enPubMed: An Enhanced Metadata Version of the PubMed Literature Database

This dataset, named enPubMed, is an enhanced version of the PubMed/MEDLINE database — one of the most important literature resources for biomedical scientists, drug developers, and healthcare professionals. Despite decades of development by the National Library of Medicine (NLM), PubMed suffers from systematically missing or incomplete metadata, including author names, affiliations, ORCID iDs, DOIs, abstracts, keywords, and references. For example, only 22% of articles contain reference sections, and only 61.57% of author names are complete. These issues negatively impact literature search, citation network analysis, biomedical knowledge mining, and the training of large language models (LLMs) in biomedicine. To address this problem, we established large-scale, dense linkages between PubMed and six additional open scholarly databases: Microsoft Academic Graph (MAG), Semantic Scholar (S2), Open Academic Graph (OAG), OpenAIRE's DOIBoost, ORCID, PubMed Central (PMC-OA), and Europe PubMed Central. Using a hybrid linking approach (combining DOI-based and optimized content-based matching) and metadata aggregation, we systematically restored and supplemented missing metadata. Key improvements (absolute percentage point increase in accessibility): Metadata Field PubMed (original) enPubMed (enhanced) Absolute Improvement Author full names 61.57% 93.62% +32.05% Author affiliations 46.48% 78.79% +32.31% Abstracts 67.46% 76.89% +9.43% Keywords 18.29% 44.20% +25.91% ORCID iDs 3.12% 11.42% +8.30% The accessibility of DOIs and references also improved, though some limitations remain. Importantly, our evaluation shows that the enhancement maintains low error rates (overall content‑based linking error ratio: 0.55%). Scientific significance and findings: Enhanced search and discovery: With over 36% of PubMed queries including author names, the substantial increase in full author name coverage directly improves author search accuracy and reduces irrelevant results. Improved literature exposure: The addition of 9.43% more abstracts and 25.91% more keywords increases the visibility of previously under‑exposed articles, promoting broader dissemination of biomedical discoveries. Richer citation networks: By restoring both internal and external citations (PubMed originally only included internal citations), enPubMed enables more complete citation network analysis, benefiting science of science (SciSci) research, research trend detection, and knowledge linkage exploration. Foundation for AI and LLMs: High‑quality metadata is critical for training reliable large language models in biomedicine. enPubMed provides a cleaner, more complete data foundation for such efforts. Unique contributions from multiple databases: Each integrated database (MAG, S2, OAG, ORCID, PMC‑OA, DOIBoost) makes exclusive contributions to specific metadata fields, demonstrating the necessity of multi‑source fusion. Important limitations: This version primarily focuses on supplementing or restoring missing metadata. It does not perform deep fusion (e.g., merging near‑duplicates with fine‑grained resolution) or global disambiguation of authors and affiliations. The accessibility of certain fields remains limited (e.g., ORCID coverage is still below 12%). Inconsistencies or ambiguities from source databases may still exist in the aggregated metadata. The current enhancement approach follows a simple rule (e.g., selecting the longest metadata) for non‑standardized fields such as affiliations and keywords, rather than a sophisticated merging strategy. Therefore, this dataset is shared as a preliminary release to enable community evaluation, feedback, and further improvements. Users are encouraged to consider these limitations when using the data for downstream research. File contents include enhanced fields for: Author full names Author affiliations ORCID iDs DOIs Abstracts Keywords