Bioactive Molecule Prediction Using Extreme Gradient Boosting.

Area of Science:

• Computational chemistry
• Cheminformatics
• Machine learning in drug discovery

Background:

• The increasing volume of chemical and biological data necessitates advanced computational methods for drug discovery.
• Computer-aided drug discovery (CADD) relies heavily on data mining and machine learning techniques.
• Traditional drug discovery methods are being augmented by sophisticated computational approaches.

Purpose of the Study:

• To evaluate the efficacy of extreme gradient boosting (Xgboost) for predicting biological activity.
• To compare Xgboost's performance against established machine learning algorithms using quantitative molecular structure descriptions.
• To assess Xgboost's capability in handling diverse and imbalanced biological activity datasets.

Main Methods:

• Utilized extreme gradient boosting (Xgboost), an ensemble method based on Classification and Regression Trees (CART) and Gradient Boosting Machine.
• Applied Xgboost to seven benchmark datasets commonly used in biological activity prediction studies.
• Compared Xgboost against Random Forest (RF), Least Squares Support Vector Machines (LSVM), Radial Basis Function Neural Network (RBFN), and Naïve Bayes (NB).

Main Results:

• Xgboost demonstrated superior performance in predicting biological activities compared to RF, LSVM, RBFN, and NB.
• The algorithm effectively identified minority activity classes within highly imbalanced datasets.
• Xgboost exhibited robust performance across datasets with both high and low molecular diversity.

Conclusions:

• Extreme gradient boosting (Xgboost) is a highly effective machine learning method for predicting biological activity.
• Xgboost offers significant advantages over other common algorithms, particularly for challenging datasets.
• The findings support the integration of Xgboost into computational drug discovery pipelines for enhanced efficiency and accuracy.