Pieza independiente

Nine billion variants

Movement 4 · The Flesh — September 2026

I The table

In September 2026 an atlas appeared with the precomputed effects of some nine billion single-letter variants. You query it in a second. And here is the thing that has to be said out loud: not one of those predictions has been checked in a body.

It is a real jump and it is worth saying what kind. We have gone from asking what a variant does to having it already written down for nearly all of them.

An atlas of predictions is not knowledge about anyone. It is a table with a margin of error.

II Three things that get confused

What it is	What it looks like	The difference
Predicted molecular effect	a diagnosis	a body has context
Statistical association	a cause	correlation cannot be implanted
Rare variant with no data	a benign variant	absence of evidence

That third row is the one that does damage in a consulting room. A variant nobody has published on is not a harmless variant; it is an unstudied one [and the difference gets paid for in decisions].

“Interpreting is not measuring, and measuring is not curing.”

III What it is actually good for

It is good for ordering the work, which is no small thing; it is almost everything a lab does before touching anything. It lets you rule out quickly what almost certainly does not matter and spend the lab's expensive time on what might. It turns a search of months into an afternoon of filtering.

It is also good for something less discussed. When the prediction fails systematically on one gene, that deviation points at biology the model does not capture [and that is worth more than a hit]. The table's error is a map of what we do not understand.

IV Responsible use [safe]

Treat each prediction as a hypothesis, with its date and its model version.
Check whether experimental evidence exists for that specific variant.
Do not change clinical behaviour on the table alone.
Keep the version you queried. Models update and answers change.
If the variant is rare, say that it is not known. That is a legitimate answer.

V Risks

The immediate risk is use outside the clinic: insurers, employers and consumer apps returning a risk score with no error bar and no family context.

The deeper risk is vocabulary [which is why it never gets corrected]. When a table is called an atlas, people read it as a map of an existing territory; this is a map of a computed one, which is a different thing and has to be said every time.