CoNECo: a Corpus for Named Entity recognition and normalization of protein Complexes

Area of Science:

• Biomedical Informatics
• Computational Biology
• Natural Language Processing

Background:

• Biomedical information extraction has advanced, yet lacks resources for protein-containing complexes.
• Existing resources fail to adequately recognize complex names across diverse organisms.

Purpose of the Study:

• To develop a dedicated corpus for Named Entity Recognition (NER) and Named Entity Normalization (NEN) of protein-containing complexes.
• To create and evaluate computational tools for identifying and standardizing complex entities in biomedical literature.

Main Methods:

• Construction of the Complex Named Entity Corpus (CoNECo) with 1621 documents and 2052 annotated entities.
• Normalization of 1976 entities to Gene Ontology (GO).
• Training and evaluation of transformer-based and dictionary-based taggers.

Main Results:

• The CoNECo corpus demonstrated robust performance in NER and NEN tasks.
• Transformer-based tagger achieved an F-score of 73.7%, and dictionary-based tagger achieved 61.2%.
• The developed taggers were applied for comprehensive annotation of open-access biomedical literature.

Conclusions:

• CoNECo provides a crucial resource for advancing protein-containing complex recognition.
• The developed methods offer effective solutions for NER and NEN in this domain.
• All resources are publicly available to facilitate further research and development.