Learning Multidimensional Electronegativity of Selected Atom Types in Organic Molecules Using Graph Neural Networks

Area of Science:

• Computational Chemistry
• Machine Learning
• Quantum Chemistry

Background:

• Electronegativity, a concept proposed by Pauling, is traditionally data-driven.
• Updating electronegativity with extensive experimental and computational data has been neglected.
• Artificial intelligence (AI) offers new possibilities for data-driven chemical concepts.

Purpose of the Study:

• To develop a data-driven method for generating multidimensional electronegativity.
• To enhance the classification of atoms and covalent bonds in organic molecules.
• To improve molecular machine learning task performance.

Main Methods:

• Utilized graph neural networks (GNNs) for a data-driven approach.
• Generated multidimensional electronegativity, focusing on 2D for detailed classification.
• Integrated the new electronegativity into molecular machine learning models.

Main Results:

• Achieved improved performance in molecular machine learning tasks.
• Demonstrated the utility of AI in refining chemical concepts like electronegativity.
• Provided a more informative, multidimensional representation of atomic electronegativity.

Conclusions:

• AI-driven methods can significantly enhance traditional chemical concepts.
• The proposed multidimensional electronegativity offers a more nuanced understanding of chemical bonds.
• This approach has broad applicability in chemical studies and AI-driven research.