Anc2vec: embedding gene ontology terms by preserving ancestors relationships

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• The Gene Ontology (GO) uses a controlled vocabulary for gene product functions and locations.
• Vector representations (embeddings) of GO terms capture meaningful information and improve downstream tasks.
• Existing GO term embeddings often fail to capture the GO's crucial structural features.

Purpose of the Study:

• To present anc2vec, a novel protocol for generating GO term embeddings.
• To preserve ontological uniqueness, ancestor hierarchy, and sub-ontology membership in GO term embeddings.
• To demonstrate the advantages of anc2vec embeddings on diverse biological tasks.

Main Methods:

• Developed a novel protocol, anc2vec, using neural networks.
• Constructed vector representations of GO terms by preserving ontological uniqueness, ancestor hierarchy, and sub-ontology membership.
• Systematically evaluated anc2vec on visualization, sub-ontology prediction, term inference, embedding retrieval, and protein-protein interaction prediction.

Main Results:

• anc2vec embeddings demonstrated superior performance compared to recent approaches across multiple tasks.
• Preserving structural features of the GO in embeddings leads to improved performance.
• Experimental results validated the effectiveness of anc2vec for various bioinformatics applications.

Conclusions:

• anc2vec effectively captures crucial structural features of the Gene Ontology.
• Improved representation of GO structure in embeddings enhances performance on diverse downstream tasks.
• anc2vec offers a valuable tool for bioinformatics research, particularly in functional genomics and protein interaction studies.