Machine Learning Classification of One-Chiral-Center Organic Molecules According to Optical Rotation.

Area of Science:

• Computational chemistry
• Machine learning in chemistry
• Stereochemistry

Background:

• Predicting the sign of optical rotation is crucial for stereoisomer identification.
• Machine learning offers a promising avenue for predicting molecular properties.
• Chiral descriptors can encode stereochemical information for computational models.

Purpose of the Study:

• To investigate machine learning algorithms for classifying organic molecules based on optical rotation sign.
• To develop predictive models for optical rotation sign across different solvents.
• To establish applicability domains for enhanced model accuracy.

Main Methods:

• Utilized diverse datasets of up to 13,080 compounds with optical rotation data.
• Employed chiral descriptors based on physicochemical and topological properties.
• Applied random forests (RF) and artificial neural networks for prediction.

Main Results:

• Achieved prediction accuracies up to 75% for dichloromethane, 82% for chloroform, and 82% for methanol.
• Identified RF probabilities and structural similarity as key factors for defining accurate applicability domains.
• Demonstrated the effectiveness of chiral descriptors in machine learning models.

Conclusions:

• Machine learning models, particularly RF and neural networks, can effectively predict the sign of optical rotation for chiral molecules.
• The choice of solvent significantly impacts prediction accuracy.
• Defining applicability domains based on model outputs and data characteristics improves predictive performance.