GraphBAN: An inductive graph-based approach for enhanced prediction of compound-protein interactions

Area of Science:

• Computational chemistry
• Bioinformatics
• Drug discovery

Background:

• Compound-protein interactions are vital for understanding molecular mechanisms and therapeutic potential in drug discovery.
• Traditional methods often struggle with predicting interactions involving entirely new compounds or proteins.

Purpose of the Study:

• To introduce GraphBAN, a graph-based framework for inductive prediction of compound-protein interactions.
• To enable robust prediction of interactions involving previously unseen compounds and proteins.

Main Methods:

• Utilized a graph-based framework (GraphBAN) for inductive link prediction.
• Employed a knowledge distillation architecture with teacher (network structure) and student (node attributes) blocks.
• Incorporated a domain adaptation module to enhance cross-dataset effectiveness.

Main Results:

• GraphBAN demonstrated superior performance against ten baseline models across five benchmark datasets.
• Empirical tests confirmed GraphBAN's effectiveness in predicting interactions for novel compounds and proteins.
• A case study involving the Pin1 protein validated the model's real-world applicability.

Conclusions:

• GraphBAN offers a robust and effective solution for predicting compound-protein interactions, particularly for novel entities.
• The framework's inductive capabilities and domain adaptation enhance its utility in early drug discovery.
• GraphBAN represents a promising tool for accelerating the identification of potential drug candidates.