Classification of Aurora B kinase inhibitors using computational models.

Area of Science:

• Medicinal Chemistry
• Computational Chemistry
• Drug Discovery

Background:

• Aurora B kinase is a key target in cancer therapy.
• Developing selective inhibitors is crucial for effective treatment.
• Predictive models can accelerate the discovery of novel inhibitors.

Purpose of the Study:

• To build and evaluate classification models for predicting Aurora B kinase inhibitor activity.
• To identify important molecular descriptors for Aurora B kinase inhibition.
• To leverage machine learning for efficient drug discovery.

Main Methods:

• Utilized Self-Organizing Map (SOM) and Support Vector Machine (SVM) algorithms.
• Developed four classification models based on selected molecular descriptors and fingerprints.
• Employed a dataset of 679 Aurora B kinase inhibitors, split into training and test sets.

Main Results:

• All four models demonstrated high prediction accuracy (>87%) on the independent test set.
• Combined use of ADRIANA.Code descriptors and MACCS fingerprints improved predictive performance.
• Identified hydrogen-bonding and atom charge descriptors as critical for Aurora B kinase inhibitor bioactivity.

Conclusions:

• Machine learning models can effectively predict Aurora B kinase inhibitor activity.
• A combination of diverse molecular descriptors enhances model accuracy.
• Understanding key molecular properties guides the design of more potent Aurora B kinase inhibitors.