Polygenic risk scores are not diagnoses. Stop selling them like they are.
A polygenic risk score is a probability estimate. It sums the small effects of thousands to millions of common variants to produce a single number that represents your estimated genetic predisposition to a trait, relative to a reference population. At their best, PRS can meaningfully rank-order risk for conditions like coronary artery disease, type 2 diabetes, and breast cancer. At their worst, they are pseudoscience in a progress bar.
The "at their worst" case is what most consumer marketing looks like. A screenshot shared on a product's landing page — "Your polygenic risk for Alzheimer's is in the 92nd percentile" — shown without the two caveats a reader needs. First: PRS are only as good as the reference population they were trained on, and the training data is still overwhelmingly European-ancestry. For anyone outside that group, transferred PRS performance degrades, often dramatically. Second: even a correctly-trained PRS is not a diagnosis. It is a relative-risk number, and the absolute-risk numbers underneath can be small enough that the 92nd percentile still represents a small lifetime probability.
The FDA drafted non-binding guidance on direct-to-consumer pharmacogenomic reporting in March 2026. It did not cover PRS reporting directly, but the framing is instructive: the agency's concern is not that the findings are wrong but that consumers will act on them without clinical context. Pharmacogenomics has a narrower action space than polygenic risk — change a dose, change a drug — and even there, the FDA explicitly asks providers to qualify findings as "suggestive" and to recommend clinician involvement before any action.
PRS reporting ought to be held to at least that standard. Nucleus, to its credit, does present polygenic findings with confidence bands and an explicit "not medical advice" framing inside the product. Outside the product — in ads, in social-media posts, in the email subject lines that lead customers to buy, and now in Nucleus' embryo-screening marketing — the framing collapses. That collapse is the specific thing genetics ethicists have been complaining about for the past eighteen months, and it is the specific thing that keeps a company like Nucleus out of an Editor's Choice slot it would otherwise be a strong contender for.
What we want to see, before we start handing out category awards on polygenic features: ancestry-aware PRS that report meaningful findings for non-European-ancestry customers instead of silently downgrading. Absolute-risk numbers, not just percentiles, on every PRS finding. Clinician-referral flows that are used, not just mentioned. And marketing copy that treats a polygenic score as a probability estimate — because that is what it is — rather than as a verdict, including when the verdict in question is about an embryo.