Drug-target interactions prediction using marginalized denoising model on heterogeneous networks

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug-target interactions (DTIs) are crucial for drug discovery and repositioning.
• Computational methods offer a cost-effective alternative to experimental approaches for identifying DTIs.
• Integrating diverse biological data and managing sparse DTIs remain significant challenges.

Purpose of the Study:

• To propose a novel computational method for predicting drug-target interactions (DTIs).
• To enhance the accuracy of drug candidate recommendations and existing drug repositioning.
• To address the challenges of data sparsity and integration in DTI prediction.

Main Methods:

• Developed a marginalized denoising model on heterogeneous networks.
• Incorporated an association index kernel matrix to calculate drug-target sharing relationships.
• Utilized latent global associations to mitigate false positives from network link sparsity.

Main Results:

• The proposed method achieved superior performance on benchmark and compiled datasets.
• Demonstrated higher scores in Area Under the Curve (AUC) compared to existing methods.
• Showcased larger values in Area Under Precision-Recall Curve (AUPR), indicating improved prediction efficacy.

Conclusions:

• The novel method's performance is attributed to the association index kernel matrix and latent global associations.
• Effectively addresses network link sparsity and improves the reliability of DTI predictions.
• Offers a valuable computational approach for identifying new drug candidates and repositioning existing drugs.