Overview
PubMed is the comprehensive biomedical literature database maintained by NIH containing over 36 million citations spanning medicine, biology, chemistry, and health sciences. It serves as the primary data source for training biomedical AI models.
What’s In It
PubMed contains 36+ million citations and abstracts from biomedical literature going back to the 1950s. PubMed Central provides full-text access to over 9 million articles. Structured metadata includes MeSH terms, affiliations, and funding sources.
How It’s Used
PubMed data trains specialized biomedical models including BioGPT, PubMedBERT, and BioMedLM. The structured metadata enables fine-grained retrieval. Clinical NLP systems use it for evidence-based question answering.
Controversies
Training on medical literature raises concerns about models providing medical advice without clinical validation. Publication bias toward positive results propagates through trained models. Older papers may contain outdated medical knowledge.