Population Stratification and Ancestry

  • Population stratification and ancestry are important concepts in population genetics, statistical genetics, genomics, and bioinformatics. They describe patterns of genetic similarity and difference among individuals and populations, shaped by evolutionary history, migration, genetic drift, natural selection, mutation, recombination, and demographic events. Understanding these patterns is essential for interpreting genetic variation, conducting genome-wide association studies (GWAS), studying human evolution, investigating disease susceptibility, and developing reliable genomic prediction methods.
  • Population stratification occurs when a study population contains subgroups with different ancestral backgrounds or genetic histories. These subgroups may differ in allele frequencies because of geographic separation, historical migration, population bottlenecks, founder effects, genetic drift, admixture, or other evolutionary processes. As a result, genetic variants may be distributed unevenly across the study population. Population structure can be subtle, even among individuals who appear geographically or culturally similar, and may be detected through genome-wide patterns of genetic variation.
  • Ancestry refers to the historical origins and genetic contributions of an individual’s ancestors. Genetic ancestry is inferred from inherited DNA variation and reflects the populations from which an individual’s ancestors may have originated. It is not necessarily identical to geographical, cultural, linguistic, national, or social identity. Human populations have continuously migrated, mixed, and exchanged genetic material throughout history. Consequently, ancestry is often complex and cannot always be represented accurately by assigning an individual to a single population or geographic category.
  • Population stratification can occur at different geographical and evolutionary scales. Broad-scale structure may reflect differences among continental or large regional populations, while fine-scale structure may occur among neighbouring countries, regions, communities, or geographically separated groups within the same country. Local population structure may also arise from historical isolation, endogamy, founder events, or limited gene flow. In animal and plant populations, structure can reflect breed formation, domestication, artificial selection, geographic separation, and the development of distinct breeding populations.
  • Genetic differences among populations are commonly measured through allele frequencies, genotype frequencies, haplotype distributions, and patterns of linkage disequilibrium (LD). An allele frequency represents the proportion of copies of a particular allele at a genetic locus within a population. Differences in allele frequencies across subgroups provide evidence of population structure. However, individual genetic variants rarely provide a complete picture of ancestry. Genome-wide information from thousands or millions of markers, particularly single-nucleotide polymorphisms (SNPs), is generally more informative for identifying genetic relationships and population structure.
  • One important consequence of population stratification is confounding in genetic association studies. A confounder is a factor associated with both the genetic predictor and the phenotype being studied. Suppose a genetic variant is more frequent in one ancestral subgroup, and that subgroup also has a higher average value of a particular trait because of environmental, social, demographic, or other genetic factors. An association analysis that ignores population structure may incorrectly attribute the difference in the trait to the genetic variant. This can produce a spurious association, meaning that a statistical relationship is detected even though the variant does not have the claimed causal effect.
  • Population stratification is therefore a major concern in genome-wide association studies. For example, if cases and controls have different ancestral compositions, allele-frequency differences may arise because of ancestry rather than disease status. A variant may appear associated with a disease simply because it is more common in the ancestral group that contributes a larger proportion of cases. This problem can affect both common and rare variant analyses, and it may lead to false-positive findings or distort estimates of genetic effects. Careful study design, ancestry assessment, and statistical correction are necessary to reduce this risk.
  • Principal component analysis (PCA) is one of the most widely used methods for examining population structure. PCA reduces high-dimensional genotype data into a smaller number of variables called principal components (PCs). Each principal component captures a direction of genetic variation, with the first component explaining the largest possible proportion of variation and subsequent components explaining additional variation under the PCA model. Individuals can be plotted using their principal component scores. Individuals with similar genome-wide genetic patterns often appear near one another, while groups with different genetic backgrounds may form distinguishable clusters or gradients.
  • PCA does not directly identify an individual’s complete ancestry or establish a definitive population label. Its results depend on the individuals included in the analysis, the genetic markers selected, quality-control procedures, linkage disequilibrium, missing data, and other technical factors. The principal components describe variation within the analysed dataset, so they should be interpreted in relation to the study design and reference populations. A continuous distribution in a PCA plot may reflect gradual geographic variation, admixture, or other patterns rather than clearly separated populations.
  • Before conducting PCA, genotype data generally require quality control. Common steps include removing poorly genotyped individuals and variants, evaluating sample contamination and identity, checking relatedness, excluding problematic genomic regions when appropriate, and addressing missing genotype data. Variants with very low minor allele frequency may be removed for some analyses because they contribute less stable information to broad population-structure estimation. Linkage disequilibrium pruning is also commonly performed so that large groups of correlated SNPs do not dominate the principal components. The appropriate quality-control thresholds depend on the dataset, organism, sequencing or genotyping technology, and research question.
  • The PLINK and PLINK 2 software packages are widely used for genotype quality control, LD pruning, PCA, and population-genetic analyses. A simplified PLINK 2 workflow for calculating principal components from a dataset in PGEN format may look like this:

# Calculate allele frequencies and perform initial quality checks

plink2 –pfile cohort \

       –freq \

       –missing \

       –out cohort_qc

Linkage disequilibrium pruning can be performed to select a set of approximately independent SNPs:

# Identify approximately independent variants

plink2 –pfile cohort \

       –indep-pairwise 200 50 0.2 \

       –out cohort_pruned

The resulting variant list can be used to calculate principal components:

# Calculate the first 10 principal components

plink2 –pfile cohort \

       –extract cohort_pruned.prune.in \

       –pca 10 \

       –out cohort_pca

  • These commands illustrate a common workflow rather than a universal analysis protocol. The window size, step size, LD threshold, number of principal components, and other settings should be selected according to the dataset and biological objectives. Researchers should also consider whether close relatives, admixed individuals, or technical batches need separate evaluation before interpreting the PCA results.
  • Ancestry can also be investigated using model-based clustering methods. STRUCTURE and ADMIXTURE are widely known examples. These methods estimate ancestry proportions under statistical models that represent individuals as having contributions from a specified number of ancestral components. The parameter (K) represents the number of assumed ancestral components in the model. An individual may be represented as having contributions from several components rather than belonging entirely to one group. The estimated proportions are model-dependent and should not be interpreted as exact percentages of ancestry from historically defined populations. Different choices of reference samples, marker sets, and (K) can produce different results.
  • ADMIXTURE is commonly used for large SNP datasets because it provides relatively efficient estimation of ancestry proportions under its modelling assumptions. A simplified command-line example is:

# Estimate ancestry proportions for K = 3

admixture –cv cohort.bed 3

  • The appropriate value of (K) is not necessarily the number of historically distinct populations. Researchers often compare model fit across several values of (K), examine cross-validation error, and evaluate whether the results are stable and biologically meaningful. STRUCTURE and ADMIXTURE simplify complex demographic histories into a limited number of components, so their output should be treated as a statistical representation rather than a literal reconstruction of ancient populations.
  • Genetic ancestry can also be inferred by comparing an individual’s genotype data with reference populations. Reference panels consist of individuals whose genetic data have been collected and characterized for a particular research purpose. Examples of widely used human population resources include the 1000 Genomes Project, the Human Genome Diversity Project (HGDP), the Simons Genome Diversity Project, and population datasets incorporated into resources such as gnomAD. These resources differ in sampling design, population representation, sequencing technology, and available metadata. Reference panels should be selected carefully because incomplete or unrepresentative sampling can lead to misleading ancestry estimates.
  • Ancestry estimation is affected by the quality and diversity of reference populations. If a person’s ancestral background is not adequately represented in the reference panel, the method may assign their genetic variation to the closest available groups, even when those groups are not accurate representations of their ancestry. Individuals with ancestry from several populations may also receive different estimates depending on the reference panel and algorithm. Therefore, ancestry estimates should be reported with information about the reference dataset, markers, software, model assumptions, and limitations.
  • Local ancestry analysis is a more detailed approach used particularly in admixed populations. Instead of estimating ancestry only across the entire genome, local ancestry methods infer the likely ancestral origin of different chromosomal segments. Recombination over generations produces chromosomes containing segments inherited from different ancestral populations. By comparing these segments with reference panels, researchers can estimate ancestry along each chromosome. Local ancestry can help investigate population history, admixture timing, ancestry-specific genetic effects, and disease-associated loci. Its accuracy depends on the quality of reference panels, the genetic markers available, recombination patterns, and the demographic complexity of the populations being studied.
  • Tools such as RFMix and LAMP-LD have been used for local ancestry inference, while other methods address related ancestry and haplotype-estimation problems. Local ancestry is not the same as identifying the precise historical individual or community from which a DNA segment originated. It estimates similarity to the reference populations and categories used by the model. If historical populations are not represented in the reference panel, their genetic contributions may be approximated by the closest available groups.
  • Population structure is also closely related to genetic drift, migration, admixture, and gene flow. Genetic drift refers to random changes in allele frequencies across generations, particularly in small populations. Population bottlenecks reduce population size and can alter genetic diversity through random loss of alleles. Founder effects occur when a new population is established by a small number of individuals, potentially producing allele frequencies that differ from those of the source population. Migration and gene flow introduce genetic variation between populations, while admixture occurs when previously differentiated populations contribute genetically to a new or existing population.
  • Linkage disequilibrium is another important feature of population structure. LD describes the non-random association of alleles at different loci. Patterns of LD vary across populations because of demographic history, recombination, genetic drift, selection, and admixture. Differences in LD can influence the performance of genome-wide association studies, fine-mapping, and genotype imputation. A genetic marker that is strongly correlated with a causal variant in one population may be weakly correlated with it in another. Consequently, association signals identified in one population may not transfer directly to another population.
  • Population stratification can be addressed through several statistical methods. One common approach is to include principal components as covariates in regression models. For a quantitative phenotype, a simplified association model can be written as:

[
Y_i = \beta_0 + \beta_1 G_i + \sum_{j=1}^{k}\gamma_j PC_{ij} + \boldsymbol{\delta}^{T}\mathbf{C}_i + \varepsilon_i
]

  • Here, (Y_i) represents the phenotype for individual (i), (G_i) represents the genotype at the genetic variant being tested, (PC_{ij}) represents the individual’s score on principal component (j), and (\mathbf{C}_i) represents other covariates such as age, sex, or technical variables. The coefficient (\beta_1) estimates the genetic association after accounting for the included covariates. This approach is widely used in GWAS, although the number of principal components and the choice of covariates require careful consideration.
  • Mixed-model association methods provide another way to account for genetic similarity and population structure. These methods incorporate a genetic relationship matrix (GRM), which estimates genetic relatedness among individuals, as a random effect in the statistical model. Linear mixed models are commonly used for quantitative traits, while generalized mixed models and related approaches can be used for binary or other phenotypes. Methods such as EMMAX, GEMMA, BOLT-LMM, SAIGE, and REGENIE are used in different GWAS settings. The appropriate method depends on phenotype type, cohort size, relatedness, case-control imbalance, computational resources, and study design.
  • Genomic relationship matrices are calculated from genome-wide markers and describe genetic similarity among individuals. They can help account for both close relationships and broader population structure. However, the relationship matrix itself depends on marker selection, allele-frequency estimates, genotype quality, and the population used to calculate it. A poorly specified model may fail to control confounding or may reduce statistical power. Researchers should therefore examine model diagnostics and evaluate whether the selected correction method is appropriate for the study population.
  • Population stratification is also relevant to genomic prediction and precision medicine. Polygenic risk scores (PRS) are calculated by combining the effects of many genetic variants, often using effect estimates from GWAS. If the discovery and target populations differ in allele frequencies, LD patterns, environmental exposures, or genetic architecture, prediction accuracy may decline when the score is transferred between them. This limitation has raised important concerns about the unequal performance of genetic prediction tools across populations. More diverse discovery cohorts, appropriate validation, and population-aware statistical methods can help improve generalizability, although they do not eliminate all sources of bias.
  • In pharmacogenomics, ancestry-related differences in allele frequencies may influence the distribution of variants affecting drug metabolism, transport, or response. However, ancestry should not be used as a simple substitute for directly testing clinically relevant genetic variants. Individual genotypes, clinical history, drug interactions, and other patient-specific factors are more directly relevant to treatment decisions. Similarly, differences in disease-associated allele frequencies across populations do not automatically imply that ancestry itself causes disease risk. Environmental, socioeconomic, healthcare, and other biological factors may contribute to observed differences.
  • Population stratification and ancestry analysis are also widely used in animal and plant genomics. In livestock, population structure can reflect breed formation, crossbreeding, selection, and the use of popular breeding males. PCA, genomic relationship matrices, and ancestry estimation can help identify breed composition, detect admixture, evaluate genetic diversity, and manage inbreeding. In plant breeding, these methods are used to characterize germplasm collections, identify population groups, control structure in association mapping, and support the selection of genetically diverse breeding material. In conservation genetics, they help identify population differentiation, gene flow, and management units, although decisions should incorporate ecological and demographic evidence as well.
  • Several software tools are available for population-structure and ancestry analysis. PLINK and PLINK 2 support genotype processing, LD pruning, PCA, and relatedness analysis. EIGENSOFT provides PCA and methods for examining population structure. ADMIXTURE and STRUCTURE estimate ancestry proportions under model-based clustering approaches. RFMix is used for local ancestry inference in admixed populations. KING and related tools estimate genetic relationships, while tools such as VCFtools and BCFtools support variant-data filtering and population-genetic summaries. R packages, including SNPRelate and related Bioconductor resources, provide additional methods for analysing large-scale genotype data.
  • Interpreting population structure requires attention to technical artifacts as well as biological history. Differences in sequencing platforms, genotyping arrays, variant-calling pipelines, batch effects, missingness, and sample quality can create artificial clusters in genetic analyses. If one population is genotyped using one technology and another population using a different technology, PCA may separate samples by technical platform rather than ancestry. Consistent quality control, careful variant harmonization, and evaluation of technical covariates are therefore essential.
  • The definition of a population also depends on the research question. A population may be defined geographically, demographically, historically, ecologically, or genetically. These definitions are not interchangeable. Genetic clusters are statistical summaries of variation and do not necessarily correspond to discrete biological populations. Human genetic variation often changes gradually across geography, and boundaries between groups may be uncertain or historically recent. Researchers should avoid treating ancestry categories as fixed, universal, or biologically absolute classifications.
  • Ethical and responsible use of ancestry information is particularly important in human genomics. Genetic ancestry estimates should not be used to make unsupported claims about identity, behaviour, intelligence, or social characteristics. Genetic differences among populations are generally complex and influenced by many evolutionary and demographic processes. Researchers should communicate uncertainty, protect genetic privacy, avoid stigmatizing interpretations, and consider the implications of using ancestry information in clinical, forensic, and population-based research. Consent, data governance, appropriate access controls, and transparent reporting are important components of responsible genomic research.
  • For reproducible population-structure analysis, researchers should document the reference genome assembly, genotyping or sequencing platform, sample inclusion criteria, variant filters, allele-frequency thresholds, LD-pruning settings, software versions, reference panels, and statistical models. PCA plots, ancestry estimates, and association results should be interpreted in the context of these methodological choices. Where possible, analyses should be repeated using alternative reference panels or modelling approaches to assess the stability of conclusions. Researchers should also distinguish exploratory ancestry analysis from clinical or causal inference.
  • Population stratification and ancestry are fundamental to understanding genetic diversity and interpreting genomic data. Population structure reflects the combined effects of evolutionary history, demographic events, migration, selection, and genetic drift, while ancestry analysis estimates genetic relationships and contributions from reference populations. These concepts are essential for controlling confounding in GWAS, improving genomic prediction, studying human and animal evolution, and understanding variation in disease-associated and pharmacogenomic markers. Reliable interpretation requires appropriate reference data, robust statistical methods, careful quality control, and awareness of the scientific and ethical limitations of ancestry inference.

Author:

Leave a Reply

Your email address will not be published. Required fields are marked *