A subspace learning aided matrix factorization for drug repurposing

Area of Science:

• Computational Biology
• Bioinformatics
• Drug Discovery

Background:

• Drug development is costly and time-consuming.
• Machine learning and computational biology offer advanced drug repositioning techniques.
• Improved synergy is needed to enhance predictive accuracy and clinical application.

Purpose of the Study:

• To present a novel approach integrating sparse subspace learning (SLSDR) and a dual-domain drug repurposing method (iDrug).
• To enhance the iDrug objective function using SLSDR for improved feature extraction from drug-disease and drug-target data.
• To develop a holistic solution for optimizing drug repositioning predictions.

Main Methods:

• Integration of SLSDR, a feature selection technique, with the iDrug method for drug repurposing.
• Utilizing matrix factorization for both subspace learning and the iDrug approach.
• Constructing drug-drug, target-target, and disease-disease similarity matrices based on SLSDR-derived features.
• Introducing a novel objective function to capture complex drug-disease interactions.

Main Results:

• The integrated approach offers notable gains in prediction accuracy (AUC, AUPR) and computing efficiency.
• Demonstrated superior performance compared to state-of-the-art drug repurposing methods.
• Preserves data geometry in both feature and sample spaces.

Conclusions:

• The proposed matrix factorization method enhances drug repurposing by integrating drug-disease and drug-target domain knowledge.
• Achieves improved accuracy in drug repurposing compared to existing state-of-the-art methods.