DeepGraphGO

DeepGraphGO predicts protein functions across multiple species using an end-to-end graph neural network that integrates protein sequence data and high-order protein network information for large-scale multi-label automated function prediction.

Key Features:

• Graph neural network: End-to-end graph neural network that integrates protein sequence representations with high-order protein network information.
• Input data: Combines protein sequence data and high-order protein network information as model inputs.
• Multispecies training: Single model trained across multiple species to aggregate samples and increase training data volume.
• Multi-label classification: Frames automated function prediction (AFP) as a large-scale multi-label classification problem.
• Benchmark performance: Demonstrated superior performance relative to DeepGOPlus, GeneMANIA, deepNF, and clusDCA on large-scale datasets.
• Ensemble integration: Incorporated into the NetGO ensemble method to further improve AFP performance.
• Challenging-protein prediction: Improved predictive accuracy for proteins that are difficult to classify.

Scientific Applications:

• Multispecies protein function prediction: Predicting Gene Ontology annotations and protein functions across multiple species using a single model.
• Automated Function Prediction (AFP) at scale: Large-scale AFP for proteomes framed as multi-label classification tasks.
• Prediction of challenging proteins: Function prediction for proteins that are typically hard to classify by sequence-only or species-specific network methods.
• Ensemble AFP workflows: Use within ensemble methods such as NetGO to enhance overall AFP performance.

Methodology:

End-to-end graph neural network that integrates protein sequence and high-order protein network information, trained as a single multispecies model on aggregated samples and evaluated on large-scale datasets against DeepGOPlus, GeneMANIA, deepNF, and clusDCA.