Automatic Extraction from Scientific Abstracts of Synonyms for Proteins and Genes

Introduction: Protein and gene names change frequently as research reveals details about these entities. 1 Because authors often use synonyms, information retrieval requires identification of these alternate names. Many biological databases — such as GenBank 2 and SWISSPROT 3 — have synonym databases; however, the databases may not be complete. Furthermore, to our knowledge, the extraction of synonyms is mainly done by laborious manual curating and review. It is desirable to automate the process due to the enormous volume of publication. We observed that many scientific abstracts have summaries of synonyms. The synonyms are often specifically proposed or mentioned and may be classified into a set of patterns to be recognized by automation. Methods: We manually classified and evaluated patterns used by authors to represent synonyms of proteins or genes in scientific abstracts. We implemented a program, SRE (for synonym recognition and extraction), to recognize and extract the terms associated with the patterns. SRE is written in Perl. The output of SRE is sets of two or more synonyms. We applied SRE to 2,312 scientific abstracts, a subset of abstracts we downloaded from PubMed by the keyword " human. " We then evaluated the precision of SRE's results, using our own judgment as the standard. Precision is the number of correct sets of synonyms of proteins or genes divided by the total sets of terms retrieved. Results: We classified several patterns that express synonyms of proteins and genes in scientific abstracts. The simplest patterns are " synonym " or " a synonym of, " such as in " Thermoactinomyces candidus should be considered a synonym of Thermoactinomyces vulgaris…, " 5 where synonyms Thermoactinomyces candidus and Thermoactinomyces vulgaris can be extracted as noun phrases before and after the string "a synonym of." To evaluate whether the patterns of " synonym " and " a synonym of " would help us to find synonyms of proteins or gene names, we retrieved all the PubMed abstracts that contained the keyword synonym and manually analyzed whether the associated terms are proteins or genes. A search on the keyword synonym for abstracts from 1966 to present retrieved a total of 540 abstracts. A subset of 30 randomly selected abstracts contained no protein or gene names; in most cases, terms were names of species. We therefore discarded this approach. " Called " and " known as " are frequently used to introduce synonyms (" ...Apo3 (also …