Structural and Electronic Features-Integrated Machine Learning Framework for High-Throughput Prediction of Organic...

Area of Science:

• Environmental Chemistry
• Computational Chemistry
• Chemical Engineering

Background:

• Predicting organic pollutant reactivity is vital for advanced oxidation processes (AOPs).
• Structural complexity and diversity of pollutants challenge current predictive models.
• Developing interpretable, high-throughput frameworks is essential for efficient AOP design.

Purpose of the Study:

• To develop a machine learning (ML) framework for predicting organic pollutant reactivity towards sulfate radicals.
• To integrate quantum chemical and molecular fingerprint descriptors for enhanced prediction accuracy.
• To establish quantitative structure-activity relationships (QSAR) and quantitative read-across structure-activity relationship (QR-SAR) models.

Main Methods:

• Utilized a machine learning approach combining RDKit and conceptual density functional theory (CDFT) descriptors.
• Identified key structure-activity relationships including E_HOMO(N), electron-donating capacity, ring structures, branching, and molecular surface areas.
• Developed a quantitative read-across structure-activity relationship model incorporating intermolecular similarity.

Main Results:

• Achieved a 2.1-fold increase in applicability domain (AD) compared to traditional QSAR, covering 74.3% across 12 pollutant classes.
• Identified quantitative reactivity thresholds for key structure-activity features.
• Experimental validation demonstrated strong predictive performance with R^2 = 0.811.

Conclusions:

• The developed ML framework offers a transparent and high-throughput method for predicting organic pollutant reactivity.
• This approach facilitates the rational design of AOPs tailored to specific pollutant profiles.
• The integrated descriptor approach enhances predictive power and applicability domain for environmental remediation strategies.