R software for QSAR analysis in phytopharmacological studies.

Area of Science:

• Computational chemistry
• cheminformatics
• bioinformatics

Background:

• Quantitative structure-activity relationship (QSAR) analysis is crucial for drug design and natural product research.
• The proliferation of descriptors necessitates robust selection methods for accurate QSAR modeling.
• Identifying relevant independent variables is challenging due to the vast number of available descriptors.

Purpose of the Study:

• To demonstrate various descriptor selection techniques for QSAR studies, including Boruta, all subsets regression, ANOVA, AIC, stepwise regression, and genetic algorithms.
• To perform comprehensive regression diagnostics using R software to assess model reliability.
• To provide researchers with accessible tools for selecting descriptors and diagnosing errors in QSAR analysis.

Main Methods:

• Utilized descriptor selection methods: Boruta approach, all subsets regression, ANOVA, AIC, stepwise regression, and genetic algorithm.
• Employed R software for regression diagnostics, including normality, linearity, residual histograms, PP plots, multicollinearity, and homoscedasticity tests.
• Developed a workflow to integrate descriptor selection and regression diagnostics for QSAR model development.

Main Results:

• The Boruta approach and genetic algorithm proved superior in selecting relevant independent variables for QSAR models.
• Regression diagnostics successfully identified and addressed model errors, enhancing QSAR model reliability.
• The implemented workflow effectively highlights various descriptor selection and diagnostic procedures for QSAR studies.

Conclusions:

• Effective descriptor selection and regression diagnostics are essential for developing reliable QSAR models.
• The Boruta approach and genetic algorithm are recommended for robust variable selection in QSAR.
• This study provides a practical and adaptable framework for researchers conducting QSAR analysis.