Prediction of factor Xa inhibitors by machine learning methods

Area of Science:

• Computational chemistry
• Medicinal chemistry
• Machine learning

Background:

• Factor Xa (FXa) inhibitors are crucial anticoagulants for thrombotic diseases.
• Existing computational methods for predicting FXa inhibitors have limitations in accuracy and compound diversity.
• Improved prediction models are needed for developing novel antithrombotic agents.

Purpose of the Study:

• To evaluate and enhance machine learning methods for predicting Factor Xa (FXa) inhibitors.
• To assess prediction performance using a diverse set of compounds and optimized molecular descriptors.
• To identify the most effective machine learning model for distinguishing FXa inhibitors and non-inhibitors.

Main Methods:

• Utilized diverse datasets of 1098 compounds (360 inhibitors, 738 non-inhibitors).
• Employed feature selection to identify relevant molecular descriptors.
• Tested multiple machine learning algorithms: C4.5 decision tree, k-nearest neighbor, probabilistic neural network, and support vector machine (SVM).

Main Results:

• Achieved high prediction accuracies: 89.1-97.5% for inhibitors and 92.3-98.1% for non-inhibitors.
• Support vector machine (SVM) demonstrated superior performance with 98.1% accuracy for inhibitors and 94.6% for non-inhibitors.
• The study confirmed the utility of machine learning for predicting FXa inhibitors on a larger, more diverse compound set.

Conclusions:

• Machine learning methods, particularly SVM, are highly effective for predicting Factor Xa (FXa) inhibitors.
• The enhanced approach improves prediction accuracy and applicability to a broader range of chemical compounds.
• This facilitates the discovery and development of new anticoagulant drugs for thrombotic conditions.