Automatic bioinformatic software named entity recognition from literature

Bioinformatics software and databases are crucial elements in contemporary life science research; however, their references in scientific literature are inconsistent and challenging to identify systematically on a large scale. The absence of a thorough and up-to-date catalog of bioinformatics resources hampers endeavors towards automated biomedical knowledge extraction and simplified data analysis.

The researchers introduce SNAIL, a hybrid named entity recognition framework aimed at automatically detecting bioinformatics software and database (SW/DB) references from biomedical texts. SNAIL combines lexical and semantic modeling techniques. The lexical component recognizes orthographic patterns and contextual clues suggestive of SW/DB names, while the semantic component utilizes contextual embeddings produced by transformer-based language models like SciBERT, augmented with an explicit token-masking technique to improve entity-focused representations.

A substantial training corpus was generated automatically through a hybrid pipeline that merges citation-hinted extraction with assistance from large language models. When evaluated on two separate benchmark datasets and real-world research articles, SNAIL significantly surpasses existing methods, including domain-specific techniques such as bioNerDS2 and general-purpose large language models like ChatGPT, Gemini, Grok, and Claude.

Implementing SNAIL in large-scale literature analysis uncovers distinct journal-level preferences across various bioinformatics subfields. These findings indicate that SNAIL offers an accurate and scalable solution for identifying bioinformatics resources in scientific texts, facilitating systematic meta-analysis of tool usage and research trends.