Drug-target interaction prediction using Multi Graph Regularized Nuclear Norm Minimization.

Area of Science:

• Computational biology
• Bioinformatics
• Drug discovery

Background:

• Accurate identification of drug-target interactions (DTIs) is vital in pharmaceutical sciences.
• Experimental validation of DTIs is resource-intensive, necessitating efficient in-silico prediction methods.
• Existing computational methods for DTI prediction can be improved by better integration of drug and target similarity information.

Purpose of the Study:

• To develop a novel computational framework for predicting drug-target interactions.
• To enhance DTI prediction accuracy by leveraging multiple drug-drug and target-target similarity measures.
• To address the lack of consensus on optimal similarity integration strategies in DTI prediction.

Main Methods:

• Proposed a Multi-Graph Regularized Nuclear Norm Minimization (MGRNNM) framework.
• Utilized known drug-target interaction networks, drug similarities, and target similarities as inputs.
• Incorporated multiple graph Laplacian regularization terms based on diverse drug-drug and target-target similarities to capture complex relationships.

Main Results:

• The MGRNNM framework demonstrated improved predictive performance on four benchmark datasets.
• The proposed method significantly outperformed existing state-of-the-art computational DTI prediction techniques.
• Cross-validation experiments using AUPR and AUC metrics confirmed the algorithm's effectiveness.

Conclusions:

• The MGRNNM framework offers a robust and accurate approach for in-silico drug-target interaction prediction.
• Integrating multiple similarity graphs enhances the prediction of DTIs by better preserving data manifold properties.
• The developed computational method can accelerate drug discovery by efficiently narrowing down potential drug-target interactions for experimental validation.