Supervised biological network alignment with graph neural networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Massive proteins with known sequences remain functionally unannotated despite sequencing advances.
• Biological network alignment (NA) transfers functional knowledge across species by finding node correspondence in protein-protein interaction (PPI) networks.
• Traditional NA assumed topological similarity implies functional similarity, but this is not always true, necessitating a data-driven approach.

Purpose of the Study:

• To introduce GraNA, a deep learning framework for supervised NA.
• To address the pairwise NA problem by learning protein representations and predicting functional correspondence.
• To integrate multi-faceted non-functional data as anchor links to guide cross-species protein mapping.

Main Methods:

• GraNA employs graph neural networks (GNNs) for supervised NA.
• It utilizes within-network interactions and across-network anchor links (e.g., sequence similarity, orthologs) for learning.
• Predicts functional correspondence between proteins in different species.

Main Results:

• GraNA accurately predicted functional relatedness and robustly transferred annotations across species.
• Outperformed existing NA methods on a benchmark dataset.
• Successfully identified functionally replaceable human-yeast protein pairs in a case study.

Conclusions:

• GraNA provides an effective deep learning framework for supervised biological network alignment.
• It enables accurate functional annotation transfer across species by leveraging diverse data sources.
• The method demonstrates significant improvements over traditional NA approaches.