Learning representations for gene ontology terms by jointly encoding graph structure and textual node descriptors.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Semantic similarity of Gene Ontology (GO) terms is crucial for bioinformatics.
• Existing word embedding methods often ignore GO's structural information and term relationships.
• Developing effective GO term representations is essential for advanced functional analysis.

Purpose of the Study:

• To propose GT2Vec, a novel model for GO term representation.
• To integrate GO graph structure and semantic descriptions for improved embeddings.
• To enhance functional bioinformatics applications through better GO term vectorization.

Main Methods:

• Utilized graph contrastive learning to capture GO graph structure.
• Employed BERT encoders for semantic descriptions of GO terms.
• Developed GT2Vec by jointly encoding graph structure and textual descriptors.

Main Results:

• Evaluated GT2Vec on a protein similarity task using benchmark datasets.
• Demonstrated the effectiveness of the joint encoding approach.
• Achieved superior performance compared to existing methods by integrating structural and semantic information.

Conclusions:

• GT2Vec provides a powerful new method for GO term representation.
• Jointly encoding graph structure and semantic descriptions significantly enhances vector representations.
• The GT2Vec model offers improved capabilities for functional bioinformatics research.