🧬 Sachs Protein Signaling

What this is

Single-cell flow cytometry of 11 signaling markers across 7,466 human T-cells, in 9 separate perturbation experiments (Sachs et al., Science 2005) — a widely used benchmark in causal discovery.

The interesting question here is not what correlates with what. It is which signaling influences are real, in which experimental context, and on what evidence.

What you can ask

• What did a reagent actually do? Every arm is contrasted against a named reference cohort — or explicitly marked as having none.
• Does a textbook causal edge survive contact with the data? Where the experimental design permits an interventional test, results are 1 confirmed and 2 contradicted; other attempted tests are reported as not evaluable. Edges without an intervention on their parent are not tested.
• Why is this individual cell unusual? Only against a relationship that demonstrably holds in its own experimental arm.
• What can't this data tell us? Often the most useful answer, and one this example will give you directly.

What makes it unusual

Of 18 causal edges, 12 have a declared functional sign with its provenance recorded; 6 have a sourced direction but no declared sign — and those 6 are never rendered as activation or inhibition, because no source says so.

Where the data contradicts the published literature, both are reported and the edge is left alone. Where an experiment has no comparable control, no effect is attributed to it. Retracted claims are retracted rather than quietly recomputed.

Every figure you see is recomputed from the data at build time, and the whole example is verified against a frozen specification before it ships.