GoVec: Gene Ontology Representation Learning Using Weighted Heterogeneous Graph and Meta-Path.

Area of Science:

• Bioinformatics
• Computational Biology
• Life Sciences

Background:

• Biomedical knowledge graphs are essential for data-intensive life science and healthcare applications.
• Current Gene Ontology (GO) representation learning methods are limited to homogeneous graphs, failing to capture diverse node types and relationships.
• Heterogeneous graphs offer a solution by integrating ontology terms and biomedical entities.

Purpose of the Study:

• To introduce GoVec, a novel method for learning representations in heterogeneous biomedical graphs.
• To enable seamless representation learning for both ontologies and biological entities.
• To enhance bioinformatics applications, particularly semantic similarity and relation extraction.

Main Methods:

• GoVec utilizes meta-path-based representation learning within a heterogeneous graph framework.
• It generates vector embeddings that capture relationships between different types of nodes.
• The method was evaluated by comparing semantic similarities with expert-defined similarities and through downstream task performance.

Main Results:

• GoVec successfully produces representations for both ontologies and biological entities.
• The generated vectors improve semantic similarity calculations and relation extraction.
• Evaluations demonstrated GoVec's superiority over state-of-the-art methods in protein-protein interaction and protein family similarity tasks.
• Qualitative analysis showed GoVec embeddings visually separate protein families based on functional semantics.

Conclusions:

• GoVec effectively addresses the limitations of homogeneous graph approaches in Gene Ontology representation learning.
• The method provides valuable vector representations for diverse bioinformatics applications.
• GoVec demonstrates significant potential for advancing computational biology and data-driven biomedical research.