DomFun

DomFun predicts protein functions by leveraging domain-based associations to assign functional annotations from Gene Ontology (GO), KEGG, Reactome and the Human Phenotype Ontology (HPO) without relying on sequence analysis.

Key Features:

• Domain-Based Prediction: Uses associations between protein domains and functional annotations to predict protein function independent of sequence data.
• Tripartite Network Analysis: Employs a tripartite network connecting homologous superfamily and FunFam domains from CATH-Gene3D with functional annotations to calculate domain–function associations using multiple indices.
• Functional Annotation Integration: Integrates GO molecular functions and biological processes, KEGG pathways, Reactome pathway terms, and HPO annotations for prediction output.
• Performance Validation: Validated using the CAFA 2 benchmark for GO annotations and a custom Pathway Prediction Performance (PPP) procedure for KEGG and Reactome annotations.
• Comparative Performance: Shows superior performance with FunFams compared to superfamilies, with stronger results for GO molecular functions than biological processes.
• Association and Aggregation Methods: Identifies Simpson's index for domain–function associations combined with Stouffer's method for aggregation as an effective approach.

Scientific Applications:

• Protein Function Prediction: Predicts functions of uncharacterized proteins to support proteomics and functional genomics studies.
• Functional Annotation and Pathway Analysis: Provides integrated annotations to support pathway-level interpretation using KEGG and Reactome terms.
• Phenotype Association: Supports phenotype association studies by mapping domain associations to HPO annotations.

Methodology:

Uses tripartite network analysis connecting homologous superfamily and FunFam domains from CATH-Gene3D with annotations from Gene Ontology (GO), KEGG, Reactome and HPO; computes domain–function associations using multiple indices including Simpson's index; aggregates associations using Stouffer's method; predictions are validated against CAFA 2 and a custom Pathway Prediction Performance (PPP) benchmark; prediction is based on domain associations without sequence analysis.