NEMine

NEMine enhances recognition of gene and protein names in biomedical text by integrating named-entity dictionaries with POS tagging and statistical NER outputs to reduce term ambiguity and variability.

Key Features:

• Dictionary-Based Enhancement: NEMine utilizes existing and additional named entity (NE) dictionaries to improve recognition accuracy.
• No Retraining Required: Incorporating protein names into the POS tagger's dictionary improves performance without retraining statistical NER models.
• POS Tagging Integration: Known NEs are identified concurrently with Part-of-Speech (POS) tagging using a general word dictionary and an NE dictionary.
• Dictionary Enrichment: Adding protein names found in training data to the POS tagger's dictionary further enhances recognition accuracy.
• Performance Evaluation: On the JNLPBA-2004 dataset, F-score increased from 73.14 to 73.78 after adding training-set protein names and up to 78.72 after enriching the dictionary with test-set protein names.

Scientific Applications:

• Protein Name Recognition: NEMine improves identification of protein names that exhibit high term variation and ambiguity.
• Biomedical Text Analysis: The approach is applicable to biomedical text mining tasks requiring accurate identification of gene and protein names in genomics and proteomics research.

Methodology:

Known NEs are identified concurrently with POS tagging using a general word dictionary and an NE dictionary; statistical NER models are trained on outputs from the POS/PROTEIN tagger tagged with correct NE labels; protein names found in training data are added to the POS tagger's dictionary to enhance performance without retraining.