and lipophilicity, all of which are responsible for the enhancement of mammalian toxicity. This aids in the development of eco-friendly chemicals and bridging data gaps, thus bypassing traditional toxicity assessment methods.