Quantum Descriptor-Based Machine-Learning Modeling of Thermal Hazard of Cyclic Sulfamidates

Area of Science:

• Organic Chemistry
• Chemical Engineering
• Computational Chemistry

Background:

• Cyclic sulfamidates are essential in organic synthesis.
• Accurate thermal criticality classification is crucial for safe process development and scale-up.
• Limited experimental data for reaction enthalpies poses a challenge for modeling.

Purpose of the Study:

• To develop predictive models for reaction enthalpies of 5-membered cyclic sulfamidates.
• To utilize quantum-chemical descriptors for improved modeling accuracy with sparse data.
• To ensure safe handling and scale-up of cyclic sulfamidates in chemical processes.

Main Methods:

• Utilized quantum-chemical descriptors, which are more chemically relevant than traditional cheminformatics descriptors.
• Developed three predictive models: Partial Least Squares (PLS) with Genetic Algorithm (GA), Least Absolute Shrinkage and Selection Operator (LASSO), and Gaussian Process Regression (GPR).
• Applied chemistry-aware modeling techniques suitable for small datasets.

Main Results:

• Achieved coefficients of determination of 0.78 (PLS-GA), 0.67 (LASSO), and 0.74 (GPR).
• Despite prediction errors around 100 J/g, all models accurately classified thermal criticality.
• Demonstrated the effectiveness of quantum-chemical descriptors in predicting reaction enthalpies.

Conclusions:

• The developed models provide a reliable framework for preliminary safety assessment of cyclic sulfamidates.
• Quantum-chemical property-based modeling is effective for small datasets in predicting thermal behavior.
• This approach supports informed decision-making in process development and scale-up for safer chemical synthesis.