DrugBank
DrugBank is a bioinformatics and cheminformatics database (in plain terms, one that joins biology with the chemistry of medicines) that gathers, for each drug, what is known across chemistry, pharmacology, and biology in one record: its chemical structure, how it works in the body, dosing, food and drug interactions, and the proteins and pathways it acts on. With more than 15,000 drug entries, each cross-referenced to related drugs and targets, it behaves more like a connected knowledge base than a flat list of medicines.
You can browse and search all of it on the web for free. To bring the data into your own pipeline you register for an account and download the full release, which arrives mainly as one large structured `XML` file with some tabular `CSV` exports, so plan for a parsing step to flatten those deeply nested records into fields or retrievable passages. Because the entries are already cross-linked, the data maps neatly onto a knowledge graph if you want to model drug, target, and pathway relationships explicitly.
That structure suits any RAG assistant fielding questions about medicines: interactions, contraindications, mechanisms of action, or drug targets, whether for a research tool, pharmacology teaching, or clinical decision support. Since the entries are hand-curated and consistently shaped, retrieval over them tends to come back cleaner than mining the same facts out of free-text medical literature.
Licensing is the thing to settle before you write a line of code. Academic and other non-commercial use sits under CC BY-NC 4.0, which requires attribution and rules out commercial products, and any commercial use needs a separate paid licence direct from DrugBank. That is exactly why we mark commercial use as prohibited under the free terms, so confirm which licence covers your project first. Treat the clinical detail as reference material rather than medical advice, and handle it with due care in anything user-facing.
If a non-commercial licence will not work for you, reach instead for openly licensed alternatives: ChEMBL and PubChem for bioactivity and chemical data, UniProt for the protein targets in depth, and RxNorm for drug naming and normalisation.
Related sources
AlphaFold Protein Structure Database
EMBL-EBI and Google DeepMind's open database of AI-predicted protein structures, covering over 200 million sequences across almost all of UniProt. Each model ships with per-residue confidence and predicted aligned error scores, and is reachable by website, API, FTP and Google Cloud bulk access. Released under CC BY 4.0 and designed to pair with UniProt and the experimental structures in the PDB.
ClinicalTrials.gov
A US National Institutes of Health database of privately and publicly funded clinical studies. It holds structured records on more than 450,000 studies, including protocols, conditions, interventions, outcomes, and results, all in the public domain.
MIMIC-III / MIMIC-IV
Freely accessible critical care databases with de-identified health data from more than 40,000 intensive care patients at Beth Israel Deaconess Medical Center. They include vital signs, lab results, medications, and clinical notes. Access needs credentialing and a data-use agreement.
MIRIAD
A million-scale medical instruction and retrieval dataset of roughly 5.8 million question-answer pairs, each grounded in a passage from peer-reviewed biomedical literature. Built for medical RAG, retrieval, and hallucination detection.