Feature selection for the imbalanced QSAR problems by using easyensemble.

Area of Science:

• * Cheminformatics and computational drug discovery.
• * Machine learning applications in pharmacology.

Background:

• * Quantitative Structure Activity Relationship (QSAR) models predict drug molecule activities, offering an alternative to costly experimental methods.
• * Predicting drug activity is challenging due to imbalanced datasets, where active molecules are rare.
• * Existing methods like asymmetric bagging and EasyEnsemble have limitations in handling data imbalance.

Purpose of the Study:

• * To develop an improved feature selection algorithm for QSAR modeling.
• * To address the challenge of imbalanced datasets in drug activity prediction.
• * To enhance the generalization performance of the EasyEnsemble classifier.

Main Methods:

• * Introduction of Prediction Risk based feature selection for EasyEnsemble (PREE), an embedded feature selection algorithm.
• * Application of PREE to drug molecule datasets.
• * Comparative analysis of PREE against asymmetric bagging and standard EasyEnsemble.

Main Results:

• * PREE demonstrated superior performance in predicting drug activities on tested datasets.
• * The proposed algorithm effectively handles imbalanced data, a common issue in drug discovery.
• * PREE improved the generalization capabilities of the EasyEnsemble classifier.

Conclusions:

• * PREE is an effective feature selection method for QSAR modeling, particularly for imbalanced datasets.
• * The algorithm offers a significant improvement over existing methods like asymmetric bagging and EasyEnsemble.
• * PREE contributes to more accurate and efficient drug discovery processes.