Graph neural network architectures for predicting the electrophilicity index: insights from 2D and 3D molecular graph representations

Key Points

• The aim is to benchmark graph neural networks for predicting the electrophilicity index of QM9 molecules.
• Comparison of 2D and 3D graph neural network architectures
• Evaluation of geometry-aware models like ALIGNN
• Assessment of computational efficiency of 2D models
• ALIGNN achieves the highest accuracy in predicting the electrophilicity index
• 2D models demonstrate superior computational efficiency compared to 3D models