Why Near-Identical Genotypes Need Careful Interpretation
A pair of samples with nearly identical genome-wide genotypes may represent the same person entered twice, two aliquots or technical replicates from one DNA sample, or two monozygotic (MZ) twins. These possibilities matter for quality control, but the genotype pattern by itself does not always distinguish them. The first task is therefore narrower: quantify how closely the samples agree, check whether the disagreements look technical, and describe the strength and limits of the evidence.
This tutorial concerns comparisons of called genotypes, especially at well-measured autosomal variants. Pairwise relatedness can help flag unusually close pairs, but near-identical pairs require attention to the actual calls as well. The same high agreement can arise from a repeated sample or from MZ twins, and interpretation also depends on sample records and laboratory information.
At a biallelic autosomal variant, genotypes can be represented as 0, 1, or 2 copies of the alternate allele. After the allele coding is harmonized, two calls match when their dosage values match. A call missing in either sample is not a match or a mismatch: it is excluded from the jointly called denominator and reported separately through call-rate or missingness summaries.
Conditions for a Meaningful Comparison
A concordance calculation is only meaningful if the two samples are compared at the same variants with the same allele definitions. In practice, comparisons should use consistently aligned variant identifiers and alleles, and should address strand or reference-allele differences before counting matches. A strand or coding mismatch can make the same biological genotype appear discordant. Multiallelic variants and sex chromosomes also require care; a clear first comparison uses well-called, diploid autosomal biallelic variants.
Call quality matters as well. Low-quality or poorly mapped variants can create apparent differences, while missing calls reduce the number of loci available for comparison. A high concordance based on a small or unrepresentative set of loci is weaker evidence than the same value based on many informative, well-measured loci. Conversely, many jointly called loci do not guarantee a sound comparison if sample contamination, allele harmonization errors, or systematic assay problems are present.
For a pairwise calculation, let \(C\) be the set of loci called in both samples, and let \(M\) be the number of those loci with matching genotypes. Then the concordance is \(M/|C|\), and the discordance rate is \((|C|-M)/|C|\). These rates are complements because every locus in \(C\) either matches or does not. Always give the denominator, not just the percentage.
Worked Example: Calculating Concordance With a Missing Call
Suppose two hypothetical samples have these calls at eight autosomal variants. The variants have already been harmonized, and calls shown are treated as usable.
| Variant | Sample A | Sample B |
|---|---|---|
| 1 | 0 | 0 |
| 2 | 1 | 1 |
| 3 | 2 | 2 |
| 4 | 0 | 1 |
| 5 | 1 | 1 |
| 6 | 0 | 0 |
| 7 | 2 | 2 |
| 8 | Missing | 1 |
At variant 8, the pair has no jointly called genotype, so that site is excluded. The remaining seven sites include six matches and one mismatch. Thus:
The pair agrees at 85.71% of the seven jointly called variants in this small illustration. That percentage alone is not enough to classify the pair: seven sites are far too few to characterize genome-wide similarity, and the mismatching call should be checked rather than automatically treated as biological. The missing call is also informative about data completeness, even though it does not enter the concordance denominator.
Why Agreement Is Strong Evidence of Close Genetic Similarity
At a single biallelic locus, unrelated people can match by chance. Under Hardy-Weinberg equilibrium, with alternate-allele frequency \(p\) and reference-allele frequency \(q=1-p\), genotype probabilities are \(p^2\), \(2pq\), and \(q^2\). For two independently sampled individuals from that population, the probability of matching is the sum of the probabilities that both have each possible genotype.
This result assumes random pairing from the same population, Hardy-Weinberg genotype frequencies, and no relationship between the two individuals. It describes a chance match at a locus, not the probability that a pair is unrelated after observing a match. It also does not account for population structure, genotyping error, or dependencies between loci.
If loci were mutually independent, the probability of matching at every one of \(L\) loci with the same allele frequency would be the single-locus probability raised to the \(L\)th power. This illustrates why many informative matches can provide strong evidence against two unrelated samples. Real genomic variants are not all independent: nearby variants may be in linkage disequilibrium, and allele frequencies vary across populations. Therefore, the simple product is an illustration, not a calibrated genome-wide probability for a real pair.
Worked Example: Chance Matching at One and Ten Loci
Consider a hypothetical biallelic variant with alternate-allele frequency \(p=0.20\), so \(q=0.80\). Under Hardy-Weinberg equilibrium, the genotype probabilities are \(0.20^2=0.04\), \(2(0.20)(0.80)=0.32\), and \(0.80^2=0.64\). For two independently sampled individuals:
So the chance that an unrelated pair matches at this one locus is 0.5136 under these assumptions. If ten such loci were mutually independent and had this same allele frequency, the chance of matching at all ten would be:
This is approximately 0.1277%. The calculation is exact only for the stated idealized assumptions. Ten nearby variants may not be independent, and the value does not directly apply to a genome-wide dataset with varied allele frequencies, population structure, and call error.
Investigating Discordant Calls
Even a true duplicate pair or a pair of MZ twins may not have perfect observed concordance. Genotyping and sequencing errors, sample swaps, contamination, low-quality calls, and allele harmonization mistakes can all produce discordances. MZ twins can also differ at a small number of sites through post-zygotic mutations or mosaicism, although such differences do not ordinarily overturn the broad conclusion of near-identical inherited genotypes.
A useful follow-up is to examine where discordances occur and whether they are concentrated. Are mismatches enriched among low-quality calls, low-depth sites, or variants with poor call rates? Does one sample have systematically more missingness or heterozygous calls? Do discrepancies cluster in a genomic region or appear across the genome? A concentration in problematic regions points toward technical review; isolated, high-quality mismatches may warrant checking sample identity and, where relevant, biological explanations.
The comparison should also report the number of jointly called loci and the missingness in each sample. For example, a high concordance calculated after excluding a substantial number of difficult sites may hide an important data-quality problem. Likewise, a small discordance rate can correspond to many mismatches in a very large dataset. The rate and the raw mismatch count answer different questions and are best considered together.
Worked Example: Interpreting a High Concordance and Its Exceptions
Two hypothetical samples are compared at 10,000 eligible autosomal variants. Each sample has 9,990 called genotypes, and 9,980 variants are called in both. Seven of those 9,980 jointly called variants are discordant. The other 9,973 match.
The samples show very high agreement at the jointly called variants, but there are still seven calls to investigate. The pair has 20 variants without calls in both samples, since \(10{,}000-9{,}980=20\). Given each sample has 9,990 calls, the missing sets must be disjoint in this example: one sample is missing ten variants that are called in the other, and vice versa. Review of call quality and allele alignment at the seven discordant sites would help determine whether these are likely technical discrepancies. The result supports close genetic similarity; it does not by itself establish whether the samples are duplicate records or MZ twins.
What the Genetic Evidence Can and Cannot Resolve
Several approaches describe related but distinct aspects of a pair. Genotype concordance directly asks how often the observed calls agree. Identity-by-state (IBS) describes sharing of the same observed alleles, without specifying how those alleles were inherited. Identity-by-descent (IBD) and kinship measures summarize sharing attributable to common ancestry under their model assumptions. A pairwise relatedness statistic can identify exceptionally close pairs, while direct concordance is particularly useful for checking whether calls themselves are nearly identical.
| Approach | What it summarizes | Important limitation here |
|---|---|---|
| Genotype concordance | Fraction of jointly called loci with equal genotypes | Depends on call quality, harmonization, and the selected loci |
| IBS sharing | Observed allele sharing at variants | Sharing alone does not establish how alleles were inherited |
| IBD or kinship estimates | Genetic sharing relative to a relationship model | Very close pairs may be hard to distinguish from one another using a single summary |
| Sample records and laboratory metadata | Collection, labeling, and processing history | Records can be incomplete or contain their own errors |
For duplicate-versus-twin interpretation, metadata are essential. A repeated sample identifier, documented replicate, or laboratory processing record may support a duplicate interpretation; recorded twin status may support an MZ-twin interpretation. Without such information, genotype data generally show that the pair is nearly identical, not which of these explanations is correct. Also, similarity does not establish that either sample is correctly labeled.
This comparison does not prescribe a universal concordance cutoff. The meaning of a value depends on the genotyping platform, variant set, call-quality filters, and the intended use. Threshold choice and decisions about which related sample to retain are separate decisions. The practical aim here is to produce a defensible pairwise comparison and identify cases that need review.
Common Mistakes and Diagnostics
- Counting missing calls as discordances. This lowers concordance for the wrong reason. Use only loci called in both samples, and report missingness separately.
- Assuming high concordance proves duplication. MZ twins can also have near-identical inherited genotypes. Use sample records and laboratory information to distinguish explanations where possible.
- Ignoring allele alignment. A strand or allele-coding mismatch can turn matching biological genotypes into apparent mismatches. Harmonize variants before comparing calls.
- Reporting a percentage without its denominator. A rate based on a few loci is not equivalent in evidential strength to the same rate based on many well-measured loci.
- Treating all mismatches as biological differences. Check call quality, missingness, and possible technical problems before interpreting discordance.
- Using an independence calculation as a genome-wide probability. Linkage, population structure, and varying allele frequencies violate the simple repeated-locus calculation.
A warning sign is disagreement between a very high concordance rate and poor sample-level quality indicators, such as extensive missingness or low-quality calls. Another is an unexpectedly large mismatch count concentrated in a small region. These patterns do not automatically invalidate the pairwise result, but they call for a closer look at variant-level and sample-level data before drawing conclusions.
Check Your Understanding
Answer each question using the definitions and assumptions in this tutorial. Short solutions follow.
- Two samples match at 4,950 of 5,000 jointly called variants. What are their concordance and discordance rates?
- Why should a locus called in only one sample not be counted as a genotype mismatch?
- Under the Hardy-Weinberg assumptions, what is the single-locus genotype-match probability when \(p=0.50\) and \(q=0.50\)?
- A pair has extremely high genotype concordance. Can that result alone determine whether the pair is a duplicate or MZ twins? Explain.
- What should be checked if many discordant calls cluster among low-quality variants?
Solutions. (1) Concordance is \(4{,}950/5{,}000=0.9900\), or 99.00%; discordance is \(50/5{,}000=0.0100\), or 1.000%. (2) A missing call provides no observed genotype in one member of the pair, so agreement cannot be assessed at that locus; report it through missingness instead. (3) Each genotype has probability 0.25, so the match probability is \(3(0.25^2)=0.1875\). (4) No. Duplicate samples and MZ twins can both show near-identical inherited genotypes; sample records and laboratory metadata are needed to distinguish them. (5) Review call quality and variant-level metrics, verify allele harmonization, and assess whether technical problems explain the cluster before treating the differences as biological.