Fine-Mapping

The central challenge fine-mapping solves is linkage disequilibrium (LD)—the non-random correlation between nearby variants. A significant GWAS signal often spans dozens or hundreds of SNPs in high LD, but only one or a few are truly causal. Fine-mapping uses an external LD reference panel (e.g., 1000 Genomes Project) to explicitly model this correlation matrix and statistically tease apart the independent signals.

• Distinguishes tag SNPs (markers) from causal variants
• Requires accurate population-matched LD estimates
• Sensitive to mismatches between GWAS and reference panel ancestry

The primary output of a fine-mapping analysis is a credible set—the minimal set of variants that, with a specified posterior probability (typically 95%), contains the true causal variant. This transforms a long list of associated SNPs into a manageable, experimentally tractable number of candidates for functional validation.

• A 95% credible set might contain 1–30 variants depending on LD complexity
• Variants within the set are ranked by Posterior Inclusion Probability (PIP)
• A single-variant credible set indicates high-resolution mapping success

Multiple algorithmic frameworks exist, each making different assumptions about the underlying genetic architecture of the locus:

• Approximate Bayes Factor (ABF): Computes a Bayes Factor for each variant independently using effect size and standard error from summary statistics; fast but assumes a single causal variant per locus
• SuSiE (Sum of Single Effects): Fits multiple single-effect regression models iteratively, allowing for multiple causal variants with distinct effect sizes
• FINEMAP: Uses a shotgun stochastic search algorithm to explore the model space of causal configurations efficiently
• PAINTOR: Integrates functional annotation data (e.g., chromatin accessibility) to prioritize variants with both statistical and biological evidence

Modern fine-mapping tools incorporate functional genomics data to break ties between statistically equivalent variants. By weighting variants based on their overlap with regulatory elements, DNase I hypersensitivity sites, or expression quantitative trait loci (eQTLs), the resolution of credible sets improves dramatically.

• Annotations serve as prior probabilities in Bayesian frameworks
• Common resources: ENCODE, Roadmap Epigenomics, GTEx
• A variant with high PIP and strong functional annotation is the top candidate for CRISPR validation

A foundational step preceding formal fine-mapping is conditional analysis, often performed with tools like GCTA-COJO. This iteratively tests for association at a locus while conditioning on the top signal, revealing whether multiple independent causal variants exist at the same locus—a phenomenon known as allelic heterogeneity.

• Stepwise conditioning identifies secondary, tertiary, and further independent signals
• Distinguishes a single strong causal variant from multiple moderate ones
• Essential for accurately modeling the number of causal variants in fine-mapping priors

Fine-mapping is often paired with colocalization analysis to determine whether the same causal variant driving a GWAS trait also regulates gene expression or protein levels in relevant tissues. This connects a disease-associated locus to a specific effector gene and mechanism.

• Coloc tests the hypothesis of a shared causal variant between GWAS and eQTL signals
• A high posterior probability for colocalization (H4 > 0.8) nominates the target gene
• Bridges the gap from statistical association to biological mechanism and drug target identification

The non-random association of alleles at different loci, which is the central confounding factor that fine-mapping aims to resolve. LD patterns arise from population history, mutation, and recombination rates. High LD means multiple variants are statistically correlated, making it impossible to distinguish the true causal variant from its proxies using single-marker tests alone.

• r² and D': Common metrics quantifying LD strength between variant pairs
• LD Blocks: Genomic regions where variants are highly correlated due to low historical recombination
• Fine-mapping explicitly models the local LD matrix to assign posterior inclusion probabilities to each variant

The primary output of a fine-mapping analysis: a set of genetic variants that collectively contain the true causal variant with a specified probability, typically 95%. A credible set narrows a GWAS locus from hundreds of correlated variants down to a handful of high-confidence candidates.

• A 95% credible set means there is a 95% probability that at least one variant in the set is causal
• Posterior Inclusion Probability (PIP) : The probability that a specific variant is causal, used to rank variants within the set
• Smaller credible sets indicate higher resolution and more successful fine-mapping

Aggregated association results from a genome-wide association study that serve as the primary input for most fine-mapping methods. These include the effect allele, beta coefficient, standard error, and p-value for each variant tested.

• Fine-mapping tools like FINEMAP, SuSiE, and PAINTOR operate directly on summary statistics plus an LD reference panel
• Using summary data eliminates the need for individual-level genotypes, enabling meta-analysis across large consortia
• Winner's Curse: The overestimation of effect sizes at significantly associated loci, which can bias fine-mapping if uncorrected

The specific genetic variant that directly alters gene function, regulation, or expression to influence the phenotype. Fine-mapping aims to distinguish this variant from the many non-functional variants in LD with it.

• Coding variants: Change amino acid sequences and are often prioritized by functional annotation
• Regulatory variants: Reside in promoters, enhancers, or other non-coding elements and affect gene expression levels
• Integrating functional genomics data (e.g., ATAC-seq, ChIP-seq, eQTLs) with statistical fine-mapping improves causal variant identification

The statistical framework underlying modern fine-mapping methods. Instead of testing one variant at a time, Bayesian approaches evaluate all possible combinations of causal variants simultaneously, assigning a posterior probability to each configuration.

• Spike-and-slab priors: Assume a small fraction of variants have non-zero effects, matching the sparse genetic architecture of most traits
• Stochastic search: Algorithms like MCMC explore the vast model space of possible causal configurations
• Methods like SuSiE use a sum of single-effects model to efficiently approximate the posterior without exhaustive search

External genomic data used to weight or prioritize variants based on their likely biological function. Integrating annotations into fine-mapping improves power by upweighting variants in functionally relevant regions.

• Conservation scores: Identify bases under purifying selection across species
• Chromatin state maps: Mark active promoters, enhancers, and transcribed regions in relevant cell types
• PAINTOR and PolyFun incorporate functional annotations directly into the fine-mapping prior, enabling functionally-informed credible sets