Automatic extraction of protein point mutations using a graph bigram association.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Point mutations are crucial for understanding protein structure and function.
• Experimentally determined point mutations and their impacts are primarily documented in peer-reviewed literature.
• Existing databases often lack comprehensive indexing of point mutations and their associated biological context.

Purpose of the Study:

• To develop and evaluate an automated application, Mutation GraB (Graph Bigram), for identifying, extracting, and verifying point mutations from biomedical literature.
• To address the challenge of linking point mutations to their specific protein and organism of origin.
• To improve the accuracy of text-mining for mutation-protein associations.

Main Methods:

• Developed a graph-based bigram traversal algorithm to identify associations between point mutations, proteins, and organisms.
• Utilized the Swiss-Prot protein database for information verification.
• Incorporated term frequency and positional data within articles to enhance mutation-protein association.
• Compared the graph bigram metric against a word-proximity metric using full-text literature from GPCR, tyrosine kinase, and ion channel families.

Main Results:

• The graph bigram metric demonstrated superior performance over the word-proximity metric across all tested protein families (GPCRs, tyrosine kinases, ion channels).
• Achieved higher F-measures for GPCRs (0.79 vs. 0.76), protein tyrosine kinases (0.72 vs. 0.69), and ion channel transporters (0.76 vs. 0.74).
• Significantly improved precision in disambiguating point mutations associated with multiple proteins (0.84 vs. 0.73).

Conclusions:

• The graph bigram metric represents a significant advancement over previous search metrics for point mutation extraction.
• Mutation GraB offers a robust solution for text-mining applications requiring accurate association of words, particularly in the context of protein mutations.
• The developed method enhances the ability to mine biomedical literature for critical information on protein variations and their functional consequences.