QSRR modeling for diverse drugs using different feature selection methods coupled with linear and nonlinear...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Quantitative Structure-Retention Relationship (QSRR) models are crucial for predicting chromatographic behavior of compounds.
• Previous QSRR models for drug retention on poly butadiene columns used Classification And Regression Trees (CART).
• Effective feature selection is vital for developing robust QSRR models.

Purpose of the Study:

• To develop accurate QSRR models for estimating the chromatographic retention of 83 diverse drugs.
• To evaluate Ant Colony Optimization (ACO) as a feature selection method for molecular descriptors in QSRR.
• To compare ACO with other feature selection techniques like Genetic Algorithms, Stepwise Regression, and Relief.

Main Methods:

• Utilized Ant Colony Optimization (ACO) for selecting optimal molecular descriptors from a large pool.
• Employed Multiple Linear Regression (MLR) and Support Vector Machines (SVMs) for QSRR model development.
• Validated models by correlating experimental retention factors with predicted values.

Main Results:

• ACO effectively identified important molecular descriptors for QSRR.
• Both MLR and SVM models demonstrated excellent correlation between experimental and predicted drug retention factors (logk(w)).
• The Support Vector Machine (SVM) model, utilizing descriptors selected by ACO, achieved the best predictive performance.

Conclusions:

• Ant Colony Optimization is a powerful tool for feature selection in QSRR studies.
• The developed QSRR models, particularly the SVM-ACO model, accurately predict drug retention, aiding in chromatographic method development.
• This approach enhances the understanding of structure-retention relationships in drug analysis.