When epigenetic clocks cross populations
A model can measure biology precisely and still fail the population where it is used. What we owe people a method was never calibrated for.
The promise, and the trap
An epigenetic clock reads the chemical marks layered on top of DNA and returns an estimate correlate: a biological age. It is one of the quietly remarkable results of modern genomics: biology, made legible.
But a model learns the population it was trained on. Point it at people it never saw and the number it returns can be precise and wrong at the same time.
A model can measure biology precisely and still fail the population where it is used.
What we owe a method's blind spots
This is not an argument against measurement. It is an argument for humility about calibration: testing whether a tool built elsewhere still tells the truth here, before we let it make decisions about people's health.
Some of my current work asks this of published age models in African-ancestry settings. The honest answer matters more than a flattering one.