Nucleophilicity Prediction Using Graph Neural Networks.

Area of Science:

• Organic Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Quantitative structure-activity relationships are crucial for understanding chemical reaction rates.
• Mayr's nucleophilicity parameter (N) is a key metric, but its experimental determination is time-consuming.
• Existing machine learning models often rely on predefined structural descriptors, potentially limiting their accuracy.

Purpose of the Study:

• To develop a data-driven model for predicting Mayr's nucleophilicity parameter (N).
• To leverage graph neural networks (GNNs) for a more natural representation of molecular structure.
• To improve prediction accuracy by integrating both structural and electronic molecular information.

Main Methods:

• A SchNet-based graph neural network (GNN) model was developed.
• The model initially used only molecular conformation and solvent type as input.
• Density Functional Theory (DFT)-calculated parameters were later incorporated as graph global features to enhance electronic information capture.

Main Results:

• The initial GNN model achieved comparable performance to benchmark studies (R² = 0.91, RMSE = 2.25).
• Incorporating DFT-calculated electronic parameters significantly improved prediction accuracy (R² = 0.95, RMSE = 1.63).
• The study demonstrated the effectiveness of GNNs in integrating diverse molecular information.

Conclusions:

• Graph neural networks (GNNs) offer an effective approach for predicting nucleophilicity parameters (N).
• Both molecular structure and electronic properties are essential for accurate N prediction.
• The proposed GNN model provides a more efficient and accurate alternative to experimental methods for determining N.