A Drug-Target Network-Based Supervised Machine Learning Repurposing Method Allowing the Use of Multiple Heterogeneous...

Area of Science:

• Computational biology
• Cheminformatics
• Machine learning

Background:

• Drug-target interactions are crucial for pharmaceutical innovation, drug lead discovery, and drug repositioning.
• In silico methods, particularly kernel-based machine learning, are widely used for identifying novel drug-target interactions.
• Selecting appropriate kernel functions and parameters significantly impacts classifier performance.

Purpose of the Study:

• To present the Kronecker regularized least squares with multiple kernel learning (KronRLS-MKL) algorithm.
• To demonstrate the integration of heterogeneous information sources into a unified chemogenomic space.
• To predict new drug-target interactions using the KronRLS-MKL method.

Main Methods:

• Utilizing multiple kernel learning (MKL) to combine various kernels, enabling the integration of multiple biological data sources.
• Implementing the KronRLS-MKL algorithm for predicting drug-target interactions.
• Describing data acquisition from heterogeneous sources and the practical application of KronRLS-MKL.

Main Results:

• KronRLS-MKL effectively integrates diverse biological data to create a comprehensive chemogenomic space.
• The method facilitates the prediction of novel and existing drug-target interactions.
• Successful application of KronRLS-MKL demonstrates its utility in drug discovery pipelines.

Conclusions:

• KronRLS-MKL is a powerful machine learning tool for predicting drug-target interactions by integrating heterogeneous data.
• The approach enhances the ability to discover new therapeutic applications and optimize drug development.
• This method offers a robust framework for leveraging multi-source biological information in drug discovery.