GO2Vec: transforming GO terms and proteins to vector representations via graph embeddings

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Semantic similarity of Gene Ontology (GO) terms is crucial for bioinformatics.
• Traditional methods rely on information content; recent approaches use word embeddings.
• GO2Vec leverages graph embeddings for GO term vector representations.

Purpose of the Study:

• To introduce GO2Vec, a novel method for learning GO term vector representations using graph embeddings.
• To combine GO graph structure and GO annotations for improved vector learning.
• To apply learned vectors to bioinformatics tasks like protein functional similarity and interaction prediction.

Main Methods:

• Utilized graph embeddings to learn vector representations for GO terms directly from the GO graph.
• Integrated information from both the GO graph structure and GO annotations.
• Applied GO2Vec to assess protein functional similarity and predict protein-protein interactions.

Main Results:

• GO2Vec demonstrated superior performance in calculating protein functional similarity on the CESSM dataset.
• The method effectively predicted protein-protein interactions on Yeast and Human datasets from STRING.
• GO2Vec outperformed both information content-based and word embedding-based measures.

Conclusions:

• Graph embeddings are effective for learning GO term representations from GO and GOA graphs.
• GO annotations significantly contribute to computing similarity between GO terms and proteins.
• GO2Vec offers an effective approach for semantic similarity analysis in bioinformatics.