Development of quantitative structure property relationship models and tool for predicting the soil adsorption...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• The soils/sediments organic carbon sorption coefficient (K_OC) is essential for assessing chemical behavior in the environment.
• Experimental K_OC determination is costly and time-consuming, necessitating predictive methods.
• Quantitative Structure-Property Relationship (QSPR) modeling offers a viable alternative for estimating K_OC values.

Purpose of the Study:

• To develop reliable quantitative structure-property relationship (QSPR) models for predicting the organic carbon sorption coefficient (K_OC).
• To create a user-friendly software tool for estimating K_OC values.
• To assess the predictive performance of the developed QSPR models and the software tool.

Main Methods:

• Development of QSPR models using 1477 experimental logK_OC values and seven machine learning algorithms.
• Identification of optimal models including univariate and multi-variable approaches, with and without logK_OW.
• Validation using internal and external datasets, assessing goodness-of-fit, robustness, and predictive ability.

Main Results:

• Three types of optimal models were developed, demonstrating excellent goodness-of-fit (R²_Train > 0.700), robustness (Q² > 0.600), and predictive ability (Q²_EXT > 0.700).
• A software tool, 'logKOC Predictor', was created using the developed models.
• External validation confirmed the tool's reliable estimation of unknown logK_OC values within its applicability domain.

Conclusions:

• The developed QSPR models and the 'logKOC Predictor' tool provide a reliable and efficient method for estimating logK_OC values.
• These predictive models can significantly aid in chemicals management by filling data gaps for K_OC.
• The tool's reliability is high for substances within the model's applicability domain and marked with 'High reliability'.