springD2A: capturing uncertainty in disease-drug association prediction with model integration

Area of Science:

• Computational biology
• Pharmacology
• Bioinformatics

Background:

• Drug repositioning is a vital strategy for efficient drug discovery, identifying new uses for existing medications.
• Traditional methods for predicting disease-drug associations often construct negative sets from unknown pairs, potentially overlooking valid associations.
• Existing negative pairs in drug repositioning models may contain true associations, but are often ignored, limiting predictive accuracy.

Purpose of the Study:

• To develop a novel method, springD2A, for predicting disease-drug associations that accounts for uncertainty in negative pairs.
• To improve the accuracy of drug repositioning models by better discriminating between confirmed positive and uncertain negative disease-drug pairs.

Main Methods:

• Introduced a spring-like penalty in the loss function for negative pairs, adjusting based on their distance in a unit sphere.
• Implemented a sequential sampling strategy where unknown disease-drug pairs are sampled as negatives proportionally to their predicted positive scores.
• Utilized ensemble schemes (parameter- and feature-based) by learning multiple models during sequential sampling to enhance performance.

Main Results:

• The springD2A method demonstrated effectiveness in capturing uncertainty within negative disease-drug pairs.
• springD2A showed improved discrimination between reliable positive pairs and uncertain negative pairs.
• Experimental results confirmed springD2A as an effective tool for advancing drug repositioning efforts.

Conclusions:

• The springD2A method offers a robust approach to handling uncertainty in negative data for disease-drug association prediction.
• This method has the potential to significantly improve the efficiency and success rate of drug repositioning strategies.
• The developed method and associated datasets are publicly available, facilitating further research in drug discovery.