K _OC-WebPredictor: An Open-Access Tool for Prediction and Insights into Soil Sorption

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• The soil organic carbon-water partition coefficient (KOC) is crucial for understanding contaminant behavior in soil.
• Experimental KOC measurement is accurate but costly and time-consuming, hindering comprehensive chemical assessment.
• A predictive approach is needed to evaluate the mobility and persistence of numerous chemicals.

Purpose of the Study:

• To develop and validate predictive models for estimating log KOC values.
• To compare the performance of quantitative structure-activity relationship (QSAR), quantitative Read-Across Structure-Activity Relationship (q-RASAR), and machine learning (ML) models.
• To provide an accessible tool for assessing chemical sorption and mobility in soil.

Main Methods:

• Development of interpretable QSAR and q-RASAR models using molecular descriptors.
• Optimization of a multiple linear regression (MLR)-based QSAR model using 824 diverse compounds.
• Application of ML approaches and statistical validation (R², Q²LOO, Q²F1, paired t-tests, Wilcoxon signed-rank tests).
• External screening of 7,612 chemicals from the U.S. EPA's CPDat inventory.
• Development of the open-access KOC-WebPredictor web tool.

Main Results:

• The optimized MLR-QSAR model demonstrated strong predictive performance (R² = 0.85, Q²LOO = 0.84, Q²F1 = 0.84).
• The QSAR model significantly outperformed the q-RASAR variant in predictive accuracy and robustness (p < 0.05).
• Mechanistic insights revealed that hydrophobicity, aromaticity, and halogenation increase sorption, while polar/phosphorus groups enhance mobility.
• The models achieved 94% applicability domain coverage for a large external chemical dataset, facilitating regulatory data gap filling.

Conclusions:

• Interpretable QSAR, q-RASAR, and ML models provide reliable alternatives to experimental KOC determination.
• The developed QSAR model offers a robust and accurate method for predicting soil sorption behavior.
• The KOC-WebPredictor tool enables efficient assessment of chemical mobility, supporting environmental risk assessment and chemical prioritization.