Enhancer-to-Gene Mapping with Regulatory Links

From a Noncoding Variant to a Candidate Gene

A noncoding variant can lie far from the gene it may affect. Its position alone therefore does not identify a target: a variant may overlap an enhancer, and that enhancer may regulate a distant gene, several genes, or none of the genes proposed by a particular assay. Enhancer-to-gene mapping uses regulatory evidence to make this connection more specific than a distance window can, but it remains an inference with uncertainty.

The key is to keep two questions separate. First, is the variant plausibly involved in a regulatory element? Second, does that element regulate the gene in the relevant biological context? Coordinates help identify overlap; chromatin contacts and regulatory-link models provide evidence about enhancer-to-gene connections. Neither kind of evidence, by itself, proves that the variant changes the gene’s expression or causes a trait.

Definition: An enhancer-to-gene link is an evidence-based proposed regulatory connection between a genomic enhancer interval and a gene, often represented by the gene’s promoter or transcription start site. A variant-to-gene mapping connects a variant to a gene through one or more such elements. The mapping is an evidential link, not a demonstration of a causal chain.

Representing the Regulatory Chain

A common chain is variant $i$ → enhancer $e$ → gene $g$. The variant must first overlap, or otherwise plausibly affect, the enhancer. The enhancer must then be linked to the gene. These are distinct steps: strong evidence for one does not automatically establish the other.

Let $r_{ie}$ denote the probability or weight that variant $i$ acts through enhancer $e$, conditional on the variant being causal. It can reflect overlap and uncertainty in the enhancer annotation. Let $q_{eg}$ denote the probability that enhancer $e$ regulates gene $g$, conditional on the stated assay, cell type, and model. A calibrated $q_{eg}$ can be used as a probability; an uncalibrated score should instead be described as a relative measure of support.

Definition: Under a specified model, the regulatory mapping weight from variant $i$ to gene $g$ is $$ m_{ig}=\sum_e r_{ie}q_{eg}. $$ This expression treats the enhancer assignments represented by $r_{ie}$ as mutually exclusive routes for the variant. It is a probability of mapping through the included routes only if the weights have probabilistic interpretations and the model’s conditional assumptions are appropriate. It need not sum to one across genes: an enhancer can regulate more than one gene.

Coordinates determine whether a variant falls inside a defined interval, subject to consistent genome build, chromosome naming, and coordinate conventions. For example, one data source may use zero-based, half-open intervals while another uses one-based, closed intervals. A one-base discrepancy can change an apparent overlap at an interval boundary. Always harmonize coordinates before interpreting overlaps.

Chromatin-contact assays measure physical proximity or contact between genomic regions, not gene regulation directly. A contact between an enhancer and a promoter supports a possible connection, but contact may occur without a measurable regulatory effect. Other link evidence may use correlated activity across samples, sequence or chromatin features, or trained prediction models. The meaning of a score depends on how it was constructed and validated.

Worked Example: Checking an Enhancer Overlap and Its Linked Genes

Suppose a hypothetical variant lies at chr7:55,100,145 on a harmonized genome build. An assay defines an enhancer at chr7:55,100,100–55,100,200, using one-based closed coordinates. The assay reports calibrated enhancer-to-gene link probabilities of 0.72 for Gene A and 0.18 for Gene B in the tissue under study. What does this evidence support?

Step 1: Check overlap. The variant coordinate satisfies $55{,}100{,}100 \leq 55{,}100{,}145 \leq 55{,}100{,}200$, so it falls inside the enhancer interval.

Step 2: Interpret the links. Conditional on this enhancer and the assay’s calibration and context, the reported link probability is 0.72 for Gene A and 0.18 for Gene B. Gene A has stronger link support in this assay. The values are not probabilities that the variant causes a trait through either gene.

Conclusion. The coordinate overlap and enhancer-to-gene results support considering Gene A as a regulatory target, with weaker support for Gene B. They do not establish that the variant changes either gene or that either gene mediates a trait association.

Link Scores, Probabilities, and Uncertainty

A score is not automatically a probability. A model might report a score of 8.4, a rank, or a value between zero and one. Only a score calibrated to the probability of a clearly defined outcome can be interpreted as a probability. A high uncalibrated score can rank one link above another under that model, but it does not mean there is an 84% chance that the enhancer regulates the gene.

Uncertainty can arise from the enhancer boundary, assay noise, cell-type mismatch, model calibration, or limited validation. Preserve this uncertainty rather than treating every proposed link as certain. If a variant could act through multiple enhancer elements, the $r_{ie}$ weights should represent the uncertainty over those routes. If the link itself is uncertain, $q_{eg}$ should reflect that uncertainty when it is calibrated to do so.

If variant $i$ has posterior probability $\pi_i$ of being causal, a simple expected contribution through the regulatory chain is $\pi_i m_{ig}$. This uses the variant’s causal probability and the conditional mapping weight; it does not turn the mapping into proof. Its use depends on having meaningful inputs and a defensible model for the conditional relationships.

Worked Example: Combining Two Possible Enhancer Routes

Consider a hypothetical variant with causal posterior probability $\pi_i=0.40$. Conditional on being causal, suppose the evidence assigns weights $r_{i1}=0.60$ and $r_{i2}=0.40$ to two alternative enhancer routes. Calibrated link probabilities to Gene C are $q_{1C}=0.70$ and $q_{2C}=0.25$. Assume the two enhancer routes are mutually exclusive in this model and that the link probabilities apply to the same relevant context.

The conditional mapping weight is $$ m_{iC}=(0.60)(0.70)+(0.40)(0.25)=0.42+0.10=0.52. $$

The variant’s expected contribution to a causal-through-Gene-C mapping under these assumptions is $$ \pi_i m_{iC}=(0.40)(0.52)=0.208. $$

This value is a model-based contribution, not a 20.8% chance that Gene C causes the trait. The interpretation depends on the causal posterior, the enhancer-route weights, the calibrated link probabilities, and the assumed chain. If the route weights were merely arbitrary scores, the arithmetic would not have a probability interpretation.

Why the Weighted Expression Works

The mapping expression follows from conditioning on the enhancer route. For a fixed variant $i$, consider the event that it is causal and maps to gene $g$ through enhancer $e$. Under the model, its probability is the probability the variant is causal, multiplied by the conditional probability of the enhancer route, multiplied by the conditional link probability:

P(C_i=1,E_i=e,L_{eg}=1)=\pi_i r_{ie}q_{eg}.

Here $C_i$ indicates that the variant is causal, $E_i=e$ denotes its enhancer route, and $L_{eg}$ denotes a regulatory link. This factorization assumes that $r_{ie}$ and $q_{eg}$ are defined conditional on the preceding steps and that the link probability applies to the relevant context. Summing over the mutually exclusive enhancer routes gives

P(C_i=1,\text{ maps through an included enhancer to }g) =\sum_e \pi_i r_{ie}q_{eg} =\pi_i m_{ig}.

The equality uses the addition rule for mutually exclusive events. If enhancer routes are not mutually exclusive, directly summing their probabilities can double-count cases; the model must instead represent their joint structure or define a suitable non-overlapping set of routes. Likewise, uncertainty or dependence among assays should not be ignored merely because several scores are available.

Worked Example: Interpreting a Context-Specific Link Estimate

Suppose a hypothetical enhancer-to-gene model reports a calibrated link estimate of 0.68 for Gene D in a relevant cell type, with a 95% confidence interval from 0.49 to 0.82. A variant overlaps the enhancer, but the assay provides no link estimate for a second cell type that may also be relevant.

The estimate supports a link in the measured cell type. Under the method’s repeated-sampling interpretation, the interval procedure would produce intervals containing the underlying link parameter in 95% of repeated applications under its assumptions. It does not mean there is a 95% probability that the true link probability lies in this particular interval.

The missing estimate for the second cell type is not evidence of a zero link there. It means the link is unmeasured or unavailable under the information provided. The mapping claim should therefore be limited to the measured context, with the second context left unresolved. The interval also indicates uncertainty: the evidence does not pin down the link probability precisely.

Context, Overlap, and Multiple Links

Regulatory activity is cell-type and condition dependent. An enhancer active in one tissue may be inactive in another, and contact data from one cell type may not transfer to a different one. Match the evidence to the biological context of the trait where possible, and name the context explicitly when reporting a link. A context mismatch weakens the relevance of the evidence; it does not necessarily make the measured result incorrect.

Enhancer intervals can overlap one another, and a variant can fall in more than one annotated element. Decide whether the mapping model treats these as distinct, mutually exclusive routes or as overlapping evidence for one regulatory region. Counting every overlapping annotation as independent support can inflate apparent evidence. Similarly, multiple assays may share data or biological signals. Agreement across correlated assays is not equivalent to independent replication.

An enhancer may have links to several genes. Do not force the probabilities to sum to one across genes unless the model explicitly defines one mutually exclusive target. If two genes have calibrated link probabilities of 0.6 and 0.5, that is not inherently contradictory: both may be regulated, and the events need not be mutually exclusive. Conversely, a missing link in a database is not necessarily a negative result; distinguish “not observed” from “tested and unsupported.”

Common Mistake: Treating a Link as a Causal Verdict

Wrong reasoning: “The variant overlaps an enhancer, and the enhancer contacts Gene A’s promoter, so Gene A is the causal gene.”
Correction: Overlap supports a possible variant-to-enhancer route, and contact supports a possible enhancer-to-gene connection. Neither alone demonstrates that the variant alters the enhancer, that the enhancer changes Gene A’s expression, or that this change mediates the trait association.
Wrong reasoning: “The link score is 0.8, so there is an 80% chance the gene is causal.”
Correction: First determine whether the score is calibrated and exactly what probability it represents. A link probability, even when calibrated, is not the probability that the gene causes the trait.
Wrong reasoning: “No link was reported for the other cell type, so the gene is not regulated there.”
Correction: An absent measurement is not a measured null. Report the evidence as missing or unavailable unless the assay actually tested and failed to support the link.

Reporting a Regulatory Mapping

A clear report makes the evidence chain and its limits visible. State the variant and enhancer coordinates, the genome build and interval convention, the source and type of link evidence, the gene and context, and whether the reported quantity is a probability or an uncalibrated score. Explain how multiple enhancer routes were handled and whether relevant measurements were missing. When posterior variant probabilities are used, distinguish those probabilities from the conditional enhancer-to-gene evidence.

Key takeaway: Enhancer-to-gene evidence can make a noncoding variant’s candidate-gene mapping more informative than proximity alone. Interpret the link in its assay and tissue context, preserve uncertainty in each step, and describe the result as support for a regulatory connection—not proof of a causal gene or mechanism.

Check Your Understanding

Answer each question by separating coordinate evidence, link evidence, and causal interpretation.

A variant overlaps an enhancer with a high uncalibrated score linking it to a distant gene. What can be concluded, and what cannot be inferred from the score alone?
In the mapping expression, what does $r_{ie}$ represent, and what does $q_{eg}$ represent?
Why may enhancer-to-gene probabilities for different genes fail to sum to one?
A link is observed in one cell type but is unavailable in another. How should the unavailable result be described?
What assumptions are needed to interpret a weighted sum over enhancer routes as a mapping probability?