Why Mechanism Claims Need Calibration
A genetic association can identify a genomic locus without identifying the causal nucleotide, the functional element, the affected gene, or the molecular process connecting that gene to the trait. Those are distinct questions. An accessible chromatin peak near a credible-set variant is evidence about a possible regulatory element; it does not, by itself, show that the variant changes that element, which gene it regulates, or how that gene affects the trait.
A useful interpretation therefore separates claims into levels and records what each observation actually supports. This is not a way to weaken every conclusion. It is a way to make strong conclusions when the evidence warrants them, while keeping uncertainty visible when it does not. In particular, evidence can be persuasive for one link in a proposed mechanism and weak or absent for another.
An Evidence Table for Locus Interpretation
Build the table around the claim, not around a list of assays. For each proposed link, record the observation, the claim it supports, its limitations, and the wording justified by the evidence. This prevents a familiar error: taking a collection of annotations and writing a single mechanistic sentence that implies every link has been demonstrated.
| Claim level | Examples of relevant evidence | What the evidence does not establish on its own | Calibrated wording |
|---|---|---|---|
| Variant | Fine-mapping support; allele-specific molecular measurements; variant-specific editing | That the variant acts through a particular element, gene, or trait pathway | “Prioritizes variant v”; “editing v altered the measured outcome in this system” |
| Element | Accessibility or histone marks; overlap with candidate variants; perturbation of the element | That a particular variant is functional or that a particular gene is its target | “Supports enhancer activity in the assayed cell type”; “perturbing the region changed expression” |
| Target gene | Chromatin contact; molecular QTL evidence; expression response after element perturbation | That the associated variant caused the gene-level effect, or that this effect mediates the trait | “Nominates gene G as a candidate target”; “supports an effect on G in the tested context” |
| Mechanism | Convergent evidence linking a variant, element or molecular process, gene, and relevant phenotype | Unmeasured links, untested contexts, or alternative causal routes | “The evidence supports a model in which…”; state remaining alternatives explicitly |
The table distinguishes measurement from interpretation. A contact assay measures a form of spatial proximity under its experimental conditions; a contact is not itself proof of regulatory action. A QTL reports a statistical relationship between genotype and a molecular phenotype in a sampled population; overlap with a trait locus is functional support, but does not automatically show that the same causal variant drives both signals. A perturbation can show that changing a region affects a measured outcome in the tested system, but the result may not identify which naturally occurring allele matters.
Also record the context of each observation: tissue or cell type, developmental or stimulation state, assay resolution, and coverage. A result in a cell type that is not relevant to the trait may be biologically interesting but less direct support for the proposed in-context mechanism. Conversely, an assay with no signal is not strong evidence against a mechanism if the relevant cell type was not tested, the region was poorly measured, or the assay had limited sensitivity.
Worked Example: Annotation Overlap Does Not Identify the Causal Variant
Worked Example: Annotation Overlap Does Not Identify the Causal Variant
Suppose a hypothetical fine-mapping analysis reports four variants in a credible set, with posterior probabilities 0.40, 0.30, 0.20, and 0.10. The first two overlap the same accessible enhancer in a relevant cell type; the other two do not. The enhancer also has an active histone mark, and a contact assay reports contact with gene G.
Variant-level calculation. The posterior mass assigned to variants overlapping the enhancer is the sum of their reported probabilities:
Under the assumptions of this fine-mapping analysis, 70% of the reported posterior probability is assigned to variants that overlap the annotated enhancer. That is a useful summary of overlap, not a 70% probability that the enhancer is causal or that a specific variant acts through G. The two overlapping variants remain distinct candidates, and the annotation does not tell us which, if either, changes enhancer function.
Element-level conclusion. Accessibility and the histone mark support the interpretation that the region has regulatory activity in the assayed cell type. They do not demonstrate that either overlapping variant alters that activity.
Target-gene conclusion. Contact with G makes G a candidate target. Contact alone does not show that the enhancer regulates G, and it does not rule out other targets.
A calibrated summary is: “Most of the reported fine-mapping posterior mass overlaps an active candidate enhancer in the assayed cell type. The enhancer contacts G, nominating G as a possible target, but variant-specific regulatory activity and the target-gene link remain unresolved.” This wording preserves the evidence without converting annotation overlap into a complete mechanism.
Reasoning Across the Proposed Chain
A mechanism claim is a connected claim, not merely a list of individually plausible observations. Let \(V\) denote the event that the proposed variant is causal, \(E\) that the proposed element mediates its effect, \(G\) that the proposed gene is affected, and \(R\) that the proposed route links the molecular change to the trait. The full mechanism requires all of these links to hold together. If probabilities for the component events were available under a common model, then:
The inequality follows because the joint event \(V \cap E \cap G \cap R\) is contained within each individual event. Thus, support for one link cannot establish the entire chain. In practice, annotation evidence is often not expressed as comparable probabilities, so the equation is a logical reminder rather than a recipe for combining assay results. In particular, multiplying support from correlated annotations as if they were independent can exaggerate confidence.
A practical review asks, link by link: What was directly measured? What inference connects that measurement to the claim? What alternative explanations remain? If an evidence table has strong entries for element activity and gene response but no variant-specific experiment, describe the element-to-gene evidence as stronger than the variant-to-element evidence. Do not let a strong result at one level conceal a weak link elsewhere.
Name the candidate variant, element or process, gene, and trait-related consequence. Keep each link explicit rather than compressing the proposal into “the locus acts through G.”
For example, chromatin accessibility supports element activity; a perturbation response supports a functional effect of the perturbed region in that assay; neither alone identifies a causal allele.
Note the assayed cell type, assay resolution and coverage, and whether the observation is variant-specific, region-level, or gene-level.
Use “consistent with,” “supports,” or “nominates” for indirect evidence. Reserve direct wording for the effect actually measured, and state unresolved links and alternatives.
Worked Example: A Regional Perturbation Supports an Element Effect
Worked Example: A Regional Perturbation Supports an Element Effect
In a hypothetical cell model, researchers use CRISPR interference at a candidate enhancer and measure expression of nearby gene G. The normalized mean expression is 1.00 in control samples and 0.58 after enhancer targeting. Assume the values are normalized to the same reference and the comparison is valid for this experiment.
The relative reduction is:
Expression of G is 42% lower in the targeted condition than in the control condition. If multiple appropriately controlled guides targeting the region give a consistent result, this supports the claim that perturbing the region can affect G expression in this cell model.
It does not establish that a particular credible-set variant causes the effect. CRISPR interference suppresses a region rather than replacing one allele with another, and the perturbation may affect local chromatin or more than one regulatory function. Nor does the expression change alone prove that altered G expression mediates the genetic association with the trait.
A calibrated conclusion is: “In the tested cell model, CRISPR interference of the candidate enhancer reduced G expression by 42%, supporting a functional enhancer-to-G link in that system. Whether a specific associated variant perturbs this link, and whether the expression change mediates the trait association, remains to be tested.” The arithmetic supports the stated effect size, while the wording keeps the experiment’s scope clear.
Worked Example: Conflicting Contexts and an Unresolved Target
Worked Example: Conflicting Contexts and an Unresolved Target
Consider a hypothetical association near two genes, G1 and G2. A candidate variant lies in an accessible region in cell type A. A chromatin-contact assay in cell type A reports contact with G1. In cell type B, a molecular QTL analysis reports a genotype-expression association for G2, but the variant driving that association has not been shown to be the same as the trait-associated causal variant. No perturbation experiment has been performed.
The accessibility result supports the possibility that the region is regulatory in cell type A. The contact result nominates G1 in that context, but does not by itself demonstrate a regulatory effect on G1. The QTL supports a genotype-expression relationship involving G2 in cell type B, but the shared mechanism is uncertain because the causal variant has not been resolved as the same and the contexts differ. The observations therefore do not justify selecting one gene as the established target.
A suitable evidence-table summary would list G1 under “contact-based candidate target in cell type A” and G2 under “QTL-supported candidate in cell type B; shared causal variant unresolved.” It would also record that neither gene has direct perturbation support in the stated evidence.
A calibrated conclusion is: “The annotations nominate G1 and G2 through different evidence and contexts. Current evidence does not resolve the target gene or establish a single variant-to-gene mechanism.” Calling G1 the target solely because the contact assay detected it, or calling G2 the target solely because of the QTL, would overstate what those observations show.
Choosing Words That Match Evidence
Mechanism language should indicate both the evidential strength and the scope of the claim. “Overlaps” is a coordinate statement. “Is consistent with” indicates compatibility without uniquely favoring the proposed explanation. “Supports” indicates evidence in favor of a claim while allowing alternatives. “Nominates” identifies a candidate for follow-up, not a settled answer. “Demonstrates” should be tied to the specific outcome directly measured, in the tested assay and context.
For example, a carefully bounded statement may say that allele-specific editing changed expression of a gene in a particular cell model. It should not silently expand that finding into proof that the variant explains the trait association in all relevant tissues. Even strong, convergent evidence can leave alternative routes open, such as another target gene, another relevant cell type, or an effect on a molecular phenotype not measured in the experiment.
Why This Discipline Matters
Calibrated claims make locus interpretation more useful, not less decisive. They show which links have direct experimental support, which depend on annotation or statistical association, and which require further testing. That distinction guides follow-up: variant-specific editing addresses a different uncertainty from perturbing a whole region, while testing another relevant cell type addresses a different limitation from identifying a target gene.
A concise report can still give a clear interpretation. State the strongest supported claim first, identify the evidence behind it, and name the most consequential unresolved alternative. When evidence supports only a candidate element or gene, say so plainly. When an assay directly demonstrates an effect, specify the measured effect and the experimental context instead of generalizing beyond them.
Check Your Understanding
For each question, distinguish the claim supported by the observation from any stronger claim that remains unresolved.
- A credible-set variant overlaps an accessible region. What does this establish about the variant, and what does it suggest about the region?
- Why does a chromatin contact between an enhancer and a gene nominate a target without proving that the enhancer regulates that gene?
- A regional CRISPRi perturbation changes gene expression. Name one conclusion supported by this result and one variant-level conclusion it does not establish.
- Why should a QTL association in one cell type be reported separately from a contact result in another cell type?
- Rewrite this claim more cautiously: “The causal variant activates an enhancer, which upregulates G and causes the trait,” when the evidence consists only of variant overlap, accessibility, and contact with G.