Integrating Similarities via Local Interaction Consistency and Optimizing Area Under the Curve Measures via Matrix...

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Experimental identification of drug-target interactions (DTIs) is costly and time-consuming.
• Computational methods accelerate drug discovery by predicting potential DTIs.
• Existing methods for fusing drug and target similarities often overlook neighbor interaction consistency.

Purpose of the Study:

• To develop a novel similarity integration method for drug-target interaction prediction.
• To incorporate local interaction consistency (LIC) into the fusion of heterogeneous drug and target data.
• To optimize drug-target interaction prediction models using Area Under the Precision-Recall Curve (AUPR) and Area Under the Receiver Operating Characteristic Curve (AUC) as direct optimization objectives.

Main Methods:

• Proposed a Local Interaction Consistency (LIC) aware similarity integration method.
• Developed two matrix factorization (MF) models optimizing AUPR and AUC via convex surrogate losses.
• Created an ensemble MF approach combining the strengths of single-metric MF models.

Main Results:

• The proposed LIC-aware similarity integration method improves DTI prediction.
• The novel MF models demonstrated superior performance in optimizing AUPR and AUC compared to existing methods.
• The ensemble MF approach effectively leveraged both AUPR and AUC for enhanced DTI prediction accuracy.

Conclusions:

• The developed methods offer a more reliable and efficient approach to predicting drug-target interactions.
• The Local Interaction Consistency (LIC) aware similarity integration is crucial for accurate DTI prediction.
• The optimization of AUPR and AUC directly within matrix factorization models leads to improved performance and discovery of potential new DTIs.