A genome-wide association study, or GWAS (pronounced “jee-woss”), is a method scientists use to scan the entire human genome looking for tiny genetic variations linked to a specific disease or trait. Instead of guessing which gene might be involved, researchers compare the DNA of thousands of people with a condition (like type 2 diabetes) to thousands without it, searching for patterns that show up more often in one group. Since the first large studies launched in the mid-2000s, GWAS has become one of the most widely used tools in genetics, with over 7,500 published studies and more than one million identified genetic associations cataloged to date.
How GWAS Works
Your DNA is made up of about 3 billion pairs of chemical “letters.” Most of that sequence is identical from person to person, but roughly every 300 letters, there’s a spot where people commonly differ by a single letter. These one-letter differences are called single nucleotide polymorphisms, or SNPs (pronounced “snips”). A modern GWAS can test hundreds of thousands to millions of these SNPs in one pass.
Researchers collect DNA samples from two groups: people who have the trait or disease being studied (cases) and people who don’t (controls). Then they compare SNP frequencies between the groups. If a particular SNP variant shows up significantly more often in the cases than the controls, that spot in the genome is flagged as potentially linked to the trait. The same approach works for traits measured on a scale, like height or blood pressure, where researchers look for SNPs that correlate with higher or lower values.
What makes GWAS distinctive is that it’s “hypothesis-free.” Older methods, called candidate gene studies, required researchers to pick a gene they already suspected was involved and test it directly. That approach worked only when the initial guess was right. GWAS flips the process: it surveys the entire genome without any prior assumptions, which means it can identify genetic regions nobody would have thought to look at. In fact, the majority of genetic variants identified through GWAS turn out to be in stretches of DNA that don’t directly code for proteins, regions that older approaches would have ignored entirely.
Why the Significance Bar Is So High
When you test millions of SNPs at once, some will appear linked to the trait purely by chance. To account for this, GWAS uses an extremely strict statistical threshold. A result is only considered “genome-wide significant” if the probability of it being a fluke falls below 5 × 10⁻⁸, or one in 20 million. For comparison, the typical cutoff in most scientific studies is one in 20.
This threshold was originally calculated using a method called the Bonferroni correction, which essentially divides the standard significance level by the number of independent tests being run. It’s deliberately conservative. Some researchers argue the bar should be raised even further now that modern studies use larger sample sizes and can detect associations with rarer genetic variants, but the 5 × 10⁻⁸ standard has held for nearly two decades because associations that clear it tend to replicate well in follow-up studies.
Reading the Results: Manhattan Plots
GWAS results are typically displayed in a chart called a Manhattan plot, named because it resembles a city skyline. The horizontal axis represents the chromosomes laid end to end, and the vertical axis shows how strongly each SNP is associated with the trait (expressed as a statistical score). Most SNPs cluster near the bottom, meaning they show no meaningful association. The ones that spike upward past the significance line are the “hits,” the genetic regions worth investigating further.
From Association to Understanding
Finding a significant SNP is only the first step. A GWAS identifies correlation, not causation. The flagged SNP might not itself cause anything. It’s often just a marker that happens to sit near the actual functional change in DNA, inherited alongside it as a package. Researchers call these markers “tag SNPs” because they tag a region of interest rather than pinpoint the exact cause.
The work that follows a GWAS hit is called functional characterization, and it can take years. Scientists first narrow down the region through fine-mapping to identify which specific variant is most likely responsible. They then check whether the variant affects how nearby genes are turned on or off in relevant tissues. This involves measuring whether people with different versions of the SNP produce different amounts of a gene’s protein product. From there, researchers may use gene-editing tools to introduce or remove the variant in cell models or animal models, then observe what changes. They can also use techniques that map the 3D folding of DNA to figure out which gene a distant regulatory region actually controls. All of this work is needed to move from “this spot in the genome is statistically associated with heart disease” to “this variant reduces the activity of a specific gene in heart muscle cells.”
Polygenic Risk Scores
One of the most practical applications of GWAS data is the polygenic risk score, or PRS. Most common diseases and traits aren’t driven by a single gene. They’re influenced by hundreds or thousands of SNPs, each contributing a tiny amount. A PRS combines all of these small effects into a single number that estimates your genetic predisposition to a condition.
To calculate a PRS, researchers take the effect size of each relevant SNP from GWAS data, then add up the contributions based on which variants you carry. The result places you on a spectrum relative to the population. Someone in the top few percent for a coronary heart disease PRS, for instance, has a meaningfully elevated genetic risk compared to someone in the middle. These scores are starting to appear in clinical settings, though they’re best understood as one factor among many, alongside lifestyle, family history, and other health data.
The Missing Heritability Problem
One of the most discussed limitations of GWAS is that the genetic variants it identifies often explain far less of a trait’s heritability than expected. The classic example is human height. Studies of families and twins estimate that about 80% of height variation comes from genetics. Yet the roughly 50 genetic variants initially found through GWAS explained only about 5% of that variation. This gap between how heritable a trait appears in families and how much GWAS can account for is called the “missing heritability” problem.
Several explanations have been proposed. Some of the missing signal likely comes from rare variants that standard GWAS arrays weren’t designed to detect. Some may involve interactions between genes, where the effect of one variant depends on the presence of another. Epigenetic changes, which alter gene activity without changing the DNA sequence itself, could also play a role. There’s even evidence that shared environmental factors within families, like a common diet or similar gut bacteria, inflate heritability estimates in family studies, making the GWAS gap look larger than it really is. Larger and more diverse studies have steadily closed the gap, but it remains an active area of investigation.
The Diversity Problem
The vast majority of GWAS participants have been of European descent. Among non-cancer studies, about 71% included European populations, 20% included Asian populations, and only 8% included underrepresented minorities. Cancer-focused studies showed a similar skew, with 67% European, 29% Asian, and just 4% from minority groups.
This matters because genetic risk variants and their frequencies differ across populations. A SNP that is common and informative in one ancestry group may be rare or absent in another. Polygenic risk scores built primarily from European-descent data perform less accurately when applied to people of African, Latin American, or Indigenous ancestry. The practical consequence is that the people who could benefit most from genomic medicine are often the ones for whom the tools work least well. Large-scale initiatives are now working to build more representative datasets, but the imbalance accumulated over nearly two decades of research will take significant effort to correct.