Machine Learning Predicts Degree of Aromaticity from Structural Fingerprints.

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Machine learning

Background:

• Aromaticity, often determined by Hückel's rule, is a complex property.
• Existing binary classifications do not fully capture variations in chemical and biological behavior.

Purpose of the Study:

• To develop a machine learning method for quantifying the degree of aromaticity in molecules.
• To create distinct categories of aromaticity for better structure-activity relationship analysis.

Main Methods:

• Quantified aromaticity for molecules in a public dataset using an extension of Raczyńska et al.'s work.
• Developed a machine learning approach to predict the degree of aromaticity for individual aromatic rings.
• Derived categories from numerical aromaticity results.

Main Results:

• Established a quantitative measure for aromaticity beyond the binary Hückel's rule classification.
• Generated distinct categories of aromaticity based on quantified data.
• Demonstrated the potential for differentiating structural patterns and their associated behaviors.

Conclusions:

• A quantitative approach to aromaticity improves the representation of chemical and biological properties.
• Machine learning-based aromaticity categorization enhances expert systems and quantitative structure-activity relationship (QSAR) models.
• This method offers a more nuanced understanding of aromatic systems.