GNNenrich: a novel method for pathway enrichment analysis based on graph neural network

Area of Science:

• Bioinformatics
• Computational Biology
• Network Biology

Background:

• Graph neural networks (GNNs) are increasingly used for biological networks (genes, proteins).
• Current enrichment methods often overlook crucial interaction data between biological entities.
• Existing approaches like over-representation analysis and gene set scoring have limitations in capturing network effects.

Purpose of the Study:

• To introduce GNNenrich, a novel GNN-based method for biological enrichment analysis.
• To leverage protein sequence properties and interaction networks for improved functional interpretation.
• To address limitations of traditional enrichment methods by incorporating network topology.

Main Methods:

• Developed GNNenrich, a novel graph neural network approach.
• Integrated multi-level embeddings incorporating protein sequence features and interaction networks.
• Utilized graph neural network architecture for functional relationship establishment.

Main Results:

• GNNenrich was evaluated against established methods like g:Profiler and EnrichNet.
• The method demonstrated the ability to reproduce results from existing approaches.
• GNNenrich provided novel interpretations supported by protein-protein interaction (PPI) data.

Conclusions:

• GNNenrich offers a powerful new perspective for biological enrichment analysis.
• The method effectively integrates network and sequence information for enhanced interpretation.
• GNNenrich successfully supports biological interpretation by leveraging protein-protein interactions.