Chi-square-based scoring function for categorization of MEDLINE citations.

Area of Science:

• Biomedical Informatics
• Computational Biology
• Text Mining

Background:

• Text categorization is crucial for identifying relevant biomedical documents.
• Efficiently identifying genetic-related literature within large databases like MEDLINE is challenging.

Purpose of the Study:

• To develop a simple, chi-square-based scoring method for categorizing MEDLINE citations by genetic relevance.
• To assess the performance of this method against established machine learning algorithms.

Main Methods:

• Constructed genetic and non-genetic document corpora using MeSH descriptors from MEDLINE citations.
• Applied chi-square tests to compare MeSH descriptor frequencies between corpora.
• Scored citations based on the relative frequency of genetic domain-typical MeSH descriptors.

Main Results:

• Achieved a predictive accuracy of 0.87, with 0.69 recall and 0.64 precision on 734 manually annotated citations.
• Chi-square scoring performed comparably to support vector machines, decision trees, and naïve Bayes algorithms.
• The method effectively identifies MEDLINE citations relevant to genetic topics.

Conclusions:

• Chi-square scoring offers an effective solution for categorizing MEDLINE citations.
• The algorithm is integrated into the BITOLA system for gene symbol disambiguation.
• This method aids in literature-based discovery and analysis of genetic research.