Probabilistic Fine-Mapping of Genetic Risk Loci

20 concepts19 papers1 proposal

Genome-wide association studies (GWAS) have been remarkably successful at discovering regions of the genome associated with complex diseases. However, a fundamental limitation persists: GWAS identifies genomic neighborhoods, not specific causal variants. Because of a biological phenomenon known as linkage disequilibrium (LD), genetic variants located close to one another tend to be inherited together. Consequently, hundreds of non-causal variants can exhibit identical statistical associations simply by being physically close to the true functional driver.

To transition from statistical association to biological mechanism—and ultimately to therapeutic targets—researchers use statistical fine-mapping. This approach isolates the true causal drivers from the background noise of co-inherited genetic markers.

The Intuition of Bayesian Fine-Mapping

Instead of relying solely on p-values, which remain uniformly inflated across an entire LD block, modern fine-mapping relies on Bayesian probability. The core idea is to combine observed genetic summary statistics with an external reference panel of genetic correlations (the LD matrix) to calculate the Posterior Inclusion Probability (PIP) for each variant. The PIP represents the probability that a specific variant is truly causal.

By calculating these probabilities, researchers can construct a credible set. As defined in work by Ng et al. (2017) and Khan et al. (2024), a 95% credible set is the smallest group of variants that collectively share a 95% probability of containing the true causal mutation. If an LD block contains 200 highly correlated variants, but a Bayesian model concentrates 95% of the posterior probability onto just two variants, the credible set size shrinks from 200 to 2. This drastically reduces the cost and scope of downstream experimental validation.

GWAS Locus (Inflated P-values due to LD)
[ x — x — X — x — x — x — X — x ]  <- Hard to tell which is causal
           |               |
     Bayesian Fine-Mapping & LD Matrix Integration
           v               v
Posterior Probabilities (PIPs)
[ 0.01 — 0.02 — 0.85 — 0.01 — 0.03 — 0.01 — 0.06 — 0.01 ]
                 ^
           Credible Set (PIP = 0.85)

To handle loci where multiple independent causal mutations might be altering the same trait, advanced frameworks like the Sum of Single Effects (SuSiE) algorithm are deployed. As highlighted by Li et al. (2024) and Verma et al. (2024), SuSiE formulates the problem as a sum of simpler single-effect regression challenges, allowing researchers to disentangle overlapping genetic signals that traditional conditional analyses might miss.

Breaking Linkage Disequilibrium via Ancestral Diversity

One of the most effective ways to refine fine-mapping resolution does not come from mathematical adjustments, but from expanding the diversity of the study populations.

Because human populations have distinct evolutionary histories, the boundaries and correlation strengths of LD blocks vary significantly across ancestries. A causal variant will remain associated with the disease across all populations, but the non-causal variants surrounding it will vary in their degree of correlation depending on the ancestry group.

Population A (Large LD Block):   [ Variant 1 — Variant 2 — CAUSAL — Variant 3 ] (All look significant)
Population B (Short LD Block):   [ Variant 1 ] [ Variant 2 — CAUSAL ] [ Variant 3 ]
                                                ^------------^
                                            Shared Signal Isolates Causal Variant

By performing multi-ancestry fine-mapping, researchers can intersect the credible sets derived from different populations. Choquet et al. (2024) utilized this approach to narrow down functional elements in skin cancers, while Ng et al. (2017) leveraged the shorter, more fragmented LD blocks characteristic of African-ancestry genomes to pinpoint adiposity mutations with high precision. Similarly, Petty et al. (2025) integrated multi-omics data within Hispanic/Latino cohorts to uncover previously overlooked causal genes for cardiometabolic traits.

Scaling Up: From Variants to Genes and Multi-Omics

While variant-level fine-mapping identifies specific base-pair changes, diseases ultimately manifest through altered biological pathways, proteins, and genes. The same probabilistic principles are therefore being abstracted upward to intermediate molecular phenotypes.

Gene-Level Fine-Mapping

In Transcriptome-Wide Association Studies (TWAS), researchers predict gene expression levels from genetic data to link expression directly to a disease. However, just as variants are correlated through LD, the expression levels of neighboring genes can also be highly correlated. To resolve this, Rowland et al. (2021) adapted Bayesian fine-mapping to gene-level data, generating credible sets of genes rather than variants. This approach quantifies the exact probability that a specific gene's expression drives the trait, which helps prioritize targets in gene-dense regions as demonstrated by Sun et al. (2021) in the context of Alzheimer's disease susceptibility.

Epigenetic and Metabolic Integration

The framework also scales to other molecular layers, such as DNA methylation and metabolomics. For instance, Sun et al. (2023) applied probabilistic causal fine-mapping to blood methylome data to isolate specific CpG methylation sites driving disease risk from nearby correlated epigenetic marks.

By combining these multi-layered fine-mapping strategies, researchers can construct complete, high-confidence causal chains:

$$\text{Causal Variant} \longrightarrow \text{Epigenetic/Expression Change} \longrightarrow \text{Protein/Metabolite Alteration} \longrightarrow \text{Clinical Phenotype}$$

This systematic prioritization is now widely used across diverse clinical domains, from identifying metabolic processes at the kidney interface (Scherer et al., 2023; Stanzick et al., 2021) and mapping inflammatory skin conditions (Broadaway et al., 2025), to identifying master regulators of tissue function (Pan et al., 2021) and computational drug repurposing pipelines (Grenier et al., 2019; Sng et al., 2025).

History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_causal_variant_identification_8c096544,
  title = {causal variant identification},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/causal-variant-identification/rdr_8c096544e30444a3af6a704433cdb70b},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_causal_variant_identification_8c096544_v1,
  title = {causal variant identification},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/causal-variant-identification/rdr_8c096544e30444a3af6a704433cdb70b/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake