Balancing Gene Ontology annotation specificity in protein function prediction based on the protein sequence large...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Accurate protein function prediction is crucial for drug discovery and precision medicine.
• Gene Ontology (GO) provides a standardized framework, but an imbalance between low- and high-specificity terms hinders comprehensive analysis.
• This imbalance creates blind spots in understanding protein function, especially in clinically relevant pathways.

Purpose of the Study:

• To present ProGO-PSL, a novel large graph architecture designed to resolve the imbalance between low- and high-specificity GO terms.
• To improve the accuracy and completeness of protein function prediction.
• To enable a more thorough functional characterization of the proteome.

Main Methods:

• ProGO-PSL leverages explicit domain identifiers from InterPro and evolutionary context from Multiple Sequence Alignments.
• A powerful imbalance learning framework is employed to fuse these complementary data sources.
• The model utilizes a large graph architecture for integrated data analysis.

Main Results:

• ProGO-PSL consistently outperforms state-of-the-art methods by 5-15% across all specificity levels.
• The model demonstrates robust generalization on both benchmark and independent test datasets.
• Interpretable representations are generated, clarifying relationships between GO terms of varying specificity.

Conclusions:

• ProGO-PSL effectively addresses the challenge of GO term imbalance in protein function prediction.
• The approach enables more complete functional characterization of proteins and the proteome.
• This work accelerates the identification of therapeutic targets in uncharacterized biological pathways.