Definition · Glossary

Data provenance

Where a piece of data came from, and what happened to it on the way. For a protein, the first question is simple: was this shape measured, or predicted?

Related: AI-ready data.

Raphael Kim· 13 Nov 2026· v1.0

In short

Data provenance is the record of where data came from, who made it, and what was done to it. In protein work, it tells you whether a shape was measured in a lab, borrowed from a related species, or predicted by a model.

01

What it is

The web’s standards body, W3C, published a standard for provenance in 2013. It describes provenance as information about the things, activities and people involved in producing a piece of data, which helps you judge how far to trust it.

In practice that comes down to three questions. Where did it come from: which organism, which sample, which place? How was it made: measured, inferred, or predicted, and by what method? What happened to it since: was it cleaned, merged, copied or corrected?

02

What “AI-ready data” means

“AI-ready data” is a phrase used mostly by data teams in medicine and drug discovery. The US National Institutes of Health runs a $130 million programme, Bridge2AI, to build datasets of this kind. Its working definition: data prepared so that it was gathered ethically and can be used well to train AI models.

Bridge2AI lists what makes a dataset AI-ready. Provenance is on that list, next to things like how well the data is described and whether it can be found and reused.

In that world, provenance is mostly about trusting a model’s output. A designer can take it one step further and ask who and what the data came from.

03

Measured, inferred, predicted

Protein databases already record some provenance, if you know where to look. UniProt’s hand-checked section, Swiss-Prot, gives every entry an evidence level. In its September 2026 release, 21% of entries had evidence at the protein level. About 67% were inferred from related species, and 2% were predicted.

So even in the most carefully checked part of UniProt, two in three entries rest on what is known about a relative.

Shapes follow the same pattern. The Protein Data Bank holds shapes measured in labs, and each entry records the method used. The AlphaFold Database holds predicted shapes, with a confidence score for every part of the chain. AlphaFold itself learned from the measured shapes, so its predictions carry their provenance too.

A ribbon drawing of a small folded protein, with spirals and arrow-shaped strands — The shape of hen egg-white lysozyme, measured from crystals. Protein Data Bank entry 1HEL. The entry records the method, how sharp the measurement was, and who did the work.Image: Jawahar Swaminathan and MSD staff, European Bioinformatics Institute, via Wikimedia Commons. Public domain.

04

What you can do as a designer

Provenance is easy to lose and hard to rebuild. It goes missing when a shape is copied into slides, a sequence is pasted into a tool, or a result is passed to someone else. A few habits help:

Tag every shape. Measured, inferred, predicted, or unknown.

Write down the source, version and date. Database entries are updated, and predictions change between releases.

Record the organism, and the place where you can find it. See digital sequence information for why the place matters.

Keep the record with the work. Put it in the caption, the file name or the project notes, so it travels when you hand the work on.

For AI outputs, note the model and settings. A design you generated has provenance too.

This is the judgement part of being an AI-native biodesigner: reading what an output can show, and what it can’t.