Supporting the curation of biological databases with reusable text mining.

Area of Science:

• Bioinformatics
• Computational Biology
• Biomedical Informatics

Background:

• Manual curation of biological databases from scientific publications is labor-intensive and costly.
• Machine learning (ML) can streamline literature filtering for curators, but requires user-friendly tools.
• Widespread adoption of ML in curation depends on adaptable, intuitive software solutions.

Purpose of the Study:

• To propose a novel document categorization method for supporting biological database curators.
• To describe the architecture of a curator-oriented tool implementing this method without requiring programming expertise.
• To demonstrate the method's feasibility by applying it to identify allergen cross-reactivity information in PubMed abstracts.

Main Methods:

• Developed a curator-oriented tool using document categorization techniques.
• Implemented the method for a real-world curation task: identifying allergen cross-reactivity data.
• Evaluated two machine learning classifier algorithms: Classification and Regression Trees (CART) and Artificial Neural Networks (ANN).
• Tested performance using composite and single-word features with various scoring functions.

Main Results:

• Both CART and ANN classifiers surpassed predefined performance targets.
• The Artificial Neural Network (ANN) classifier achieved the highest performance.
• The method demonstrated effectiveness in identifying specific biomedical information relevant to curation.

Conclusions:

• The proposed method and tool can significantly assist biological database curators.
• Machine learning, when implemented in intuitive tools, can effectively reduce manual curation workload.
• The system shows promise for improving the efficiency and accuracy of biomedical literature curation.