Predictive Modelling of Pesticide Properties for Risk Assessment: A Curated Dataset and QSAR Evaluation for 110 Active Ingredients

Abstract

Pesticides play a critical role in global food security. However, some of them also pose significant environmental and health risks, particularly in tropical regions where data gaps hinder accurate risk assessments. This study compiles and curates properties for 110 pesticide active ingredients commonly used in Cameroon and other tropical African countries, focusing on partitioning between octanol, air, and water (KOW, KOA, and KAW) and between organic carbon and water (KOC), solubilities in water (SW) and in octanol (SO), vapour pressure (VP), and biodegradation half-lives (HLbiodeg). Empirical data were harmonized using thermodynamic adjustment, and missing properties were predicted using common quantitative structure–activity relationships (QSARs), including Poly-Parameter Linear Free Energy Relationships (PP-LFERs) with empirical Abraham solute descriptors, IFS QSAR, EPI Suite™ and OPERA. Model performance was evaluated against final adjusted empirical values, revealing strong reliability for KOW and SW but greater uncertainty for KAW, KOA, VP, and HLbiodeg. Consensus values from multiple models improved prediction performance. The study highlights the scarcity of experimentally derived data for key properties, especially for ionizable and large-molecule pesticides. The study also demonstrates that reactivity parameters like HLbiodeg often play a more influential role compared to partitioning properties in determining the levels of human and ecological exposures, expressed as intake fraction (iF) or Drinking Water Exposure Potential (DWEP). These findings underscore the need for expanded experimental datasets to refine predictive models and support evidence-based pesticide regulation in tropical regions.