From a Noncoding Credible Set to a Regulatory Hypothesis
A noncoding association signal can point to a locus without identifying which variant changes which gene. Annotation helps build and compare possible mechanisms: a variant may lie in a regulatory element, the element may affect a gene, and the gene’s activity may influence the trait. Each link needs evidence of its own. A credible-set variant overlapping an active enhancer is not, by itself, proof that the variant changes enhancer function or that the enhancer regulates a particular gene.
This tutorial works through a hypothetical locus associated with a circulating inflammatory trait. The exercise is to compare plausible regulatory explanations across assays and tissues, not to declare a gene causal from annotation alone. The fine-mapping model in the example assumes one causal variant at the locus and supplies posterior probabilities for candidate variants. That assumption matters: in a model allowing multiple causal variants, posterior inclusion probabilities need not sum to one.
The useful question is therefore not simply “Which variant has the strongest annotation?” Instead, ask what the fine-mapping results leave plausible, which variants map to which elements, what evidence connects those elements to genes, and whether the assays were measured in a biologically relevant context. Keep these evidence steps distinct when forming a conclusion.
A Hypothetical Locus and Its Evidence
Suppose the credible set contains four noncoding variants. Their PIPs sum to 0.95, so the set has 95% posterior coverage under the assumed single-causal-variant model. The remaining 0.05 is posterior mass outside the set. Two candidate genes, GENE-A and GENE-B, are expressed in different potentially relevant cell types. The table records hypothetical assay results; “not measured” means that the relevant assay or tissue data are unavailable, not that the variant lacks activity.
| Variant | PIP | Regulatory-element and assay observations | Gene-link and tissue observations |
|---|---|---|---|
| v1 | 0.42 | Overlaps element E1; accessible and marked by H3K27ac in stimulated monocytes | E1 is linked to GENE-A and has a chromatin contact with its promoter in stimulated monocytes |
| v2 | 0.28 | Also overlaps E1; accessible and marked by H3K27ac in stimulated monocytes | Same E1-to-GENE-A link and contact evidence as v1 |
| v3 | 0.15 | Overlaps a separate element, E2; accessible and marked by H3K27ac in hepatocytes | E2 is linked to GENE-B in hepatocytes; no corresponding contact is detected in the assayed map |
| v4 | 0.10 | Intronic; relevant accessibility and histone-mark tracks are not measured in the available data | Enhancer-link and contact evidence are not measured in the relevant tissues |
Accessibility indicates that DNA is available to regulatory proteins in the assayed cells; H3K27ac is commonly associated with active regulatory elements. An enhancer-to-gene link provides a proposed connection between an element and a gene. A chromatin contact supports spatial proximity in the assayed cells, but proximity alone does not establish a functional effect. Tissue context helps assess whether an observed signal is relevant to the trait, but it does not settle the mechanism.
Worked Example: Quantifying Posterior Mass in One Regulatory Element
E1 overlaps v1 and v2. What is the total posterior mass assigned to variants in E1, and what fraction of the posterior mass in the 95% credible set does that represent? Assume the supplied PIPs refer to mutually exclusive possibilities for the single causal variant.
The posterior mass for a causal variant in E1 is the sum of the PIPs for its overlapping variants:
Thus, 0.70 of the full posterior mass is assigned to v1 or v2, which overlap E1. Conditional on the causal variant being inside the credible set, the corresponding fraction is:
Under the stated model, this is about 73.7% of the posterior mass conditional on the causal variant being in the set. It is not a 73.7% probability that E1 is functional, that GENE-A is the target, or that the proposed mechanism is true. The calculation groups fine-mapping probability by overlap; functional and gene-link evidence must still be assessed separately.
Compare the Evidence Along the Proposed Mechanism
A regulatory hypothesis has several links: a variant may alter a regulatory element, the element may affect a target gene, and altered gene activity may contribute to the trait. The available assays do not all address the same link. Accessibility and histone marks characterize an element’s state in a tissue. Enhancer-link and contact evidence address possible element-to-gene relationships. Fine-mapping PIPs describe variant-level support under the statistical model. Tissue context bears on relevance. Treating these as separate evidence types helps avoid turning several measurements of one biological feature into several independent confirmations.
Identify which credible-set variants carry substantial PIP. Here v1 and v2 together carry 0.70 posterior mass and both overlap E1.
E1 is accessible and H3K27ac-marked in stimulated monocytes; E2 has those observations in hepatocytes. These results support activity in the measured tissues, not variant-specific function.
E1 has both an enhancer link and a detected contact to GENE-A in stimulated monocytes. E2 has an enhancer link to GENE-B in hepatocytes, while a contact is not detected in the assayed map.
Compare the full evidence chains and their tissue contexts. Record absent, negative, and unmeasured results accurately, and state what remains unresolved.
Worked Example: Comparing Two Candidate Regulatory Mechanisms
Compare the proposed mechanisms “v1 or v2 alters E1, which regulates GENE-A in stimulated monocytes” and “v3 alters E2, which regulates GENE-B in hepatocytes.” Which has the stronger annotation support, and why?
For the GENE-A hypothesis, v1 and v2 carry \(0.42+0.28=0.70\) posterior mass and both overlap E1. E1 is accessible and H3K27ac-marked in stimulated monocytes, and the available data include both an E1-to-GENE-A link and a contact with the GENE-A promoter in that tissue. These observations form a comparatively coherent chain from higher-PIP variants to an active element and then to a candidate gene in one context.
For the GENE-B hypothesis, v3 carries 0.15 PIP and overlaps E2, which is accessible and H3K27ac-marked in hepatocytes. E2 has a reported link to GENE-B in hepatocytes, so this remains a plausible alternative. However, the assayed chromatin-contact map does not detect a corresponding contact. That is a measured non-detection in that map, not proof that a contact never occurs or that the link is false.
On these annotations, the GENE-A mechanism has stronger overall support, particularly because more posterior mass overlaps its candidate element and multiple types of gene-link evidence are available in stimulated monocytes. This comparison is not a formal probability calculation for either gene. The observations do not establish that v1 or v2 changes E1 activity, that the contact is regulatory, or that GENE-A mediates the trait association. GENE-B remains a reasonable alternative supported by a distinct variant and tissue context.
A Short Justification for Keeping the Claims Separate
The posterior-mass calculation above follows from the single-causal-variant assumption. Let \(C\) denote the event that the causal variant is in the credible set, and let \(E\) denote the event that it is one of the variants overlapping E1. Since v1 and v2 are distinct candidate variants within E1, their mutually exclusive posterior probabilities add:
Because E is contained in C, the conditional probability is \(P(E\mid C)=P(E)/P(C)\). Here \(P(C)=0.95\), giving \(0.70/0.95=0.7368\). This is a statement about where the causal variant falls under the fine-mapping model. It does not condition on, or incorporate, the functional assay results. To infer a gene-level probability would require a separate, explicitly specified model for how variant, element, gene, and assay evidence relate. The table alone does not provide such a model.
Worked Example: Handling Missing Evidence Without Calling It Negative
For v4, the relevant accessibility, histone-mark, enhancer-link, and contact data were not measured. What posterior mass is not resolved by the described annotations, if the residual 0.05 outside the credible set also has no annotation summary?
The in-set variant with unmeasured relevant assays is v4, with PIP 0.10. Adding the residual posterior mass outside the set gives:
Thus 0.15 posterior mass is not resolved by the described annotations in this example. This arithmetic does not show that these variants are nonfunctional; their relevant assays are unavailable, and the residual mass is outside the summarized set. It also does not mean that the remaining 0.85 has a known mechanism. v3 has some supporting annotations but still has uncertainty about its gene connection and the functional effect of the variant.
Interpreting the Integrated Result
A useful locus interpretation reports both the leading hypothesis and the boundaries of what the evidence supports. For this hypothetical locus, one defensible summary is: “The annotations favor a regulatory mechanism involving E1 and GENE-A in stimulated monocytes, because v1 and v2 account for 0.70 posterior mass within the credible set and E1 has activity and gene-link evidence in that context. A hepatocyte E2-to-GENE-B mechanism remains plausible through v3. The available data do not establish which variant changes element function, whether either gene mediates the trait, or what mechanism applies to unmeasured v4 and the residual posterior mass.” This communicates a relative comparison without presenting annotation as proof.
The interpretation should reflect assay design. A contact not detected in a particular map is not equivalent to an experiment demonstrating no regulatory relationship. A missing track is not a negative result. An enhancer link may be indirect or method-dependent, while accessibility and histone marks can be correlated indicators of element activity. Multiple observations from the same element and tissue therefore need not be independent evidence. Earlier material on correlated annotations and missingness is useful when deciding how much weight to place on such patterns.
Tissue matching also requires restraint. A tissue can be biologically plausible for a trait and still not be the causal context, and an assay in a relevant tissue can be incomplete or use a cell state that differs from the one involved in the association. In this example, stimulated monocytes and hepatocytes offer two distinct contexts. The coherent set of observations for GENE-A makes it the stronger annotation-based hypothesis here, but does not rule out the GENE-B route or another mechanism.
Common Pitfalls
- Equating overlap with function. The mistaken claim is “v1 overlaps an active enhancer, so v1 changes enhancer activity.” Overlap and element activity do not demonstrate a variant-specific effect; that requires evidence addressing the effect of the allele or a suitable perturbation.
- Counting related assays as independent confirmations. Accessibility and H3K27ac at E1 both support an active regulatory state, but they are not necessarily independent. The contact and enhancer-link observations also need to be interpreted in light of how they were obtained.
- Calling unmeasured data negative. No available track for v4 does not mean that v4 lies outside an active element. State whether a signal was detected, not detected in a specified assay, or not measured.
- Turning a variant-level posterior into a gene-level probability. The 0.70 mass for v1 and v2 is not a 70% probability for GENE-A. It describes posterior mass on variants under the stated model; gene attribution requires additional assumptions and evidence.
Check Your Understanding
Use the hypothetical locus and the principles above to answer these questions.
- Under the stated single-causal-variant model, what posterior mass is assigned to variants overlapping E1, and what is that mass conditional on the causal variant being inside the 0.95 credible set?
- Which observations support an E1-to-GENE-A hypothesis, and which part of the proposed mechanism remains unproven?
- Why is the absence of a detected E2 contact not equivalent to evidence that E2 cannot regulate GENE-B?
- What does the 0.10 PIP for v4 mean when the relevant tissue assays were not measured?
- Why should the 0.70 posterior mass for v1 and v2 not be reported as a 70% probability that GENE-A is causal?