A simple approach for protein name identification: prospects and limits

Area of Science:

• Bioinformatics
• Computational Biology
• Biomedical Informatics

Background:

• Vast biological knowledge resides in unstructured biomedical journal articles.
• Automated identification and mapping of gene/protein names to database identifiers are crucial for information extraction and integration.
• Current methods require efficient tools for gene and protein name recognition.

Purpose of the Study:

• To develop a simple and efficient approach for identifying gene and protein names in text.
• To map identified names to unique database identifiers.
• To evaluate the performance of the developed approach in a benchmark task.

Main Methods:

• Utilized synonym lists mapping database identifiers to gene/protein synonyms.
• Generated and curated synonym lists for yeast, mouse, and fly from public databases.
• Employed exact text matching against MEDLINE abstracts.
• Applied rule-based and support vector machine-based post-filters to enhance precision.

Main Results:

• Achieved high recall and precision with F-measures of 0.897 for yeast, 0.764/0.773 for mouse, and 0.768 for fly.
• Performance was competitive with top submissions in the BioCreAtIvE challenge.
• Demonstrated the effectiveness of post-filtering, particularly the support vector machine, for challenging nomenclature.

Conclusions:

• The developed approach effectively identifies gene and protein names and maps them to database identifiers.
• Contextual filtering is essential for improving accuracy, especially for organisms with complex nomenclature like fly.
• The support vector machine-based post-filter significantly enhanced performance for challenging protein name identification tasks.