DeepGraphGO: graph neural network for large-scale, multispecies protein function prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Automated protein function prediction (AFP) is a critical multi-label classification task.
• Existing network-based AFP methods suffer from species-specific model training and neglect protein sequence information, limiting their performance.
• There is a need for powerful network-based AFP methods that overcome these limitations.

Purpose of the Study:

• To develop an end-to-end, multispecies graph neural network-based method for AFP.
• To leverage both protein sequence and high-order protein network information for improved prediction accuracy.
• To create a single model capable of performing AFP across multiple species.

Main Methods:

• Proposed DeepGraphGO, a novel graph neural network architecture.
• Implemented a multispecies strategy enabling a single model for all species.
• Utilized both protein sequence and network topology data.

Main Results:

• DeepGraphGO significantly outperformed state-of-the-art methods, including DeepGOPlus, GeneMANIA, deepNF, and clusDCA.
• Demonstrated the effectiveness of the multispecies strategy, leading to a larger training sample size.
• Showcased improved performance on challenging proteins and enhanced results when integrated into the NetGO ensemble method.

Conclusions:

• DeepGraphGO represents a significant advancement in network-based AFP.
• The multispecies approach and integration of sequence/network data offer a powerful solution for AFP.
• DeepGraphGO provides a robust and scalable method for predicting protein functions across diverse species.