Modeling and predicting chiral stationary phase enantioselectivity: An efficient random forest classifier using an...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Chiral separations are crucial in pharmaceutical and chemical industries.
• Predicting chiral column performance is often empirical and time-consuming.
• Enantioselectivity prediction requires robust computational methods.

Purpose of the Study:

• To develop a novel, accurate prediction approach for chiral column effectiveness.
• To reduce the reliance on empirical trial-and-error methods in chiral separations.
• To enable structure-based prediction of enantioselective behavior.

Main Methods:

• Combined a random forest classifier with an optimized discretization method.
• Trained models on enantioselectivity data divided into four classes.
• Employed over-sampling, down-sampling, and binary classification aggregation for optimization.
• Utilized layered fingerprints as descriptors for 41 chiral stationary phases.

Main Results:

• Achieved successful structure-based prediction of enantioselective behavior for 34 chiral columns.
• Demonstrated high performance using metrics like area under the ROC curve, Kappa indices, and F-measure.
• Validated the predictive capability of the developed learning methodology.

Conclusions:

• The novel prediction approach significantly improves the accuracy of forecasting chiral column performance.
• This method offers a more efficient alternative to traditional trial-and-error techniques.
• The findings contribute to advancing computational strategies in chiral chromatography.