Why Positional Rules Need to Be Compared
A genomic variant can be close to one gene, lie within another gene’s body, and fall inside a wider window around several genes. These are not contradictory observations: they reflect different operational rules for translating genomic coordinates into candidate-gene assignments. A positional assignment is therefore a rule-based annotation, not evidence by itself that a gene mediates an association.
This tutorial compares three common rules: assigning a variant to the gene with the nearest transcription start site (TSS), assigning it to every gene whose body contains it, and assigning it to genes whose TSSs fall within specified upstream or downstream windows. The central question is not which rule is universally correct, but how much the candidate list and its ranking depend on the rule chosen.
Coordinate Conventions and Mapping Rules
Assume variants and genes have been placed on the same reference genome build and chromosome. A gene body is the genomic interval from its annotated start to its annotated end, inclusive. The TSS is the coordinate at which transcription begins: it is at the lower-coordinate end of a gene on the positive strand and at the higher-coordinate end on the negative strand. The strand matters when defining upstream and downstream, but not when calculating absolute genomic distance.
For variant \(i\) at coordinate \(x_i\) and gene \(g\) with TSS \(t_g\), the absolute TSS distance is \(d_{ig}=|x_i-t_g|\). The nearest-TSS rule assigns each variant to a gene minimizing this distance. A tie requires a prespecified convention, such as assigning all tied genes or using a deterministic tie-break; it should not be resolved silently.
The gene-body rule assigns a variant to gene \(g\) if its coordinate lies within that gene’s annotated body. A variant in overlapping gene bodies can map to more than one gene, while a variant outside every body maps to none. This rule does not automatically assign nearby intergenic variants.
A TSS-window rule assigns a variant to a gene if it lies within a specified distance of that gene’s TSS. A symmetric window of width \(W\) includes coordinates satisfying \(d_{ig}\leq W\). A directional rule instead specifies an upstream distance \(U\) and downstream distance \(D\), measured in the direction of transcription. On the positive strand, upstream is toward lower coordinates; on the negative strand, upstream is toward higher coordinates. Unless a rule explicitly excludes gene bodies, a TSS window can include both intragenic and intergenic variants.
| Rule | Assignment criterion | Typical mapping outcome |
|---|---|---|
| Nearest TSS | Assign to the gene with the smallest absolute TSS distance | One gene per variant, except for ties |
| Gene body | Assign to every gene body containing the variant | Zero, one, or multiple genes |
| TSS window | Assign to each TSS meeting a prespecified distance rule | Zero, one, or multiple genes |
Turning Variant Assignments into Candidate Scores
When fine-mapping supplies posterior probabilities for variants, those probabilities can summarize the positional support for each gene under a chosen rule. Let \(q_i\) be the posterior probability that variant \(i\) is causal, with \(\sum_i q_i=1\) when the analysis represents one causal variant among the listed variants. Let \(A_{ig}=1\) if the chosen positional rule maps variant \(i\) to gene \(g\), and \(A_{ig}=0\) otherwise. Define the positional score
If every variant maps to exactly one gene, \(S_g\) is the posterior probability assigned to that gene by this positional rule, and the gene scores sum to one. If variants can map to multiple genes, the same posterior mass can contribute to several scores, so the gene scores need not sum to one. If some variants are unmapped, the scores can sum to less than one. In either case, a score is conditional on the rule; it is not a probability that the gene is biologically causal.
The formula follows by defining the indicator \(A_{ig}\) for whether variant \(i\) maps to \(g\). If the causal variant is \(i\), its contribution to gene \(g\)’s positional assignment is \(A_{ig}\). Averaging that indicator over the variant posterior distribution gives
This is a posterior-weighted positional summary. Its interpretation depends on the mapping definition and on the posterior probabilities supplied to it; changing the window changes the indicators \(A_{ig}\), even when the variant probabilities stay fixed.
Worked Example: Applying the Three Rules
Consider four hypothetical variants and three hypothetical genes on the same chromosome. The gene coordinates are: A, positive strand, body 1000–1200 and TSS 1000; B, negative strand, body 1600–1900 and TSS 1900; and C, positive strand, body 2450–2700 and TSS 2450. The variants have the following coordinates and posterior probabilities: \(v_1=(1050,0.35)\), \(v_2=(1350,0.30)\), \(v_3=(1850,0.20)\), and \(v_4=(2150,0.15)\). The probabilities sum to \(1.00\).
Nearest TSS. For \(v_1\), the distances to A, B, and C are 50, 850, and 1400 bp, so it maps to A. For \(v_2\), the distances are 350, 550, and 1100 bp, so it maps to A. For \(v_3\), they are 850, 50, and 600 bp, so it maps to B. For \(v_4\), they are 1150, 250, and 300 bp, so it maps to B. Thus \(S_A=0.35+0.30=0.65\), \(S_B=0.20+0.15=0.35\), and \(S_C=0\). The nearest-TSS ranking is A, then B, then C.
Gene body. Variant \(v_1\) lies in A’s body, and \(v_3\) lies in B’s body. Neither \(v_2\) nor \(v_4\) lies in any of the three bodies. Therefore \(S_A=0.35\), \(S_B=0.20\), and \(S_C=0\); posterior mass \(0.30+0.15=0.45\) remains unmapped. A still ranks above B, but the total assigned mass is lower than under nearest TSS.
Symmetric 500-bp TSS window. Here a variant maps to every gene with TSS distance at most 500 bp. \(v_1\) and \(v_2\) map to A; \(v_3\) maps to B; and \(v_4\), 250 bp from B and 300 bp from C, maps to both B and C. The scores are \(S_A=0.35+0.30=0.65\), \(S_B=0.20+0.15=0.35\), and \(S_C=0.15\). A ranks first, followed by B and C. The scores sum to 1.15 because \(v_4\)’s posterior mass is counted for both B and C.
Symmetric 1000-bp TSS window. Under this wider rule, \(v_1\) maps to A and B; \(v_2\) maps to A and B; \(v_3\) maps to A, B, and C; and \(v_4\) maps to B and C. Consequently, \(S_A=0.35+0.30+0.20=0.85\), \(S_B=0.35+0.30+0.20+0.15=1.00\), and \(S_C=0.20+0.15=0.35\). The ranking changes: B is now first, ahead of A, then C. The scores sum to 2.20 because several variants map to multiple genes. This is a ranking under the 1000-bp rule, not a probability distribution over genes.
Directional Windows and Strand
A symmetric window treats the two genomic directions equally. A directional window can instead use different upstream and downstream distances, which makes strand orientation essential. The following example uses upstream \(U=500\) bp and downstream \(D=400\) bp. Coordinates at a TSS are included; a variant at a boundary is included if it meets the stated distance.
Worked Example: Applying an Upstream–Downstream Rule
Suppose hypothetical gene P is on the positive strand with TSS 10,000, and gene Q is on the negative strand with TSS 10,600. Three variants have coordinates and posterior probabilities \(x=(10{,}100,0.45)\), \(y=(10{,}800,0.35)\), and \(z=(10{,}400,0.20)\). Apply the directional windows \(U=500\) bp and \(D=400\) bp to each gene.
For positive-strand P, upstream coordinates are below 10,000 and downstream coordinates are above it. Variant \(x\) is 100 bp downstream of P, so it maps to P. Variants \(y\) and \(z\) are 800 and 400 bp downstream, respectively; \(y\) is outside the 400-bp downstream window, and \(z\) lies on its included boundary, so \(z\) also maps to P.
For negative-strand Q, upstream coordinates are above 10,600 and downstream coordinates are below it. Variant \(y\) is 200 bp upstream of Q and maps to Q. Variant \(x\) is 500 bp downstream of Q, outside the 400-bp downstream window, and \(z\) is 200 bp downstream, so \(z\) maps to Q. The scores are \(S_P=0.45+0.20=0.65\) and \(S_Q=0.35+0.20=0.55\). Variant \(z\) maps to both genes, so the scores sum to 1.20. Reversing Q’s strand when determining upstream and downstream would change its assignments and would be an annotation error.
Quantifying Sensitivity Across Rules
A useful sensitivity analysis holds the variant set, posterior probabilities, genome build, and gene annotations fixed, then varies only the positional rule. For each rule, record the mapping outcomes, gene scores, and ranking. Report whether the top-ranked gene changes, how many candidates enter the mapped set, and whether the scores include duplicated or unmapped mass. A ranking change is substantive evidence that the positional prioritization depends on the chosen definition; it is not a reason to select whichever rule gives the preferred answer.
One simple numerical summary is the change in rank of a gene between two rules. If A ranks first under a 500-bp window but second under a 1000-bp window, its rank shift is one place. Such a summary should accompany, not replace, the score table: it does not reveal why the ranking changed. In the worked example, B overtakes A under the 1000-bp window because the wider window assigns additional posterior mass to B, including variants already assigned to other genes.
The window should be specified before interpreting results, including whether it is measured from the TSS or gene boundary, whether upstream and downstream widths differ, whether the gene body is included, and how boundaries and ties are handled. Distances are meaningful only on a consistent coordinate system and reference build. A result that changes after a small window adjustment may indicate that several genes are geographically close to the credible variants, rather than that one positional rule has discovered the true target.
Common Pitfall and Interpretation
- Wrong reasoning: “The gene with the largest score under the widest window must be the target, because the widest window captures the most evidence.” A wider window mechanically admits more genes and can count the same variant posterior mass for several genes.
- Correction: Treat the score as conditional on the mapping rule. Show scores and mapping multiplicity across prespecified alternatives, and describe positional support as one component of candidate prioritization—not as causal proof.
- Wrong reasoning: “A variant outside a gene body cannot be assigned to that gene.” This is true only under the gene-body rule; nearest-TSS and TSS-window rules can assign intergenic variants.
- Correction: State the positional rule alongside every assignment. Positional mappings are not interchangeable with regulatory links or other functional evidence.
Positional mapping is useful because it makes a transparent, reproducible first pass from variant coordinates to genes. Its value is greatest when the rule is explicit and sensitivity is visible. A stable top candidate across reasonable rules has more robust positional support than one whose rank depends on a single arbitrary cutoff, but neither pattern alone establishes a biological mechanism.
Check Your Understanding
Answer these questions using the definitions and rules above.
- A variant is 200 bp from one TSS and lies inside a different gene body. Which gene or genes are assigned under nearest-TSS and gene-body rules?
- Under a symmetric 300-bp window, a variant is 250 bp from each of two TSSs. How does it contribute to the two gene scores?
- Why can scores from a multi-gene TSS-window rule sum to more than one even when variant posterior probabilities sum to one?
- A negative-strand gene has TSS coordinate 20,000. In which genomic direction is its upstream window?
- In the first worked example, what changed between the 500-bp and 1000-bp windows that caused B to overtake A?