Predicting drug-target interactions using matrix factorization with self-paced learning and dual similarity...

Area of Science:

• Computational biology
• Drug discovery
• Bioinformatics

Background:

• Drug repositioning (DR) identifies new uses for existing drugs, reducing costs and risks.
• Traditional DR methods are slow, expensive, and have high failure rates.
• Existing computational methods, including matrix factorization (MF), face challenges with noisy, incomplete data and limited learning capacity.

Purpose of the Study:

• To develop an improved computational method for drug repositioning.
• To address limitations of current matrix factorization methods in handling data noise and insufficient similarity information.
• To enhance the accuracy and efficiency of predicting potential drug-target interactions.

Main Methods:

• Proposed a novel approach: self-paced learning with dual similarity information and matrix factorization (SPLDMF).
• Integrated self-paced learning to mitigate issues caused by data noise and missing values.
• Incorporated dual similarity information to enhance the model's learning capacity and predictive accuracy.

Main Results:

• The SPLDMF model demonstrated superior performance across multiple benchmark and extended datasets.
• Achieved high predictive accuracy, with Area Under the ROC Curve (AUC) of 0.982 and Precision-Recall Curve (PRC) of 0.815.
• Outperformed state-of-the-art methods in predicting drug-target interactions.

Conclusions:

• The proposed SPLDMF approach is effective for predicting drug-target interactions.
• The method successfully overcomes challenges associated with noisy and incomplete data in drug repositioning.
• SPLDMF offers a promising computational strategy for accelerating drug discovery and development.