Modelling the quality of enantiomeric separations using Mutual Information as an alternative variable selection...

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning

Background:

• Quantitative structure-property relationship (QSPR) models are crucial for predicting chemical properties.
• Variable selection is essential for building robust and interpretable QSPR models.
• Traditional variable selection methods may have limitations in certain applications.

Purpose of the Study:

• To introduce and evaluate Mutual Information (MI) as a variable selection technique for QSPR.
• To compare the performance of MI-based variable selection with established methods.
• To assess the effectiveness of MI in modeling the enantioselectivity of molecules.

Main Methods:

• Mutual Information (MI) for variable selection.
• Support Vector Machines (SVM) for QSPR modeling.
• Comparison with Partial Least Squares (PLS) and Genetic Algorithms (GA).
• Modeling enantioselectivity across three chiral stationary phases.

Main Results:

• Variable selection is a necessary step in QSPR modeling.
• MI coupled with SVM provides a valuable alternative to PLS and GA.
• MI-based methods improve model interpretability by using fewer variables.
• Successful modeling of enantioselectivity using the proposed approach.

Conclusions:

• Mutual Information is an effective variable selection criterion for QSPR.
• The combination of MI and SVM offers a robust and interpretable modeling strategy.
• This approach enhances the predictive power and understanding of structure-property relationships.