Browse by type of data
Narrow by year, organization, author, or keywordFour datasets to start from
A cross-section of the index, from large computed collections to in-situ imaging series.
DFT estimates of solvation energy
130,258 G4MP2-quality solvation energy calculations covering molecules across a range of solvents. Consistently computed throughout, which makes it a practical starting point for property-prediction models.
10.18126/jos5-wj65130,258 calculationsG4MP2loads with Foundry-ML
Semantic segmentation of dendrites
Labelled training and test data from an in-situ solidification X-ray tomography experiment at Argonne.
10.18126/d4kn-cz3pXCTpixel labels
X-ray CT of composite microstructure
In-situ observation of a composite microstructure during polymer-to-ceramic conversion.
10.18126/wr84-hg0a3D imagingin situ
D3TaLES: redox-active molecules
High-throughput DFT across 35,729 molecules, with more than 90 computed properties per structure.
10.18126/v5sj-6q9335,729 molecules90+ properties
Two ways in
Publish your data
Free, with a citable DOI at the end — it satisfies most funders’ data-management plans.
Start a submissionSign in with Globus. Most universities and national labs support institutional login.Create an accountJoin the MDF group
Upload your files. From a laptop, a Globus endpoint, or Google Drive — terabytes included.What makes a good dataset
Get your DOI. Describe the dataset once; revisions publish as new versions, so citations keep resolving.
Search in your own words. Keywords or a plain-language description, across every collection at once.
Check it before you download. Metadata, formulas, file listings, and version history on every record page.
Cite what you used. Each DOI is pinned to a version, with citation text ready to copy.
Would you rather do this in a terminal?
mdf publishes, searches, and clones datasets from the command line — and ships an agent skill that drafts your metadata, gated on your approval.
pipx install mdf-cli
Load ML-ready data in a few lines of Python
Foundry-ML serves structured, validated datasets directly into your notebook. f.get_dataset(doi) returns arrays that are ready to train on, so there is no cleaning step first.
- 60+
- collections ready to load
- 5
- lines from import to arrays
pip install foundry-ml
# Import Foundry-MLfrom foundry import Foundryf = Foundry()# Load a dataset by DOIdoi = '10.18126/qsdl-qj6x'ds = f.get_dataset(doi)# Get data as a dictionaryX, y = ds.get_as_dict()
In the spotlight
Longer write-ups on notable dataset releases, with the scientific context alongside the access details.
An electronic-structure dataset for molecular machine learning
500 terabytes of electronic densities, wavefunctions, and molecular-orbital information from more than four million high-accuracy DFT calculations.
500 TB / 4M+ calculations / CC-BY-4.0
Read the spotlight