Variable selection by an evolution algorithm using modified Cp based on MLR and PLS modeling: QSAR studies of...

Area of Science:

• * Cheminformatics and Computational Toxicology
• * Quantitative Structure-Activity Relationship (QSAR) studies

Background:

• * Variable selection is crucial for developing robust QSAR models using methods like Multiple Linear Regression (MLR) and Partial Least Squares (PLS).
• * Traditional methods can be computationally intensive and may not always identify the most informative molecular descriptors.

Purpose of the Study:

• * To develop and evaluate an evolution algorithm (EA) for efficient variable selection in MLR and PLS modeling.
• * To apply the developed QSAR approach for predicting the carcinogenicity of aromatic amines.

Main Methods:

• * Modified Cp statistic as the objective function within an EA for selecting molecular descriptors.
• * Application of the EA-enhanced variable selection for both MLR and PLS model formulation.
• * Prediction of carcinogenicity using the developed QSAR models for a set of aromatic amines.

Main Results:

• * The EA successfully identified parsimonious sets of informative descriptors for MLR models.
• * For PLS modeling, the EA selected a larger set of descriptors, leading to models based on a few latent variables.
• * The developed QSAR procedures demonstrated predictive capability for aromatic amine carcinogenicity.

Conclusions:

• * The evolution algorithm provides an effective strategy for variable selection in QSAR, optimizing descriptor combinations.
• * The proposed method enhances the predictive accuracy of MLR and PLS models for toxicological endpoints like carcinogenicity.
• * This approach aids in the computational assessment of chemical safety and the design of safer molecules.