eduKateSG · Genetics · The Gold Standard of Education
Test local-ancestry associations without confusing ancestry, alleles and causes
Choose the explanation you need, or follow all twelve chapters in order.
Full chapter index · Independent practice · Genetics subject guide
Local ancestry records which ancestral source a model assigns to a particular chromosomal segment. In an admixed population, that local label can help find trait associations that a single genome-wide ancestry proportion would miss. It can also create false confidence when ancestry inference, environmental structure or allele effects are not separated.
This guide starts with the estimand, builds a transparent fictional test, and then follows the result through confounding checks, replication and interpretation. The aim is not to turn ancestry into a biological essence. It is to show how a local-ancestry signal can become a testable, limited claim.
Continue from The Gold Standard of Genetics within The Gold Standard of Education: learning that transfers into life and the How X Works learning library.
Full chapter index
See the mechanism · Chapters 1–2
Build the worked model · Chapters 3–5
Test what changes · Chapters 6–8
Check evidence and limits · Chapters 9–10
Practise and continue · Chapters 11–12
CHAPTER 1 OF 12 · See the mechanism
1. Define the association question before choosing a model
Write the outcome, study population, genomic region, ancestry sources and unit of analysis. A local-ancestry association test asks whether the number or probability of ancestry copies at a locus is associated with an outcome after specified adjustment. That differs from testing a particular allele and from comparing broad population labels.
For a diploid autosomal locus, inferred local ancestry may be coded as zero, one or two copies from source A. Modern methods may retain posterior probabilities instead of forcing a hard call. State which representation is used, because uncertainty discarded at this step cannot be recovered by a later regression.
Define whether the target is discovery, fine mapping, effect heterogeneity or prediction. A locus can show an ancestry association because causal variants differ in frequency across sources, because local ancestry tags an environmental difference, because the ancestry model is wrong, or because chance survived a large search. Those possibilities require different follow-up.
Keep labels operational. “Source A” means the reference component used by the inference system. It is not a complete account of a person, a culture or a historical population.
Evidence check: for Define the association question before choosing a model, preserve the observation before the interpretation. Record outcome, population, ancestry sources and estimand. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 2 OF 12 · See the mechanism
2. Separate global ancestry, local ancestry and genotype
Global ancestry is an average across the analysed genome. Local ancestry varies from position to position because recombination has created a mosaic. Genotype records the alleles carried at a variant. These variables are related, but they are not interchangeable.
Global ancestry can track broad population structure and correlated environments. Local ancestry can capture regional structure that a global average does not. Genotype may carry the narrower allelic mechanism. Putting all three into one model without a causal question can create unstable estimates or block part of the signal being studied.
Draw a simple diagram before modelling. Include source ancestry, local ancestry, genotype, measured covariates, environment and outcome. Mark which arrows are hypotheses rather than facts. This makes clear whether local ancestry is the exposure of interest, a confounder to adjust, or a proxy used for discovery.
The same variable can play different roles in different analyses. A local-ancestry term may be central in admixture mapping and an adjustment variable in ancestry-aware association testing. Name the role every time.
Evidence check: for Separate global ancestry, local ancestry and genotype, preserve the observation before the interpretation. Record global ancestry, local ancestry and genotype separately. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 3 OF 12 · Build the worked model
3. Build the data pipeline before reading a p-value
Start with genotype quality, genome build, strand alignment, sample relatedness and missingness. Infer global ancestry with declared references, then phase and infer local ancestry with a method appropriate to the number of sources, admixture history and marker density. Record software versions and parameters.
Evaluate inference with simulated or masked truth where possible. Accuracy can differ by ancestry, chromosome, marker density and tract length. Short old tracts are harder to classify than long recent tracts. A single average accuracy number can hide the region that drives the claimed association.
Inspect posterior certainty and switch rates. Implausibly frequent ancestry changes may reflect phasing or map error. Long regions with low information may be overconfidently assigned when the reference panel is sparse.
Keep a reproducible table from raw sample identifiers to the final analytic set. Report exclusions by reason and, when ethically appropriate, compare whether exclusions differ across relevant groups. Technical cleaning can change the population to which the conclusion applies.
Evidence check: for Build the data pipeline before reading a p-value, preserve the observation before the interpretation. Record phasing, reference panels and posterior uncertainty. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 4 OF 12 · Build the worked model
4. Work a fictional local-ancestry association
Imagine 800 unrelated participants in a two-way admixed study. The outcome is a continuous biomarker. At one preregistered region, each person has zero, one or two inferred source-A ancestry copies. The model adjusts for age, sex, study site and global ancestry.
Suppose the fitted local-ancestry coefficient is 1.8 biomarker units per source-A copy, with a standard error of 0.6. The test statistic is 1.8 / 0.6 = 3.0. This is evidence against a zero coefficient in that model, but it is not yet evidence for a specific causal allele.
Now add the lead genotype. If the local-ancestry coefficient falls to 0.4 while the allele coefficient remains stable, the allele may explain much of the regional signal. If the local-ancestry coefficient remains 1.8, multiple variants, untyped variation, structural variation or non-genetic structure may remain.
The worked result is fictional and deliberately simple. Real analyses must account for genome-wide search, kinship, phenotype transformation, imputation and local-ancestry uncertainty.
| Quantity | Value | Role | Limit |
|---|---|---|---|
| Participants | 800 | Analytic sample | Fictional and unrelated |
| Local ancestry | 0, 1 or 2 A copies | Regional exposure | Inferred with error |
| Coefficient | 1.8 units per A copy | Conditional association | Not a causal allele effect |
| Standard error | 0.6 units | Sampling uncertainty | Excludes all model error |
Evidence check: for Work a fictional local-ancestry association, preserve the observation before the interpretation. Record every fictional coefficient and calculation. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 5 OF 12 · Build the worked model
5. Control broad structure without erasing the question
Adjustment for global ancestry is often useful because broad ancestry can correlate with many genetic and non-genetic variables. Yet global adjustment alone may not remove local population structure. Conversely, conditioning on local ancestry can remove the admixture-mapping signal the study intended to test.
Choose covariates from the estimand and causal structure, not from a ritual list. Compare models with global ancestry, principal components, study site, socioeconomic variables and local ancestry where each has a justified role. Report how coefficients and uncertainty change.
Relatedness needs explicit treatment. Mixed models, kinship matrices or family-aware designs can reduce inflated precision. If the kinship model was trained from the same variants being tested, document how proximal contamination was handled.
Avoid interpreting “adjusted for ancestry” as “free from confounding.” Ancestry variables are imperfect summaries. Residual geography, recruitment, treatment access and environmental exposure can remain.
Evidence check: for Control broad structure without erasing the question, preserve the observation before the interpretation. Record covariates from a causal question. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 6 OF 12 · Test what changes
6. Distinguish confounding from mediation and effect heterogeneity
A confounder influences both the tested exposure and outcome without lying on the causal path of interest. A mediator transmits part of an effect. An effect modifier changes the association across contexts. These concepts cannot be settled by software defaults.
Suppose source-A local ancestry raises the probability of carrying a functional allele, and the allele changes the biomarker. Genotype then mediates or explains the ancestry association in a discovery model. If a site-specific exposure is also correlated with local ancestry because recruitment followed geography, that exposure may confound the association.
An ancestry-specific allele effect is a different hypothesis: the same allele has different estimated effects depending on the ancestry background on which it occurs. Testing it requires enough copies in each background and careful phasing. Sparse cells produce unstable interactions.
Use stratified counts before fitting interactions. A mathematically estimable coefficient can still be scientifically unsupported when one ancestry-background cell contains only a few carriers.
Evidence check: for Distinguish confounding from mediation and effect heterogeneity, preserve the observation before the interpretation. Record confounding, mediation and effect heterogeneity. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 7 OF 12 · Test what changes
7. Correct the search and keep the region visible
Admixture mapping usually has fewer effectively independent tests than a dense variant scan because ancestry tracts are long, but it still requires a study-specific threshold. The effective number depends on admixture history, sample size, marker set and local-ancestry correlation.
Use permutation, simulation or a defensible analytic threshold. Preserve the regional association curve, not only the most significant position. A broad plateau may be expected when neighbouring markers share the same ancestry tract.
Fine mapping should condition on the discovery process. Test variants, haplotypes and structural features within the region; compare models; and quantify what remains unexplained. Do not call the closest named gene causal merely because it is convenient.
Separate discovery and interpretation graphics. One shows evidence across the region. Another can show functional annotations, ancestry-specific allele frequencies and credible sets. Mixing them encourages readers to see annotation as proof.
Evidence check: for Correct the search and keep the region visible, preserve the observation before the interpretation. Record search threshold and regional fine mapping. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 8 OF 12 · Test what changes
8. Carry ancestry-inference error into uncertainty
Hard local-ancestry calls treat uncertain positions as known. That can reduce power, bias effects or create differential error when accuracy relates to outcome-linked groups. Posterior dosage or multiple imputation can carry more uncertainty forward.
Repeat the analysis using alternative credible reference panels, ancestry methods and call thresholds. Mask low-confidence regions and inspect whether the conclusion depends on one panel or algorithm. Simulate data under the fitted admixture history and pass it through the full pipeline.
Negative controls are valuable. Test outcomes that should not plausibly respond to the region, or genomic regions unlikely to affect the outcome but with similar inference quality. Systematic associations across controls point to structure or technical artefact.
Report calibration as well as discovery. Quantile plots, null simulations and genomic inflation summaries help, but no single diagnostic proves that confounding is absent.
Evidence check: for Carry ancestry-inference error into uncertainty, preserve the observation before the interpretation. Record ancestry-inference error and negative controls. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 9 OF 12 · Check evidence and limits
9. Design replication that can actually test the claim
An exact replication preserves the outcome definition, ancestry contrast, region, effect direction and model as far as possible in an independent sample. A conceptual replication may use another population or measurement while testing the same mechanism. Label the distinction.
Replicating local ancestry is difficult when the second population has different admixture timing, source panels or tract distributions. Failure may reflect poor power or a different linkage pattern; success may still reproduce a shared bias. Predefine what counts as support.
Use harmonised genome builds and allele orientation. Re-run local ancestry rather than copying labels across incompatible references. Report sample overlap and relatedness between discovery and replication cohorts.
Combine studies only after inspecting heterogeneity. A meta-analysis average can hide opposite ancestry-specific effects. Show cohort estimates, uncertainty and the assumptions behind any fixed- or random-effects synthesis.
| Check | Preserve | Allow to differ | Failure exposed |
|---|---|---|---|
| Exact replication | Outcome, region, direction, model | Independent sample | Chance or study-specific bias |
| Panel sensitivity | Question and sample | Reference surrogates | Ancestry-call dependence |
| Regional fine mapping | Discovery region | Variant tags and LD | Closest-gene storytelling |
| Heterogeneity | Cohort estimates | Effect size | Transport limits |
Evidence check: for Design replication that can actually test the claim, preserve the observation before the interpretation. Record exact and conceptual replication. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
CHAPTER 10 OF 12 · Check evidence and limits
10. Limit the biological and social interpretation
A local-ancestry association identifies a statistical relationship under a model. It does not show that ancestry itself is a molecular mechanism. The mechanism may involve one or more variants, regulation, structural variation, environment or their interaction.
Do not turn a regional result into a hierarchy of populations. Within-group variation is substantial, source labels depend on available references, and social categories do not map cleanly onto inferred genomic ancestry.
Consider consent, governance, benefit and group-level harms. A technically public dataset may still support interpretations that communities did not anticipate. Engage relevant communities when the result could stigmatise or alter access to resources.
Clinical use needs separate validation. Association does not establish transportability, calibration, decision benefit or fairness. Keep research discovery, causal explanation and clinical action as different claims.
Evidence check: for Limit the biological and social interpretation, preserve the observation before the interpretation. Record biological, social and governance limits. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
Case A is the fictional region: the local-ancestry coefficient is 1.8 units and falls after the lead genotype enters. The cautious interpretation is that a regional allelic signal may explain much of the ancestry association.
Case B keeps the same biology but the replication panel has poor source representation. Local-ancestry switches increase and the coefficient weakens. This does not by itself refute the mechanism; it shows that the measurement system changed.
Case C reproduces the coefficient until study site is added, after which it approaches zero. Recruitment geography and environment were carrying the apparent ancestry signal. The correct response is not to hide the adjusted model but to revise the claim.
Case D finds the same local-ancestry association in a new cohort while different lead variants tag it. That pattern may support a regional mechanism while showing that fine-mapping depends on ancestry-specific linkage. Transfer means preserving the question while allowing the evidence path to change.
Evidence check: for Run a changed-context transfer test, preserve the observation before the interpretation. Record the changed reference-panel context. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.
Archive consent and governance, cohort definitions, outcome code, genome build, genotype filters, reference panels, phasing, local-ancestry parameters, posterior uncertainty, global ancestry, covariates, kinship treatment, search threshold, regional plots, fine-mapping, negative controls, replication protocol and exact reporting language.
Common questions: Is local ancestry the same as ethnicity? No. Does global ancestry adjustment remove every ancestry-related confounder? No. Does a significant ancestry region identify a causal gene? No. Should uncertain ancestry calls be treated as exact? Usually not. Does non-replication always invalidate discovery? No, but it demands a power, measurement and heterogeneity audit.
Continue with ancestry deconvolution and multiple-pulse dating, ancestry-specific recombination and X–autosome contrasts, and haplotype sharing and ancestry tracts. Read the open overview of admixture mapping, the local-ancestry adjustment study, and the comparison of global and local adjustment.
Evidence check: for Keep the evidence route and continue, preserve the observation before the interpretation. Record the narrowest supported regional association. Then state the inference, one credible alternative, and the additional observation that would distinguish them. This compact audit keeps a fluent explanation tied to evidence that another reader can inspect.
Changed-context transfer: repeat this chapter’s decision after changing one consequential condition while holding the others fixed. Calculate or predict what should change, name the threshold or mechanism, and compare with the original case. If the conclusion survives, state the range tested; if it reverses, explain why. Transfer is demonstrated by the changed case, not asserted from the first example.
Practical decision record: write one sentence for what is known, one for what remains uncertain, and one for the next safe action. Include the evidence owner, the latest useful decision time, and the condition that would stop or reverse the action. This turns a technical result into a reviewable choice without pretending uncertainty has disappeared.