Advancing chemical safety prediction: an integrated GNN framework with DFT-augmented cyclic compound solution.

Area of Science:

• Computational Chemistry and Cheminformatics
• Machine Learning for Materials Science
• Chemical Engineering and Safety

Background:

• Assessing safety-critical physicochemical properties of rapidly proliferating chemical substances is challenging.
• Existing methods for predicting properties like Heat of Combustion (HoC), Vapor Pressure (VP), and Flashpoint have limitations.
• A unified and robust approach is needed for simultaneous prediction of multiple safety properties.

Purpose of the Study:

• To develop an integrated Graph Neural Network (GNN) based approach for predicting HoC, VP, and Flashpoint.
• To improve the prediction accuracy for challenging chemical structures, particularly cyclic compounds.
• To create a real-time prediction system for practical application in chemical safety assessment and emergency response.

Main Methods:

• Developed a unified prediction model using Graph Neural Networks (GNNs) on comprehensive datasets.
• Implemented a hybrid approach combining DFT calculations and Random Forest modeling for cyclic compounds.
• Integrated the prediction model into a real-time system using Flask, supporting SMILES notation and structure drawing.

Main Results:

• Achieved high prediction accuracy with mean absolute errors of 126 J/mol for HoC (R²=0.993), 0.617 log units for VP (R²=0.898), and 14.42 °C for Flashpoint (R²=0.839).
• Significantly improved HoC prediction for cyclic compounds with an R² of 0.918 using the specialized hybrid approach.
• The real-time system allows for input of chemical structures and comparison with experimental data and benchmarks.

Conclusions:

• The integrated GNN framework provides a robust and unified solution for predicting multiple safety-critical properties.
• The specialized treatment for cyclic compounds enhances the reliability of machine learning models in chemical safety.
• The real-time prediction system offers practical utility for chemical safety assessment and emergency response planning.