GO2Vec

GO2Vec generates vector representations of Gene Ontology (GO) terms and proteins to capture semantic relationships for computing functional similarity and predicting protein-protein interactions.

Key Features:

• Graph embedding: Uses graph embedding techniques to derive vector representations directly from the structure of the GO graph.
• Integration of GO data: Combines the GO graph and GO annotations to capture comprehensive semantic information of GO terms and annotated proteins.
• Term and protein vectors: Produces vector representations for GO terms and associated proteins that encode functional properties and interrelations.
• Semantic similarity capability: Encodes semantic relationships that support computation of functional similarity between proteins.
• Interaction prediction capability: Enables prediction of protein-protein interactions using the generated vectors.
• Benchmark performance: Demonstrated superior performance relative to information content-based measures and word embedding approaches in benchmark evaluations.

Scientific Applications:

• Functional similarity calculation: Computing functional similarity between proteins using shared vector representations of GO annotations.
• Protein-protein interaction prediction: Predicting potential protein-protein interactions by modeling proteins in a shared embedding space.

Methodology:

Combines the GO graph and GO annotations and applies graph embedding techniques to transform GO terms and associated proteins into vector representations; evaluated using the CESSM dataset for functional similarity and Yeast and Human datasets from STRING for protein-protein interaction prediction, with comparisons to information content-based measures and word embedding approaches.