PredGO

PredGO predicts Gene Ontology (GO) functional annotations for proteins by integrating AlphaFold-predicted three-dimensional structures with sequence, interaction, and expression data.

Key Features:

• Integration of Structural Information: Utilizes three-dimensional structural data predicted by AlphaFold to increase the number of proteins with available structural information for functional annotation.
• Heterogeneous Feature Fusion: Employs a pre-trained language model alongside geometric vector perceptrons and attention mechanisms to extract and fuse features including sequence homology, protein-protein interactions, gene co-expression, and other non-structural clues.
• High Coverage and Accuracy: Computational results demonstrate superior performance in predicting protein functions compared to existing approaches, attributed to combining predicted structural information with non-structural data.
• Extensive Annotation Capability: Has annotated over 205,000 UniProt entries for human proteins, with approximately 90% of annotations based on predicted structures.

Scientific Applications:

• Functional Genomics: Enabling interpretation of gene function and interaction networks through GO term prediction.
• Proteomics Research: Enhancing proteome annotation by providing functional insights derived from structural and non-structural data fusion.
• Drug Discovery: Supporting identification of potential drug targets by elucidating protein functions and interactions.

Methodology:

Combines AlphaFold-predicted structures with sequence homology, protein-protein interactions, gene co-expression and other non-structural clues using a pre-trained language model, geometric vector perceptrons, and attention mechanisms.