Exploring regression-based QSTR and i-QSTR modeling for ecotoxicity prediction of diverse pesticides on multiple avian species

Abstract

Ensuring the protection of endangered bird species from pesticide exposure plays a vital role in safeguarding ecosystem integrity. The task of predicting pesticide toxicity and conducting risk assessments has become increasingly challenging in recent times. Within this research endeavor, we have undertaken the development of regression-based quantitative structure-toxicity relationship (QSTR) and interspecies (i-QSTR) models. These models were constructed employing an extensive dataset of 664 pesticides following the guidelines set forth by the Organization for Economic Co-operation and Development (OECD). Our primary objective was to identify the fundamental characteristics responsible for the toxicity of pesticides on various avian species, including the mallard duck (MD), bobwhite quail (BQ), and zebra finch (ZF). By evaluating various globally accepted internal and external statistical parameters, we have demonstrated that our models exhibit reliability and robustness. An intelligent consensus algorithm was used to make the models more predictive. As a result of intelligent consensus prediction (ICP), test compound consensus predictability (winner model is CM3) showed better results than individual models. An attempt has been made to interpret the descriptors of the developed model from a mechanistic perspective, catering to principle 5 of OECD guidelines, in which the presence of phosphate, oxygen, ether linkage, carbamates and halogens in the backbone structure of pesticides is associated with avian toxicity. Finally, we have concluded that groups that are linked with the electronegativity and lipophilicity of a compound may escalate pesticide-induced toxicity. Developed i-QSTR models can be employed for the prediction of species-specific pesticide toxicity.