Development of models predicting biodegradation rate rating with multiple linear regression and support vector...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Toxicology

Background:

• Biodegradation is crucial for removing organic chemicals from the environment.
• Assessing biodegradability is key to understanding the environmental persistence of chemicals.
• Predictive models are needed to evaluate biodegradability efficiently.

Purpose of the Study:

• To develop quantitative structure-activity relationship (QSAR) models for predicting primary and ultimate biodegradation rates.
• To evaluate the performance of multiple linear regression (MLR) and support vector machine (SVM) algorithms for biodegradability prediction.
• To establish models with high predictive accuracy and defined applicability domains.

Main Methods:

• Development of four QSAR models using MLR and SVM algorithms on a dataset of 171 organic compounds.
• Splitting the dataset into two subsets based on carbon atom number (≤9 for MLR, >9 for SVM).
• Validation of models using determination coefficient (R²), cross-validation (Q²LOO), and external validation (Q²ext). Applicability domains visualized using the Williams plot.

Main Results:

• MLR models identified nArX (number of X on aromatic ring) as a key descriptor for both primary and ultimate biodegradation.
• SVM models achieved R² and Q² values over 0.9, demonstrating excellent goodness-of-fit, robustness, and external predictive power.
• The developed models showed satisfactory predictive abilities for biodegradability.

Conclusions:

• The developed QSAR models are effective tools for predicting the biodegradability of organic chemicals.
• These models can aid in assessing the environmental persistence and potential risks of chemicals.
• The study highlights the utility of MLR and SVM in environmental chemistry modeling.