Automatic categorization of diverse experimental information in the bioscience literature

Area of Science:

• Bioinformatics
• Computational Biology
• Data Curation

Background:

• Biocuration involves extracting experimental information from scientific literature into computable biological knowledge databases.
• A critical bottleneck in biocuration is the manual identification of relevant papers for specific data types, which is time-consuming.
• Existing methods require extensive manual review, leading to delays in knowledge base updates.

Purpose of the Study:

• To develop an automated method for identifying scientific papers relevant to specific biocuration data types.
• To reduce the time and labor involved in the manual screening of literature for biocuration.

Main Methods:

• Utilized the machine learning algorithm Support Vector Machine (SVM) for automated classification of scientific papers.
• Developed a procedure to leverage training papers from diverse literature sources for improved identification of low-occurrence data types.
• Implemented a system for automatic categorization of experimental data types.

Main Results:

• Successfully tested the SVM-based method on ten data types from WormBase, fifteen from FlyBase, and three from Mouse Genomics Informatics (MGI).
• The system is operational at WormBase, automatically associating new publications with ten data types.
• The method demonstrates applicability across diverse experimental data types and literature corpora.

Conclusions:

• The developed automated methods are effective for a variety of data types with training sets of several hundred to a few thousand documents.
• The system is fully automatic and can be integrated into various literature-based database workflows.
• This approach significantly contributes to automating the labor-intensive biocuration process, enhancing efficiency in biological knowledge management.