PFresGO

PFresGO predicts protein functions and identifies functionally important residues by integrating Gene Ontology (GO) hierarchies with self-attention and cross-attention deep-learning models to improve functional annotation accuracy.

Key Features:

• Integration of Gene Ontology (GO) Hierarchies: Incorporates hierarchical structures within Gene Ontology (GO) graphs to model relationships among biological processes, molecular functions, and cellular components.
• Attention Mechanism Utilization: Uses self-attention operations to model inter-relationships among GO terms and dynamically update embeddings of GO terms and proteins.
• Cross-Attention for Latent Space Projection: Applies cross-attention operations to project protein representations and GO embeddings into a shared latent space to capture global sequence patterns and local functional residues.
• Identification of Functionally Important Residues: Analyzes attention weight distributions to pinpoint functionally significant residues within protein sequences.

Scientific Applications:

• Protein function annotation across GO categories: Annotates protein functions across Gene Ontology categories and enables performance comparison with existing methods.
• Identification of functional residues and domains: Identifies critical residues and potential functional domains by interpreting attention-derived residue importance.

Methodology:

PFresGO employs self-attention and cross-attention operations adapted from natural language processing and deep-learning techniques to update embeddings, project protein and GO embeddings into a shared latent space, and analyze attention weight distributions to identify functionally important residues.