Bayesian eQTL Analysis — Bayesian Expression Quantitative Trait Loci Analysis
Also known as: Bayesian eQTL mapping, probabilistic eQTL analysis, Bayesian QTL mapping for gene expression, eQTL fine-mapping
Bayesian eQTL analysis identifies genetic variants (eQTLs) that regulate gene expression by combining genotype and RNA-seq data within a probabilistic framework. Unlike frequentist approaches that rely on p-value thresholds, the Bayesian formulation produces posterior probabilities of association, enabling principled fine-mapping of causal variants and coherent uncertainty quantification across thousands of gene-SNP pairs simultaneously.
Key highlights
- Produces interpretable posterior probabilities and credible sets instead of arbitrary significance thresholds.
- Handles multiple causal variants per locus (allelic heterogeneity) more naturally than stepwise frequentist approaches.
- Provides coherent uncertainty quantification that propagates through downstream colocalisation and Mendelian randomization analyses.
- Can incorporate functional genomic priors to increase power to detect causal regulatory variants.
- Automatically accounts for multiple testing through the prior, reducing the need for ad hoc correction procedures.
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 eQTL analysis when the goal is fine-mapping — identifying the most probable causal regulatory SNP rather than simply detecting that a locus is associated. It is especially appropriate when sample sizes are moderate (200–1000 individuals), when multiple causal variants per locus are plausible, or when the results will be colocalized with GWAS summary statistics. Do not use it as a drop-in replacement for standard eQTL pipelines (e.g., Matrix-eQTL or FastQTL) when computational resources are limited and discovery breadth rather than fine-mapping precision is the primary goal; in that case, run a frequentist discovery step first and apply Bayesian fine-mapping only to significant loci.
Strengths & limitations
- Produces interpretable posterior probabilities and credible sets instead of arbitrary significance thresholds.
- Handles multiple causal variants per locus (allelic heterogeneity) more naturally than stepwise frequentist approaches.
- Provides coherent uncertainty quantification that propagates through downstream colocalisation and Mendelian randomization analyses.
- Can incorporate functional genomic priors to increase power to detect causal regulatory variants.
- Automatically accounts for multiple testing through the prior, reducing the need for ad hoc correction procedures.
- Computational cost for MCMC-based methods can be prohibitive at whole-genome scale; variational approximations (SuSiE) mitigate but do not eliminate this.
- Results depend on prior specification; poorly calibrated priors on effect size or causal variant proportion can bias posterior probabilities.
- Requires accurate linkage disequilibrium (LD) reference panels; mismatched LD structure inflates false-discovery rates in fine-mapping.
- Trans-eQTL analysis remains challenging due to the vastly larger search space and smaller expected effect sizes.
- Less familiar to reviewers accustomed to p-value thresholds, requiring more explanation in manuscripts.
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 eQTL analysis different from standard (frequentist) eQTL analysis?
Standard eQTL analysis (e.g., Matrix-eQTL) tests each SNP–gene pair with a linear regression and reports a p-value, then applies a multiple-testing correction such as Bonferroni or FDR. Bayesian eQTL analysis instead computes a posterior probability of association using a prior on effect sizes, making multiple-testing correction implicit. The key practical difference is in fine-mapping: Bayesian methods output credible sets and posterior probabilities for each SNP being causal, which frequentist p-values cannot directly provide.
What sample size is needed for Bayesian eQTL fine-mapping?
Discovery of cis-eQTLs is feasible from around 100–200 samples, but credible sets narrow to single-SNP resolution typically only with 500 or more samples. Trans-eQTL mapping requires thousands of samples due to much smaller expected effect sizes. The GTEx consortium (v8: ~700 donors) demonstrated that Bayesian fine-mapping at the tissue-specific level is informative in the 500–1000 sample range.
Which software implements Bayesian eQTL fine-mapping?
SuSiE (Sum of Single Effects; Wang et al. 2020) is the most widely used tool for multi-causal Bayesian fine-mapping and integrates with coloc for colocalisation. FINEMAP (Benner et al. 2016) uses stochastic search. FastBMA provides scalable Bayesian model averaging. All three can operate on summary statistics plus an LD reference panel, enabling analysis without individual-level data.
Can Bayesian eQTL analysis handle multiple causal variants per gene?
Yes, and this is one of its principal advantages. Methods such as SuSiE explicitly model L independent causal signals simultaneously. Frequentist stepwise conditioning can also detect multiple signals, but it does not quantify uncertainty over the number of causal variants; Bayesian methods report posterior probabilities for each signal and construct separate credible sets for each, giving a richer and more honest picture of allelic heterogeneity.
How do I colocalize Bayesian eQTL results with GWAS?
The coloc R package (Giambartolomei et al. 2014; Wallace 2021) is the standard tool. It takes the posterior probabilities or Bayes factors from both the eQTL and GWAS analyses over a shared genomic region and estimates the posterior probability that the two traits share a single causal variant (H4). SuSiE-coloc extends this to handle multiple causal variants per locus, reducing false-negative colocalisation caused by allelic heterogeneity.
Sources
- 1.Stephens, M., & Balding, D. J. (2009). Bayesian statistical methods for genetic association studies. Nature Reviews Genetics, 10(10), 681–690.
- 2.Guan, Y., & Stephens, M. (2011). Bayesian variable selection regression for genome-wide association studies and other large-scale problems. Annals of Applied Statistics, 5(3), 1780–1815.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Bayesian eQTL analysis. ScholarGate. https://scholargate.app/bioinformatics/bayesian-eqtl-analysis