PANDA2: protein function prediction using graph neural networks.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• High-throughput sequencing generates vast protein data, necessitating efficient functional annotation methods.
• Current protein annotation relies on slow, costly experiments, creating a need for computational solutions.
• Integrating Gene Ontology (GO) hierarchical structures into machine learning for protein function prediction is challenging.

Purpose of the Study:

• To develop PANDA2, a deep learning system for accurate and rapid protein function prediction.
• To leverage graph neural networks and transformer models for enhanced functional inference.
• To provide a publicly accessible tool for protein function annotation.

Main Methods:

• Utilized graph neural networks to model the Gene Ontology directed acyclic graph (DAG) topology.
• Integrated features from transformer protein language models for comprehensive sequence analysis.
• Developed a deep learning system named PANDA2.

Main Results:

• PANDA2 achieved top rankings in the CAFA3 benchmark, excelling in Cellular Component Ontology (CCO) and Biological Process Ontology (BPO).
• The system demonstrated superior performance compared to leading predictors like DeepGOPlus, GOLabeler, and DeepText2GO on an independent dataset.
• PANDA2 secured first place in CCO and BPO, and second in Molecular Function Ontology (MFO) across benchmarks.

Conclusions:

• PANDA2 offers a highly accurate and efficient computational approach for protein function prediction.
• The system's performance surpasses current state-of-the-art methods, addressing limitations of experimental annotation.
• PANDA2 provides a valuable, freely accessible resource for the scientific community to accelerate functional genomics research.