NLM-Chem-BC7

NLM-Chem-BC7 provides annotated corpora and evaluation resources for automatic recognition, normalization, and indexing of chemical entities in full-text PubMed Central articles to support Chemical Identification and Chemical Indexing tasks and MeSH-linked entity recognition algorithm development.

Key Features:

• Comprehensive Corpora: Includes a Chemical Identification Corpus of 204 full-text PubMed Central (PMC) articles annotated for chemical entity spans (named entity recognition) and normalization (entity linking) to MeSH by 12 NLM indexers, and a Chemical Indexing Corpus of 1333 recent PMC articles with manual indexing of chemical substances by NLM experts as a gold-standard for MeSH indexing prediction.
• Manual Annotation and Verification: Both corpora were manually curated by NLM indexers, with the Chemical Indexing Corpus further enriched post-challenge by manual verification of candidate terms from challenge predictions not present in the original MeSH indexing and by algorithmic merging of chemical entity annotations to retain highest-confidence entries.
• Algorithm Evaluation and Improvement: The corpora are used to train and test algorithms for chemical entity recognition and MeSH indexing, enabling assessment of accuracy and reliability of extraction and indexing methods.

Scientific Applications:

• Text-Mining Enhancement: Provides high-quality annotated data to improve precision and normalization of chemical information extraction from biomedical literature.
• Research and Development: Supports development and refinement of chemical named entity recognition, normalization, and MeSH indexing algorithms.

Methodology:

Expert manual annotation of full-text PMC articles by NLM indexers; normalization by linking identified entities to MeSH; manual verification of candidate terms from challenge predictions not present in original MeSH indexing; algorithmic merging of chemical entity annotations to retain highest-confidence annotations; and post-challenge statistical validation to refine the corpus.