Partial order relation-based gene ontology embedding improves protein function prediction.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Protein annotation is crucial for understanding biological functions.
• Gene Ontology (GO) is a standard framework for describing protein functions.
• Existing GO term embedding methods often fail to capture the full topological structure of the GO directed acyclic graph (DAG).

Purpose of the Study:

• To develop a novel GO term representation learning method that incorporates partial order relationships.
• To improve the quality of GO term embeddings for enhanced biological data analysis.
• To develop a superior protein function prediction method based on improved GO term representations.

Main Methods:

• Proposed PO2Vec, a novel method for GO term representation learning utilizing partial order relationships.
• Evaluated PO2Vec against existing embedding methods on downstream biological tasks.
• Developed PO2GO, a protein function prediction method leveraging PO2Vec embeddings.

Main Results:

• PO2Vec achieved superior performance compared to existing methods in various downstream biological tasks.
• The PO2GO method demonstrated enhanced performance across multiple metrics, including annotation specificity.
• PO2GO showed strong few-shot prediction capabilities on benchmark datasets.

Conclusions:

• High-quality representation of GO structure is critical for computational protein annotation.
• PO2Vec effectively captures topological information in GO, leading to improved embeddings.
• The PO2GO method offers a significant advancement in protein function prediction accuracy and efficiency.