Multi-label literature classification based on the Gene Ontology graph

Area of Science:

• Bioinformatics
• Computational Biology
• Text Mining

Background:

• The Gene Ontology (GO) provides a structured vocabulary for gene and protein functions.
• Manual GO annotation struggles to keep pace with biomedical literature growth.
• Automated text mining is crucial for efficient GO annotation.

Purpose of the Study:

• To enhance automatic multi-label classification of biomedical literature for GO annotation.
• To investigate graph-based algorithms leveraging GO structure for improved accuracy.

Main Methods:

• Evaluated three graph-based multi-label classification algorithms (one novel stochastic, two hierarchical).
• Compared graph-based methods against a conventional flat multi-label algorithm.
• Utilized the Gene Ontology graph structure to inform classification.

Main Results:

• Graph-based methods significantly improved prediction of GO terms from text.
• These methods enhance the accuracy of automatic protein annotation.
• Classifiers suggested closely related GO annotations, aiding curators.

Conclusions:

• Graph-based multi-label classification offers superior potential for literature-based protein annotation.
• Leveraging GO structure improves upon conventional flat classification methods.
• This approach facilitates more efficient and accurate GO annotation.