Bayesian GWAS — Bayesian Genome-Wide Association Study
Also known as: Bayesian GWAS, Bayesian genome-wide association analysis, Bayesian GWA study, BF-GWAS
Bayesian GWAS applies Bayesian statistical inference to genome-wide association studies, replacing classical p-value thresholds with Bayes factors and posterior probabilities. This framework naturally incorporates prior knowledge about effect sizes and variant frequencies, quantifies evidence for association on a continuous scale, and supports principled fine-mapping of causal variants within associated loci. It is widely used in complex trait genetics, population genomics, and translational research where uncertainty quantification and multi-variant modeling matter.
Key highlights
- Produces posterior probabilities of association that are directly interpretable and actionable, unlike p-values.
- Naturally incorporates prior information about effect sizes, variant function, and LD structure.
- Enables principled fine-mapping with credible sets, often reducing the causal candidate list from hundreds of LD-linked variants to tens.
- Operates on summary statistics, enabling meta-analysis and multi-study integration without sharing individual-level data.
- Handles multiple causal signals within a locus (multi-causal fine-mapping) more explicitly than stepwise conditional approaches.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use Bayesian GWAS when you need probabilistic, continuously scaled evidence for genetic associations rather than binary significant/non-significant calls; when the goal includes fine-mapping to identify probable causal variants within associated loci; when integrating functional genomic priors (e.g., variant annotation, LD structure) into the association analysis; or when comparing evidence across studies or combining summary statistics without individual-level data access. It is especially valuable for rare-variant analyses and multi-ancestry studies where classical p-value calibration is problematic. Do NOT use Bayesian GWAS as a substitute when you simply need a standard frequentist GWAS for regulatory or pre-specified confirmatory analysis — classical approaches remain the norm for primary discovery reporting in most consortium settings. Avoid it when prior specification is arbitrary and results are highly prior-sensitive without sensitivity analyses.
Strengths & limitations
- Produces posterior probabilities of association that are directly interpretable and actionable, unlike p-values.
- Naturally incorporates prior information about effect sizes, variant function, and LD structure.
- Enables principled fine-mapping with credible sets, often reducing the causal candidate list from hundreds of LD-linked variants to tens.
- Operates on summary statistics, enabling meta-analysis and multi-study integration without sharing individual-level data.
- Handles multiple causal signals within a locus (multi-causal fine-mapping) more explicitly than stepwise conditional approaches.
- Results are sensitive to the choice of prior distributions for effect sizes and prior probability of association; poorly calibrated priors yield misleading PPAs.
- Computation at true genome-wide scale (millions of SNPs, multi-variant models) is demanding; approximate methods trade off accuracy for speed.
- Credible sets assume a single causal variant per locus in many implementations; violations inflate uncertainty.
- Less familiar to reviewers and consortia accustomed to p-value thresholds, potentially complicating publication and replication reporting.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is Bayesian GWAS different from standard GWAS?
Standard GWAS tests each SNP with a frequentist regression and reports a p-value; a threshold (typically 5×10⁻⁸) declares significance. Bayesian GWAS computes a Bayes factor and posterior probability for each SNP, quantifying how much more likely the data are under association versus no association, and incorporating a prior. The output is a ranked probability list rather than a binary significant/non-significant decision, enabling finer prioritization and uncertainty quantification.
Do I need individual-level genotype data to run Bayesian GWAS?
No. Wakefield's approximate Bayes factor (ABF) is computed from GWAS summary statistics — the estimated effect size (beta or log-OR) and its standard error from standard regression. This makes Bayesian fine-mapping accessible from publicly released GWAS summary data without accessing the original cohort.
What is a credible set and how does it differ from a confidence interval?
A 95% credible set is the smallest collection of variants whose posterior probabilities of being causal sum to at least 0.95. It is directly interpretable as 'there is a 95% posterior probability that the true causal variant is in this set.' A classical confidence interval is a frequentist construct that does not carry this direct probability interpretation. Credible sets from well-calibrated Bayesian fine-mapping are often far smaller than the full LD block.
Which tools implement Bayesian GWAS and fine-mapping?
Widely used tools include FINEMAP and SuSiE for multi-causal fine-mapping; CAVIAR and CAVIARBF; coloc and HyPrColoc for colocalisation; and PRS-CS or LDpred2 for Bayesian polygenic scoring. Approximate Bayes factors can be computed from any standard GWAS summary output with simple post-processing scripts.
How do I choose the prior probability of association pi?
A common default is pi = 1 / (number of independent variants tested), placing a prior expectation of roughly one causal variant genome-wide — conservatively small. Empirical estimates from large GWAS suggest pi in the range 10⁻⁵ to 10⁻⁴ for common complex traits. Always report sensitivity analyses across a range of pi values to show that conclusions are robust to prior choice.
Sources
- 1.Stephens, M., & Balding, D. J. (2009). Bayesian statistical methods for genetic association studies. Nature Reviews Genetics, 10(10), 681–690.
- 2.Wakefield, J. (2009). Bayes factors for genome-wide association studies: comparison with P-values. Genetic Epidemiology, 33(1), 79–86.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Bayesian GWAS. ScholarGate. https://scholargate.app/bioinformatics/bayesian-gwas