A hybrid named entity tagger for tagging human proteins/genes

Area of Science:

• Biomedical Informatics
• Computational Biology
• Natural Language Processing

Background:

• Accurate extraction of gene/protein names is crucial for biomedical literature analysis.
• Existing Named Entity Recognition (NER) tools often lack high performance for human gene/protein tagging.
• A precise tagger for human genes and proteins is essential for current text mining research.

Purpose of the Study:

• To develop a novel hybrid approach for improved human gene and protein name recognition in biomedical texts.
• To address limitations and common errors found in existing NER taggers.
• To enhance the accuracy of Named Entity Recognition (NER) for human gene/protein identification.

Main Methods:

• A hybrid approach combining machine learning (Conditional Random Fields), manually curated rules, and a novel abbreviation identification algorithm.
• Development of a domain-specific algorithm tailored for human gene and protein nomenclature.
• Utilizing the JNLPBA2004 corpus for experimental evaluation.

Main Results:

• The proposed hybrid approach achieved a precision of 80.47% and an F-score of 75.77% on the JNLPBA2004 corpus.
• The system demonstrated superior performance compared to most existing state-of-the-art NER systems for human gene/protein tagging.
• A recall of 71.60% was achieved, indicating areas for future methodological refinement.

Conclusions:

• The developed hybrid NER system offers a significant advancement in accurately identifying human genes and proteins.
• The approach effectively surmounts common errors in existing taggers, improving scientific literature analysis.
• Further research is needed to enhance recall and achieve even higher accuracy in gene/protein name extraction.