NetGO

NetGO predicts protein functions by integrating sequence data and protein-protein interaction networks to improve large-scale automated function prediction (AFP) framed as multi-label classification of gene ontology terms.

Key Features:

• Integration of Sequence and Network Information: NetGO employs a learning-to-rank framework to combine sequence-based features with protein-protein interaction networks for multi-label function prediction.
• Extensive STRING network usage: NetGO leverages STRING network data covering over 2000 species to incorporate cross-species interaction information into predictions.
• Homology transfer: NetGO applies homology transfer to propagate network-derived information to proteins not directly present in the STRING database.
• GOLabeler-based enhancement: NetGO builds on GOLabeler results from CAFA3 by adding network-derived features to improve predictive performance.
• CAFA3-style evaluation: NetGO was evaluated using the same time-delayed settings as CAFA3 and demonstrated improved performance relative to GOLabeler and other AFP methods.

Scientific Applications:

• Large-scale protein function prediction: Enables genome- and proteome-wide assignment of gene ontology terms for proteins using combined sequence and network information.
• Systems biology analyses: Supports inference of functional relationships and network-contextualized annotations across species.
• Disease mechanism investigation: Assists in identifying protein functions relevant to disease processes through network-informed annotation.
• Drug target identification: Aids prioritization of candidate drug targets by providing function annotations informed by interaction networks.
• Evolutionary and comparative genomics: Facilitates cross-species functional inference and comparative annotation via STRING's multi-species networks.

Methodology:

NetGO uses a learning-to-rank machine learning framework that integrates sequence-based features with protein-protein interaction networks from STRING (>2000 species), applies homology transfer for proteins absent from STRING, builds on GOLabeler, and was evaluated using CAFA3 time-delayed settings.