A Comparative Study of Atom-Centered And Edge-Neighborhood Graph Representations In Qspr Modeling of Green Pesticides

Abstract

Pesticides are indispensable for agricultural productivity, yet their widespread use creates environmental risks such as persistence, bioaccumulation, and ecotoxicity. Early prediction of these risks is therefore important, especially because experimental evaluation of new chemical compounds is costly, time-consuming, and ethically constrained. In this context, Quantitative Structure-Property Relationship (QSPR) modeling provides a useful computational tool for the rational assessment of environmentally friendly pesticides.This study compares two hydrogen-inclusive molecular graph representations for a structurally diverse set of sixteen pesticide molecules: the atom-centered G representation and the bond-centered edge-neighborhood N_e (G) representation. Simple Linear Regression-based QSPR models were developed for six physicochemical and environmental fate properties, and the predictive behavior of topological indices derived from both representations was comparatively evaluated. The results indicate a complementary, property-dependent pattern. The G-based models were slightly more favorable for extensive bulk-related properties such as molar refractivity and polarizability, whereas N_e (G)-based models yielded modestly higher coefficient of determination (R²) values for soil organic carbon-water partition coefficient (logK_oc), bioconcentration factor (logBCF), and octanol-water partition coefficient (logP) within the descriptor families examined. These differences are interpreted as controlled, dataset-specific trends rather than evidence of universal representational superiority. Overall, the findings suggest that hydrogen-inclusive edge-neighborhood graph representations may provide an informative complementary framework for QSPR modeling and early environmental risk assessment of pesticides.