High-Throughput Photocatalysis for Generating Reliable Datasets Analyzed by Machine Learning.

Area of Science:

• Environmental Science
• Materials Science
• Computational Chemistry

Background:

• Photocatalysis offers an eco-friendly approach for water contaminant removal.
• Assessing novel photocatalytic materials against limited analytes may not reflect real-world performance.
• A need exists for efficient methods to evaluate photocatalytic materials across various contaminants.

Purpose of the Study:

• Introduce a high-throughput photocatalysis (HTP) method for rapid material assessment.
• Develop machine learning (ML) models to predict contaminant photodegradation.
• Correlate molecular structure with photocatalytic reactivity.

Main Methods:

• Developed a modular high-throughput photocatalysis (HTP) experimental setup.
• Collected photodegradation data for multiple analytes under varying conditions.
• Applied linear regression, random forest (RF), and neural network (NN) ML models using molecular fingerprints.
• Utilized SHapley additive exPlanations (SHAP) for structure-activity relationship analysis.

Main Results:

• RF and NN models accurately estimated photodegradation percentages for unknown molecules without overfitting.
• HTP method demonstrated time-effective evaluation of photocatalytic materials.
• SHAP analysis successfully identified molecular substructures influencing photocatalytic degradation rates.
• Generated reactivity heatmaps that align with detailed dye degradation studies.

Conclusions:

• HTP combined with ML provides a robust platform for evaluating photocatalytic materials.
• The developed models can predict degradation efficiency and identify reactive sites within molecules.
• This approach accelerates the discovery and optimization of materials for water remediation.