simDEF

simDEF measures semantic similarity of Gene Ontology (GO) terms by constructing and comparing definition vectors derived from GO textual definitions to assess functional similarity among proteins.

Key Features:

• Definition-Based Similarity: simDEF uses the Gloss Vector measure from natural language processing to construct optimized definition vectors for GO terms.
• Cosine Similarity Calculation: simDEF computes semantic similarity between proteins by calculating the cosine of the angle between their respective definition vectors.
• Validation and Performance: When validated on a yeast reference database, simDEF enhanced sequence homology correlation by up to 50% versus traditional measures, improved gene expression correlation by more than 4% in the biological process hierarchy, and increased PPI predictability F1 score by more than 2.5% in the molecular function hierarchy.

Scientific Applications:

• Function Prediction: simDEF provides a definition-based functional similarity measure to improve protein function prediction.
• Protein-Protein Interaction Evaluation: simDEF improves prediction and evaluation of PPIs, including observed F1 score gains in the molecular function hierarchy.
• Gene Expression Correlation: simDEF yields improved correlation with gene expression data within the biological process hierarchy of GO.

Methodology:

Build definition vectors for GO terms using natural language processing (Gloss Vector) and compare these vectors using cosine similarity to assess semantic similarity.