Comparing usability of matching techniques for normalising biomedical named entities

Area of Science:

• Biomedical Informatics
• Computational Biology
• Natural Language Processing

Background:

• Biomedical term normalization is crucial for linking entities to reference databases.
• Accurate term normalization enhances data integration and analysis in biomedical research.
• Evaluating different string matching techniques is essential for optimizing this process.

Purpose of the Study:

• To evaluate the performance of exact, rule-based, and string-similarity-based matching techniques for biomedical term normalization.
• To compare these methods using both a gold-standard dataset and a practical curation tool.
• To assess the impact of different matchers on curation efficiency and accuracy.

Main Methods:

• Precision and recall were measured against a gold-standard dataset for each matching technique.
• The matchers were integrated into a curation tool to assess their effect on curator efficiency.
• Curation speed was measured for normalizing protein and tissue entities with assisted curation.

Main Results:

• The rule-based matcher demonstrated superior performance on the gold-standard dataset.
• String-similarity-based and exact string matching techniques significantly improved curation efficiency.
• The choice of matcher impacts performance depending on whether accuracy or speed is prioritized.

Conclusions:

• Different string matching approaches offer distinct advantages in biomedical term normalization.
• Rule-based systems are precise for data evaluation, while similarity and exact matching accelerate practical curation tasks.
• Optimizing term normalization requires selecting the appropriate matching strategy based on specific research goals.