May 20, 2026 · 7 min read

Polygenic risk scores (PRS) for non-scientists

Almost every modern genetics report talks about polygenic risk scores, or PRS. The term sounds intimidating but the underlying idea is straightforward. Here's the plain-English version, plus what a PRS can and can't tell you about your health.

One gene, many genes

Some conditions are caused by a single gene. Cystic fibrosis, sickle cell anaemia, Huntington's disease — these are "monogenic" diseases where one mutation deterministically causes the condition. Genetics there is binary: you have the variant or you don't.

Most common conditions are different. Type 2 diabetes, heart disease, depression, obesity, most cancers — these are polygenic. Hundreds or thousands of genetic variants each contribute a tiny bit to your overall risk. No single one is decisive. The total is what matters.

What a PRS actually computes

A polygenic risk score adds up the contributions of all known risk variants for a given condition, weighted by how strong each one is. Conceptually:

PRS = Σ (your dose of variant i × effect size of variant i)

"Dose" is 0, 1 or 2 — how many copies of the risk allele you carry at that position (you have two copies of each chromosome, so the maximum is 2). "Effect size" comes from genome-wide association studies (GWAS) — peer-reviewed papers that measured how much that variant moved disease risk in thousands of people.

How to read your score

Raw PRS numbers are not meaningful on their own. They become meaningful when compared against a population distribution — your score's percentile. A percentile of 80, for example, means your genetic loading for that condition is higher than 80% of people in the reference population.

Most reports — including ours — classify scores into three tiers:

• Low — below the 50th percentile of the reference population.
• Moderate — between the 50th and 80th.
• High — above the 80th.

A "high" PRS is not a diagnosis. It typically translates to 1.5–3× lifetime risk increase compared to the population average — meaningful, but not destiny. Lifestyle, environment, and other genes you didn't inherit also matter, often more than the polygenic component.

The limitations

There are four big caveats to keep in mind:

1. Ancestry bias. ~90% of all GWAS participants are of European descent. PRSs trained on European data underperform — sometimes drastically — on African, East Asian, South Asian and Hispanic/Latino populations.
2. Chip coverage. Microarray-based tests (23andMe, AncestryDNA, MyHeritage) only measure ~0.02% of the genome. Most GWAS variants are covered, but some PRS calculations would benefit from whole-genome sequencing.
3. Effect size uncertainty. Each effect size is itself an estimate with a confidence interval. Aggregating many small effects amplifies noise as much as signal. A "high" PRS in our report and "high" elsewhere may come from different variant lists.
4. Population vs individual. Even a perfectly calibrated PRS tells you about the average risk of people like you, not your personal outcome. Two people with identical scores can have radically different lives.

What it's actually useful for

With those caveats, a PRS is a starting point for a conversation, not a verdict. For some conditions — coronary artery disease, type 2 diabetes, breast cancer — a high PRS is now considered actionable enough by some clinicians to motivate earlier or more aggressive screening. For most conditions, it's information you can use to inform lifestyle choices: high cardiovascular PRS → take blood pressure more seriously; high lipid PRS → get a fasting lipid panel earlier.

Crucially: a low PRS does not protect you. Smokers with low cardiovascular PRS still get heart disease.

How we compute yours

Our reports compute a PRS for 11 trait categories — type 2 diabetes, lipid profile, cardiovascular, Alzheimer's, depression, cancer, and others — using variants from the GWAS Catalog with genome-wide significance (p < 5×10⁻⁸). Each result shows the score, the percentile, the confidence (how many of the known variants we found in your file), the top contributing variants, and a plain-language explanation. The methodology page has the full details.