Novel artificial intelligence-based identification of drug-gene-disease interaction using protein-protein interaction

Area of Science:

• Bioinformatics
• Artificial Intelligence in Medicine
• Computational Biology

Background:

• Identifying effective drugs for diseases is challenging due to complex drug-gene-disease interactions.
• Current disease- or drug-centric approaches are complex and require extensive prior information.
• A gene-centric approach is simpler but lacks methods for identifying relevant genes without predefined diseases or drugs.

Purpose of the Study:

• To develop and evaluate a novel artificial intelligence-based approach for drug repositioning.
• To identify genes critical for diseases and effective drugs targeting them using unsupervised methods.
• To perform drug repositioning using only protein-protein interaction (PPI) data.

Main Methods:

• Applied tensor decomposition (TD)-based unsupervised feature extraction (FE) to PPI networks.
• Utilized an unsupervised, gene-centric approach without specifying target diseases or drugs.
• Evaluated the method's ability to identify known drug-target relationships and discover new ones.

Main Results:

• TD-based unsupervised FE successfully identified genes associated with cancers and their corresponding drugs using only PPI data.
• The method identified hub proteins, which also showed utility in drug repositioning, particularly for cancer drugs.
• TD-based unsupervised FE demonstrated broader applicability beyond cancer, identifying drugs for other diseases and showing potential for in vivo applications.

Conclusions:

• TD-based unsupervised FE is a powerful and versatile tool for drug repositioning.
• This AI-driven method effectively leverages PPI data for drug discovery without needing additional information.
• The approach offers advantages over traditional methods, including identifying drugs for diverse diseases and potential in vivo efficacy.