NetGO 3.0

NetGO 3.0 predicts protein functions to improve automated function prediction (AFP) for proteins by integrating Evolutionary Scale Modeling (ESM)-1b embeddings and logistic regression within the NetGO framework.

Key Features:

• Multi-source integration: Combines information from multiple data sources to enhance functional annotation at scale.
• ESM-1b embeddings: Uses Evolutionary Scale Modeling (ESM)-1b protein language model embeddings as primary sequence-derived representations.
• Self-supervised representations: Leverages embeddings derived from self-supervised learning applied to protein sequences to capture informative features for unannotated proteins.
• LR-ESM model: Trains a logistic regression (LR) model on ESM-1b embeddings, referred to as LR-ESM.
• Performance parity: Demonstrates performance on par with the best-performing components of NetGO 2.0.
• Framework integration: Integrates LR-ESM into the existing NetGO 2.0 framework to extend prediction capability.

Scientific Applications:

• Automated function prediction (AFP): Supports large-scale prediction of Gene Ontology and other protein function annotations.
• Annotation of uncharacterized proteins: Extracts informative sequence representations for proteins lacking experimental annotations.
• Improved prediction accuracy: Enhances the accuracy and comprehensiveness of protein function predictions compared to prior NetGO components.

Methodology:

Each protein is represented using ESM-1b embeddings derived from self-supervised learning on protein sequences, and a logistic regression model (LR-ESM) is trained on those embeddings and integrated into the NetGO 2.0 framework.