Gotrees: predicting go associations from protein domain composition using decision trees

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• The Gene Ontology (GO) provides a standardized vocabulary for protein function, but manual annotation struggles to keep pace with data generation.
• Existing automated methods, like InterPro2GO/PFAM2GO, rely on single protein domains, potentially leading to false positives and limited sensitivity for less studied proteins.

Purpose of the Study:

• To develop a more sensitive and precise automated method for assigning Gene Ontology terms to proteins.
• To improve upon existing homology-based annotation tools by considering the entire functional domain content of proteins.

Main Methods:

• Proteins were modeled using their complete set of functional domains.
• Individual decision tree classifiers were trained for each GO term based on known protein assignments.
• The novel method's performance was compared against the InterPro2GO approach.

Main Results:

• The developed method demonstrated high sensitivity, specificity, and precision, with robustness to sparse data.
• Compared to InterPro2GO, the new approach significantly improved sensitivity for Molecular Function (22%), Biological Process (27%), and Cellular (50%) GO terms.
• A slight decrease in precision was observed, but overall performance was enhanced.

Conclusions:

• Modeling proteins by their entire functional domain content offers a more sensitive approach to Gene Ontology annotation.
• This method provides a valuable tool for annotating less studied proteins and addressing the limitations of single-domain-based predictions.
• The improved sensitivity enhances the comprehensiveness of protein function databases.