The Computational Drug Repositioning Without Negative Sampling.

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Computational drug repositioning accelerates drug development but faces challenges with unvalidated associations and limited feature representation.
• Existing models often use invalid negative sampling and inner products, hindering accuracy.

Purpose of the Study:

• To propose a novel framework, PUON (Positive and Unlabelled Outer Neighborhood), for computational drug repositioning.
• To address limitations of existing methods by avoiding negative sampling and improving feature information modeling.

Main Methods:

• Developed the PUON framework utilizing validated (Positive) and unvalidated (Unlabelled) drug-disease associations.
• Incorporated an Outer Neighborhood-based classifier to model cross-feature information within latent factors.
• Compared PUON against 6 popular baseline methods.

Main Results:

• PUON demonstrated superior performance across 6 evaluation metrics in extensive experiments.
• The framework effectively models drug-disease associations without relying on negative sampling.
• Outer Neighborhood-based classification improved the utilization of latent factor features.

Conclusions:

• The PUON framework offers a more robust and accurate approach to computational drug repositioning.
• This method overcomes key limitations of previous techniques, paving the way for more efficient drug discovery.
• PUON's performance suggests significant potential for accelerating the identification of new drug-disease relationships.