GTPLM-GO: Enhancing Protein Function Prediction Through Dual-Branch Graph Transformer and Protein Language Model...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein-protein interaction (PPI) networks are crucial for understanding protein functions.
• Existing graph neural network (GNN) methods struggle with over-smoothing, hindering accurate protein function prediction.
• Integrating local and global information in PPI networks remains a challenge.

Purpose of the Study:

• To propose GTPLM-GO, a novel method for protein function prediction.
• To address the limitations of GNNs in modeling PPI networks by integrating local and global information.
• To enhance protein function prediction accuracy by combining graph-based and language model-based approaches.

Main Methods:

• Developed GTPLM-GO, a dual-branch Graph Transformer and protein language model.
• Employed a graph neural network and a linear attention-based Transformer encoder for collaborative local-global information modeling.
• Integrated PPI network information with functional semantic encoding from a protein language model.

Main Results:

• GTPLM-GO effectively models both local and global information within PPI networks.
• The method successfully integrates network topology with protein sequence-derived functional semantics.
• Experimental results show GTPLM-GO outperforms existing network-based and sequence-based methods across various PPI network scales.

Conclusions:

• GTPLM-GO offers a superior approach to protein function prediction by overcoming GNN limitations.
• The dual-branch architecture effectively captures complex relationships in PPI networks.
• This method advances the field of bioinformatics by improving the accuracy and scope of protein function prediction.