Written and maintained by CASRAI Editorial Board
Last updated
A genome-wide association study (GWAS) does not test one hypothesis — it tests hundreds of thousands to millions of single-nucleotide polymorphisms (SNPs) against a trait, one regression or association test per variant. Running that many tests at the conventional α = 0.05 threshold would produce an overwhelming number of false positives by chance alone, so GWAS uses a far stricter bar: genome-wide significance, conventionally set at p < 5×10-8. This page explains where that number comes from, how it differs from the “suggestive” threshold researchers also report, and when a study is justified in departing from it.
What genome-wide significance actually controls
Genome-wide significance is a multiple-testing correction, not an arbitrary convention. For m independent tests each run at α = 0.05, the probability of at least one false positive across the whole study — the family-wise error rate (FWER) — is 1 − (1 − 0.05)m. A GWAS chip or imputed dataset routinely carries hundreds of thousands to several million SNPs, so left uncorrected, a genome scan is all but guaranteed to throw off “significant” hits that are pure noise. See CASRAI’s guide on the Bonferroni correction and multiple comparisons for the general version of this problem outside genomics; genome-wide significance is that same logic applied at genome scale.
Why 5×10-8, specifically
The 5×10-8 threshold is a Bonferroni correction (0.05 divided by the number of tests) calibrated to roughly one million independent tests — the effective number of independent common SNPs in the human genome once linkage disequilibrium (LD) is accounted for, estimated from the International HapMap Project’s Phase II data. Because nearby SNPs are correlated (inherited together in LD blocks), the true number of independent tests in a genome scan is far smaller than the number of SNPs actually genotyped — commonly one million rather than several million. Dividing 0.05 by roughly one million independent tests lands close to 5×10-8. The International HapMap Consortium and the Wellcome Trust Case Control Consortium (WTCCC, 2007) independently converged on near-identical thresholds using this reasoning, and 5×10-8 has been the de facto standard for common-variant GWAS in European-ancestry populations since. It replaced an earlier era of candidate-gene studies that used far looser thresholds and consequently produced a large body of associations that failed to replicate — the strict bar exists specifically because that history showed what happens without it.
Suggestive significance: the other threshold GWAS papers report
Alongside genome-wide significance, GWAS results tables and Manhattan plots typically mark a second, looser line: suggestive significance, conventionally p < 1×10-5. This threshold traces to Lander and Kruglyak’s 1995 statistical guidelines for linkage studies, later adapted for SNP-array GWAS, and is set so that under the null hypothesis, a genome scan is expected to produce roughly one suggestive hit by chance. Suggestive hits are not treated as confirmed associations — they are candidates for follow-up in an independent replication cohort, meta-analysis, or a targeted look at a specific locus. A result that clears genome-wide significance but has not been independently replicated is still conventionally reported as unconfirmed; replication in a separate sample remains the standard that turns a genome-wide-significant signal into an accepted association.
How this differs from ordinary statistical significance
Outside genomics, “statistically significant” almost always means p < 0.05 — see CASRAI’s guides on what a p-value is and statistical significance for that baseline. Genome-wide significance is not a different concept, it is the same p-value logic under a correction sized to the actual number of tests being run. A p-value of 0.001 would be highly significant in a single hypothesis test and would be reported prominently in most quantitative research; in a GWAS, it does not even clear the suggestive line, because 0.001 is nowhere near strict enough once roughly a million independent comparisons are in play. Researchers moving between single-hypothesis analysis and genome-scale analysis need to recalibrate what “significant” means for the design in front of them, not carry one field’s threshold into the other. CASRAI’s guide on the relationship between p-values and confidence intervals and on how to report p-values cover the general reporting discipline this scales up from.
When is a looser threshold defensible?
5×10-8 was calibrated for chip-based common-variant GWAS in moderate-sized cohorts. Two conditions have pushed researchers to reconsider it rather than treat it as fixed forever:
- Very large sample sizes. Work revisiting the standard threshold has found that in cohorts above roughly 100,000-120,000 participants, loosening the bar to around 5×10-7 still returns a high proportion of true positives among the newly captured hits (roughly 90-93% true positives at that relaxed threshold in large-cohort analyses, versus a meaningfully lower true-positive share — roughly 77-87% — in more modest cohorts of around 20,000). The practical guidance is not “5×10-8 is wrong,” it is that threshold choice should match study purpose and sample size: a relaxed threshold is defensible for exploratory work feeding into gene-set or pathway enrichment analysis, where some false positives are tolerable, and indefensible when a resource-intensive, high-confidence follow-up depends on the result.
- Whole-genome sequencing versus SNP arrays. A GWAS built on whole-genome sequencing tests many more variants, including rare ones with weaker LD correlation to their neighbors, than a genotyping-array study does. That changes the effective number of independent tests the Bonferroni logic above depends on, which is why sequencing-based studies increasingly report their own, often stricter, genome-wide thresholds rather than reusing 5×10-8 unmodified.
False discovery rate (FDR) control — the Benjamini-Hochberg procedure and its q-value framing — is the other tool researchers reach for as an alternative to a single fixed Bonferroni-style cutoff, particularly for large studies where FWER control is more conservative than the analysis needs. See CASRAI’s guide on q-values and the false discovery rate for how that alternative is calculated and when it fits better than a fixed genome-wide threshold.
Where this shows up in practice
Genome-wide significance is the line drawn on a Manhattan plot — the standard GWAS results figure, with SNP genomic position on the x-axis and -log10(p) on the y-axis — above which a variant is flagged as a locus of interest. It is also the threshold journals and databases like the NHGRI-EBI GWAS Catalog use to decide what counts as a reportable “hit” in the first place. Because GWAS results routinely feed into controlled-access data repositories for downstream re-analysis, researchers working with this kind of data should also understand the data-governance side: CASRAI’s guides on dbGaP, dbGaP submission and controlled-access data requests, and the NIH Genomic Data Sharing (GDS) Policy cover what happens to GWAS-scale data after the analysis stage. For the broader context of what genomics research covers as a field, see CASRAI’s What Is Genomics? overview, and for the general statistical foundation this page builds on, see the Research Methods hub.
Frequently asked questions
What p-value counts as genome-wide significant?
p < 5×10-8 is the conventional threshold for common-variant GWAS in European-ancestry cohorts, derived from a Bonferroni correction sized to roughly one million independent tests (the effective number of independent common SNPs once linkage disequilibrium is accounted for).
Is 5×10-8 the same in every GWAS?
No. It is the standard for chip-based common-variant GWAS in populations with LD patterns similar to those used to derive it. Whole-genome-sequencing studies, studies in populations with different LD structure, and very large cohorts (where a relaxed threshold around 5×10-7 has been shown to still return mostly true positives) may justify a different threshold, and should state their reasoning explicitly rather than default silently to 5×10-8.
What is the “suggestive” significance threshold?
Conventionally p < 1×10-5, tracing to Lander and Kruglyak’s 1995 linkage-study guidelines. It flags candidate loci expected to occur roughly once per genome scan by chance under the null, worth following up but not confirmed until independently replicated or validated at genome-wide significance.
Does clearing genome-wide significance mean the association is real?
It means the association is unlikely to be a chance false positive given the number of tests run — it does not by itself confirm biological validity or rule out confounding (population stratification, batch effects, genotyping error). Independent replication in a separate cohort remains the field’s standard for treating an association as established.
This page reflects the conventions most consistently reported for common-variant GWAS in the published genomics literature as of 2026; researchers should confirm the threshold expected by their specific journal, consortium, or funder before finalizing an analysis plan.
