ElectroPredictor: An Application to Predict Mayr's Electrophilicity E through Implementation of an Ensemble Model...

Area of Science:

• Computational chemistry
• Machine learning in chemistry
• Organic reactivity prediction

Background:

• Electrophilicity (E) is a key parameter for understanding organic molecule reactivity.
• The theoretical electrophilicity index (ω) has limitations in predicting E for diverse compounds.
• Accurate prediction of E is essential for designing new chemical reactions and materials.

Purpose of the Study:

• To develop a robust ensemble model for predicting electrophilicity (E) in organic molecules.
• To assess the predictive power of various descriptors and machine learning algorithms.
• To identify limitations of the theoretical electrophilicity index (ω) for diverse chemical structures.

Main Methods:

• Creation of a robust ensemble model using Mayr's reactivity database.
• Combination of topological and quantum mechanical descriptors with machine learning algorithms.
• Validation using statistical parameters, training/test partition, applicability domain, and y-scrambling tests.

Main Results:

• The global ensemble model achieved excellent predictability with Q5-fold² = 0.909 and Qext² = 0.912.
• The theoretical electrophilicity index (ω) was found to be a poor descriptor for predicting E, particularly for neutral compounds.
• A Python application, ElectroPredictor, was developed for predicting E values.

Conclusions:

• Ensemble models incorporating nonlinear machine learning and topographic descriptors are necessary for precise E prediction.
• Separating charged and neutral compounds improves prediction accuracy.
• Virtual screening of the QM9 dataset identified 10 potential new Mayr's electrophiles.