What is a polygenic score
Many common conditions and traits are influenced by thousands of genetic variants, each with a tiny effect. A polygenic score adds these effects up into one number and compares it with other people. This page explains how such a score is calculated, how to read it, and why it is not a diagnosis.
- What it is
- a weighted sum of many small genetic effects, compared with a reference population
- What it is not
- a diagnosis, or the probability that you will get the condition
- Scores in greatgenes
- 15 published scores from the PGS Catalog, weights used as published
- Coverage threshold
- a score is assessed only if your chip covers at least 50 % of it
- Shown openly
- results at least 1 standard deviation from the average
- Reference population
- European participants of 1000 Genomes (genotypes for published scores, allele frequencies for GWAS traits)
How a score is calculated
In short: for each variant in the score, count how many copies of the “effect” letter you have and multiply by the variant’s weight from a large study; then add everything up.
The weights come from genome-wide association studies, which compare large numbers of people, often hundreds of thousands, with and without a condition (or with different values of a trait such as cholesterol). Researchers publish the finished scores — the list of variants and their weights — in the PGS Catalog, where each has an identifier such as PGS000028.
greatgenes uses 15 published scores, chosen because consumer chips cover them well, with their published weights as they are: no weights of our own and no guessing of unmeasured positions. The calculation runs in your browser.
Percentile and z: how to read the result
In short: z says how many standard deviations your score is from the average of the reference population; the percentile is the share of that population expected to have a lower score.
The raw sum on its own means little, so it is compared with what is expected in a reference population. greatgenes uses the genomes of the European participants of 1000 Genomes: for published scores it computes the score for each of them on exactly the variants found in your file and takes the average and spread from these real values, so variants that are inherited together (for example in the HLA region) are not treated as independent. Your result is expressed as z (distance from the average in standard deviations) and as a percentile, assuming a normal distribution. Traits from the GWAS Catalog use European allele frequencies instead; there neighbouring variants are removed beforehand.
| z | Percentile | Word in the report |
|---|---|---|
| at least +2 | about the 98th or higher | well above average |
| at least +1 | about the 84th or higher | above average |
| between −1 and +1 | between the 16th and the 84th | near average |
| at most −1 | about the 16th or lower | below average |
| at most −2 | about the 2nd or lower | well below average |
In the report, only results at least 1 standard deviation from the average (and with enough coverage) are shown openly. The rest are folded into “Details” under the table with a one-line count; the full table, sorted by distance from the average, is always there.
Coverage: how much of the score your chip reads
In short: if your chip misses too much of a score, the result is not given at all, rather than given wrongly.
No chip reads every variant of every score. greatgenes measures coverage as the share of the score’s variance that your chip’s positions explain, so a missing variant with a large weight counts for more than one with a tiny weight (Choi, Mak, O’Reilly 2020). Below 50 % the score is marked “not assessed” and no percentile is given.
Ancestry: who the scores were built on
In short: most genetic studies, and therefore most scores, are based on people of European ancestry. For people of other ancestries the scores are less accurate.
- About 79 % of participants in genome-wide association studies are of European descent, while people of European descent make up about 16 % of the world’s population (Martin et al., Nat Genet 2019).
- In the same study, prediction accuracy relative to Europeans was on average 1.6-fold lower in Hispanic/Latino Americans, 1.7-fold lower in South Asians, 2.5-fold lower in East Asians and 4.9-fold lower in Africans (Martin et al., Nat Genet 2019).
- greatgenes does not know your ancestry. It compares every score with European allele frequencies. If your ancestry is not European, treat the percentile as a rough indication at best.
Scores for one sex
2 of the scores apply to one sex only: breast cancer (PGS000028, built on women) and prostate cancer (PGS000044). The score itself is calculated from the same kind of variants for everyone; what depends on sex is whether the result applies to you (Mavaddat et al. 2015; Schumacher et al. 2018).
- greatgenes determines sex from the X and Y chromosome data in the file. If it is known, the result for the other sex is not shown.
- If sex cannot be determined (some files have no X or Y data), results for both sexes are shown, each marked “for women” or “for men”, with one sentence above the table. Use the ones that apply to you.
Why a score is not a diagnosis
In short: a polygenic score tells you where you stand compared with other people, not whether you will get a condition.
- A higher score means a somewhat higher chance than average, not a certainty. Many people with a high score never develop the condition, and many people with a low score do.
- The score leaves out lifestyle, environment, family history, age and rare variants with large effects.
- The result is not a diagnosis and not medical advice. If a result worries you, discuss it with a doctor or a clinical geneticist, who can put it together with your family history and other risk factors (Torkamani, Wineinger, Topol 2018).
Other limits of chip data: What a consumer DNA chip can and cannot tell you. The full rules of the report: How greatgenes works: polygenic scores and traits.
Analyse your file — in your browser, without uploading it.
References
- Lambert SA et al. The Polygenic Score Catalog as an open database for reproducibility and systematic evaluation. Nat Genet 2021. doi:10.1038/s41588-021-00783-5
- Choi SW, Mak TS, O’Reilly PF. Tutorial: a guide to performing polygenic risk score analyses. Nat Protoc 2020. doi:10.1038/s41596-020-0353-1
- Martin AR et al. Clinical use of current polygenic risk scores may exacerbate health disparities. Nat Genet 2019. doi:10.1038/s41588-019-0379-x
- Torkamani A, Wineinger NE, Topol EJ. The personal and clinical utility of polygenic risk scores. Nat Rev Genet 2018. doi:10.1038/s41576-018-0018-x
- Mavaddat N et al. Prediction of breast cancer risk based on profiling with common genetic variants. J Natl Cancer Inst 2015. doi:10.1093/jnci/djv036
- Schumacher FR et al. Association analyses of more than 140,000 men identify 63 new prostate cancer susceptibility loci. Nat Genet 2018. doi:10.1038/s41588-018-0142-8