How to read a genetics study without getting fooled
Genetics headlines follow a template: "Scientists find gene for X." Sometimes there is a solid discovery underneath. Often there is a study of 80 undergraduates that will never replicate. You do not need a degree to tell the difference; you need five questions and the discipline to ask them every time.
1. Find the N
Sample size is the first thing to check and the strongest single filter. Common genetic variants have small effects, and detecting a small effect reliably takes a lot of people.
For scale: the modern studies behind chronotype used 697,828 people. Body mass index, around 700,000. When a study of 120 people claims a common variant explains who gets anxious, the honest reading is "interesting, prove it": samples that small can only reliably detect effects far larger than common variants actually have, so surprising hits in small samples are usually noise.
2. Ask if it replicated
One study is a hypothesis. The same association, found again by an independent team in an independent group of people, is a finding. Modern genome-wide studies build replication in, reporting a discovery sample and a replication sample in the same paper. If a claim rests on a single cohort and nobody has found it twice, treat it as pending.
3. Beware the candidate-gene era
From the 1990s through the 2000s, the standard method was to pick a biologically plausible gene, genotype a small group, and test for an association. It produced thousands of papers and famous "genes for" depression, intelligence, and novelty-seeking. When genome-wide methods and large samples arrived, most of those findings failed to replicate.
The wreckage still circulates, especially in supplement marketing and older wellness tests. If a claim traces to one small study from 2004 and nothing since, that is not a hidden gem, it is a known failure mode.
4. Check who was studied
Most large genetic studies to date were run in people of European ancestry. Associations usually point the same direction in other populations, but effect sizes and the accuracy of combined scores can transfer imperfectly. A finding from UK Biobank applies less precisely to a person of, say, East Asian or African ancestry, and honest reports say so rather than pretending one number fits everyone.
5. Demand absolute numbers
"Forty percent higher risk" can mean a jump from 1 in 100 to 1.4 in 100. Relative numbers create drama; absolute numbers create understanding. Good papers, and good reports, give you the rates for both groups: carriers and non-carriers. If you can only find the relative number, someone chose it for effect. We wrote a separate piece on this habit: Small leans, not verdicts.
Where to check a claim yourself
Two free tools cover most of it. PubMed lets you find the actual paper and read the abstract, where N and replication are usually stated outright. The GWAS Catalog (ebi.ac.uk/gwas) lists, for any variant, every genome-wide study that hit it, with sample sizes, so you can see in one glance whether a claim rests on 200 people or 200,000.
This checklist is also, as it happens, a description of how Helisoma is built. Every card in your report links the study behind it, states the effect as a lean with its honest size, and prefers large, replicated findings, with the rare exploratory ones labeled in the underlying data rather than dressed up as facts. If you want a report you can audit, that is the one we tried to make.
Sources
- Jones SE et al. Genome-wide analyses of chronotype in 697,828 individuals. PubMed 30696823
- Yengo L et al. GWAS of height and BMI in ~700,000 individuals. PubMed 30124842
- The GWAS Catalog, a curated database of published genome-wide associations. ebi.ac.uk/gwas