Skip to content
The
New
Alchemist
Subscribe
THE NEW ALCHEMIST
FrontierBiological datasetsWorking brief

Biological datasets

The data behind the model

Before an AI model learns biology, someone has to decide what to measure. The story starts with the cells, the experiments and the people missing from the dataset.

A cell atlas is also a record of choices: whose tissue was collected, which cells survived, and what the assay could see.

A shared reference for the lung

The integrated Human Lung Cell Atlas v1.0 brought together 49 datasets. Its healthy core draws on 107 individuals, with cell-type annotations agreed by six experts. The work of making studies comparable is part of the achievement: a shared reference lets other researchers place their own samples in context.

Human Lung Cell Atlas · Data and methods

The experiment inside the dataset

The HCA Data Portal sets standards for describing assays, experimental designs and metadata. Those details matter when data is reused. A model needs more than a cell’s measurements; researchers also need to know where the sample came from and how it was collected.

How the HCA Data Portal works

Where the story goes next

  • Does a model work on a new donor, a new laboratory and a new disease—not just a held-out portion of the same experiment?
  • What can observational cell atlases teach us, and where do we need experiments that perturb a gene or a cell?
  • Who can access and reuse the data, and who is represented in it?

Editorial preview · Coverage in development

Also on our radar

How much of an organ can we grow?A protein made to find anotherWhat changed in Moderna’s vaccine?

Letters from The New Alchemist

The next story, in your inbox.

New essays and episodes from Deeya Kotecha.

Subscribe to the newsletter