Identifying Complementary Therapeutic Relationships for Drug Repurposing via Spectral-Spatial Graph Contrastive...

Area of Science:

• Computational biology
• Pharmacology
• Artificial intelligence

Background:

• Drug repurposing accelerates novel therapy discovery by finding new uses for existing drugs.
• Current deep learning methods often assume drug similarity equates to therapeutic similarity, limiting the identification of functionally diverse yet therapeutically relevant drugs.
• Successful repurposing cases highlight the need to model both drug interaction and dissimilarity attributes.

Purpose of the Study:

• To propose MPGCL, a unified spectral-spatial graph contrastive learning framework.
• To identify repurposed drugs that exhibit complementary therapeutic relationships with existing treatments.
• To uncover "metformin-like" drug repurposing discoveries by modeling functional diversity.

Main Methods:

• MPGCL utilizes a spectral-spatial graph contrastive learning approach.
• A mid-pass spectral filter extracts drug dissimilarity signals.
• A low-pass spatial filter captures interaction-aware features, enhanced by a contrastive learning objective to differentiate drug profiles.

Main Results:

• MPGCL consistently outperforms existing drug repurposing methods on real-world benchmarks.
• The framework effectively identifies promising drug candidates with dissimilarity attributes for combination therapy.
• Case studies, particularly in breast cancer, validate the model's predictive capabilities.

Conclusions:

• MPGCL provides a robust framework for discovering novel therapeutic relationships.
• The model moves beyond conventional similarity-based paradigms in drug repurposing.
• This approach enhances the identification of functionally diverse drugs for improved therapeutic outcomes.