Read your report

Polygenic scores (PRS), in plain words

By Aimosti · 5 min read · 2026-06-30

A polygenic score is not a measurement of your fate. It adds up many tiny genetic nudges into a single leaning, and reading it well means knowing what it leaves out.

The claim. A polygenic score is a precise measurement of your genetic risk, like a cholesterol reading, and the percentile next to it is a verdict on whether you will get the disease.

Verdict. A polygenic score adds up many small-effect variants into a single statistical tendency, read against a reference group. It is a leaning, not a measurement and not a diagnosis, and its accuracy depends on whether your ancestry matches that reference group.

A polygenic score sums many common variants, each with a small effect, weighted by published effect sizes from genome-wide association studies. The total is compared with a reference population and expressed as a percentile, which is an ordinal rank rather than an amount. Because most of the risk for a complex trait comes from the rest of the genome, environment and chance, the score captures a tendency, not an outcome. Its accuracy also depends on how well the reference population matches the reader's own ancestry.

Many small effects, added up

A single common variant rarely moves the odds of a complex trait by much. A polygenic score works by summing many of them. For each scoring variant the engine counts how many copies of the effect allele you carry, zero, one or two, multiplies that by a curated weight taken from published genome-wide association studies, and adds the products together.

The result is one number that stands in for the combined push of hundreds of tiny effects. No single marker is doing the work, and dropping any one of them barely changes the total. That is the whole idea: a complex trait is shaped by many genes at once, so a useful summary has to be a sum, not a hunt for one decisive letter.

What the percentile means, and what it does not

Your raw sum on its own says nothing. It becomes readable only once it is compared with a reference group: the engine works out where your score sits on that group's distribution and reports the result as a percentile. A percentile is an ordinal rank, a position in a line, not a quantity of risk.

Sitting high on that distribution, to describe the shape and not a result, would mean your score is higher than most of the people in the reference group, no more and no less. It would not mean a matching percent chance of anything. Most of the risk for a complex disease still comes from the rest of your genome, your environment and plain chance, none of which this one number captures. A high percentile is a leaning, not a sentence, and a low one is not an all-clear.

Why 'observed, the rest assumed reference' is a strength, not a defect

A score may list a few dozen scoring variants, and your file might not carry a reported call at every one of them. When a marker is missing, the engine does not invent a genotype. It treats that position as the common reference version, the one most people carry, and it keeps count of how many markers were actually observed versus assumed.

The report says this in plain words. It shows a markers-used line, N of M scoring variants covered, and for a variant-only file it states outright that the remaining positions were read as the common genotype, an inference rather than a measurement. That disclosure is the opposite of a defect. A score that silently pretended every position had been measured would be hiding its own coverage.

Naming the observed-versus-assumed split is what lets you weigh the number: a percentile built on most of the markers is a firmer leaning than one built on a handful. The report goes further and declines to place you at all when too few of the scoring variants were found, calling that a coverage gap, not a result.

Why a European-derived score travels poorly to other ancestries

Most of the large studies that produced these scores were run in people of European ancestry. The effect sizes, the chosen markers and the reference distribution all carry that history. The same stretch of DNA can be inherited in different patterns in different populations, so a marker that stands in for risk in one group may stand in less well, or not at all, in another. A score tuned to one population therefore loses accuracy when read against another.

The report states this directly, in its own words: 'Polygenic scores travel poorly across ancestries: if your genetic ancestry differs, the percentile is materially less accurate.' Every percentile is computed against a named reference ancestry, shown as a European reference tag beside the number, and the same caveat is shown to everyone rather than hidden, because hiding it would defeat the point.

Some scores carry a further limit. A breast-cancer score, for instance, is built in women and is labelled as not applicable to men rather than reassuring. Reading a score well means reading its caveats first.

How to read it in your own report

On a real Aimosti report each polygenic score sits in the Polygenic scores section, tagged as a tendency and not a diagnosis, and the sensitive disease scores stay reveal-gated: they appear only when you choose to open them. Each card shows the trait, the named source score from the PGS Catalog, a confidence tier, and the percentile with its reference ancestry and its coverage line.

The worked sample report shows the same layout with a placed, fully covered example that is computed live rather than hardcoded, so it is the safest place to see how the pieces fit before reading your own numbers.

What Aimosti would (and wouldn't) show you

On a real report each polygenic score appears behind an explicit opt-in, labelled with its trait, its source score from the PGS Catalog, a confidence tier, and a percentile against a named reference ancestry. We show how many scoring variants your file actually covered, and we state plainly when the rest were read as the common reference genotype. The European-reference caveat and the not-a-diagnosis line sit next to the number, never in the small print.

What we won't claim

We won't call a polygenic score a measurement, a prediction that you will or will not develop the condition, or a diagnosis. We won't present a European-reference percentile as equally accurate for every ancestry, and we won't hide how many markers were assumed rather than observed. A percentile is a leaning read against a reference group, not a verdict on you.

Bottom line. A polygenic score adds up many small-effect variants into a single statistical tendency, read against a reference group. It is a leaning, not a measurement and not a diagnosis, and its accuracy depends on whether your ancestry matches that reference group.

This is the kind of answer we give. See what your file says.

See a sample report Get my report: €119