Resolving Duplicate And Close-Relative Candidates

Why Candidate Pairs Need More Than a Relatedness Estimate

A pair of samples with a very high relatedness estimate may represent the same biological individual measured twice, or two genuinely close relatives. Those possibilities have different implications for data quality and downstream analysis. A relatedness estimate identifies a pair worth investigating; it does not, by itself, establish whether the samples are duplicates.

Resolving the pair requires evidence with different roles. Relatedness estimates summarize allele sharing. Genotype concordance asks whether the two samples have the same genotype at the same markers. Sample identifiers and provenance records help establish whether the labels refer to the same person, separate relatives, or a possible labeling error. These sources should be checked together, with uncertainty retained when they do not give a clear answer.

Definition: A duplicate-sample candidate is a pair of records suspected to represent the same biological sample or individual. A close-relative candidate is a pair whose high allele sharing may instead arise because the records represent different related individuals. Genotype concordance is the proportion of jointly observed, eligible markers at which the two samples have the same called genotype.

What the Genetic Evidence Can Show

Use the scale and definition associated with the estimator being reported. Under the usual kinship-coefficient convention, a duplicate pair is expected to have kinship near \(0.5\), while a first-degree relative pair is expected to be near \(0.25\). The PI_HAT quantity, when defined as \(Z_2+\tfrac{1}{2}Z_1\), is near \(1\) for duplicates and near \(0.5\) for first-degree relatives. These are reference expectations, not exact cutoffs: estimates vary with marker information, missingness, and technical quality.

In this convention, \(Z_1\) and \(Z_2\) are estimated proportions of the genome at which a pair shares one or two alleles identical by descent, respectively. A pair of samples from the same diploid genotype is expected to share both alleles throughout, so \(Z_2\) is near one and \(Z_0\) and \(Z_1\) are near zero. Close relatives share less consistently. The precise relatedness estimate can suggest a relationship class, but it generally cannot identify the exact relationship or distinguish every type of close relative.

Concordance provides a separate check. If two high-quality measurements represent the same underlying genotype, their calls should agree at nearly all reliable, overlapping markers. A genuine relative will generally have many different genotype calls, even when the relatedness estimate is high. However, raw concordance depends strongly on marker allele frequencies, so a high value is not, on its own, proof of duplication.

Key distinction: Relatedness measures allele sharing across loci; concordance measures exact genotype agreement at jointly called loci. A pair can be highly related without having nearly identical genotypes.

Calculating and Interpreting Concordance

First harmonize marker identity, genome build, strand, and allele coding so that a call for one sample is directly comparable with the call for the other. Then retain markers meeting the chosen quality criteria where both samples have a usable genotype. If \(M\) is the number of such markers and \(A\) is the number with identical genotype calls, the observed concordance is \(A/M\). Markers missing in either sample are not counted as matches or mismatches.

For a biallelic marker, a match means the same unordered diploid genotype, such as \(AA\) with \(AA\), or \(AG\) with \(AG\). A heterozygous call and a homozygous call are discordant. Quality filters matter: a large fraction of low-quality calls can make concordance misleading, while too few jointly called markers leave the comparison weakly supported.

Worked Example: Checking a Likely Duplicate Pair

Two hypothetical records, S-041 and S-316, have a kinship estimate of \(0.498\). Their estimated IBD proportions are \(Z_0=0.001\), \(Z_1=0.004\), and \(Z_2=0.995\). At 50,000 high-quality autosomal markers called in both samples, 49,970 genotype calls match. The records have different sample labels.

Relatedness check. Using the PI_HAT definition,

\text{PI\_HAT}=Z_2+\frac{1}{2}Z_1 =0.995+\frac{1}{2}(0.004) =0.997.

The kinship estimate is also near \(0.5\), and the IBD estimates place nearly all of the genome in the two-allele-sharing state. Together these values are strongly compatible with duplicate genotypes.

Concordance check.

C=\frac{A}{M} =\frac{49{,}970}{50{,}000} =0.9994.

Thus the samples agree at 99.94% of jointly called markers, with \(50{,}000-49{,}970=30\) discordant calls. This close agreement supports a duplicate-sample explanation, although the different labels still need to be checked against sample records. The genetic data alone cannot determine whether these are two aliquots from one person, repeat measurements, or—in unusual cases—genetically indistinguishable twins.

Why Raw Concordance Is Not a Universal Duplicate Threshold

Genotype matching can be common even for unrelated individuals at markers where one allele is very frequent. To see why, suppose unrelated individuals are sampled from a population in Hardy–Weinberg equilibrium at a biallelic marker, with allele frequency \(p\) for one allele and \(1-p\) for the other. The genotype probabilities are \((1-p)^2\), \(2p(1-p)\), and \(p^2\). Under independence between the two individuals, the probability of an exact genotype match is the sum of the squared genotype probabilities.

P(\text{match}) =(1-p)^4+\{2p(1-p)\}^2+p^4.

For \(p=0.5\), the three genotype probabilities are \(0.25\), \(0.50\), and \(0.25\), so the match probability for unrelated individuals at this marker is

$$ 0.25^2+0.50^2+0.25^2 =0.0625+0.25+0.0625 =0.375. $$

When one allele is rare, the common homozygous genotype dominates, and the chance of an unrelated match at that marker is much higher. Therefore, a concordance percentage should be interpreted in the context of which markers were included, their frequencies, the population being studied, and the genotype-calling process. A value that is informative on a panel of common, quality-controlled markers may not be comparable to the same percentage calculated on a different panel.

Practical rule: Treat concordance as supporting evidence, not as a universal pass/fail threshold. Compare pairs using a defined set of reliable, overlapping markers, and consider whether their allele frequencies and data-processing history make the percentage interpretable.

Worked Example: A Close Relative Is Not a Genotype Duplicate

In a hypothetical data set, two samples have a kinship estimate of \(0.247\), with \(Z_0=0.015\), \(Z_1=0.970\), and \(Z_2=0.015\). They have identical calls at 30,700 of 50,000 jointly called markers. Their labels and collection records identify them as two different people.

PI_HAT calculation.

\text{PI\_HAT} =Z_2+\frac{1}{2}Z_1 =0.015+\frac{1}{2}(0.970) =0.500.

The kinship estimate is near the usual first-degree reference of \(0.25\), and PI_HAT is near \(0.5\). The IBD pattern is therefore compatible with close relatives. It does not, by itself, establish which first-degree relationship applies.

Concordance calculation.

C=\frac{30{,}700}{50{,}000}=0.614.

The samples match at 61.4% of jointly called markers, far below the near-complete agreement expected from two high-quality measurements of the same genotype. The relatedness estimate and record evidence support treating this as a close-relative pair rather than a duplicate pair. The concordance value is descriptive for this marker set; it should not be generalized as the expected percentage for every pair of first-degree relatives.

Combining the Evidence and Recording Decisions

A defensible resolution uses a sequence of checks rather than a single cutoff. Begin with the candidate pair and confirm that marker and allele coding are aligned. Check the relatedness estimate and its uncertainty, then calculate concordance among quality-controlled, jointly called markers. Finally, compare the evidence with identifiers and provenance: sample aliases, collection records, plate or batch positions, and any documented repeat sampling. An identifier match can support a duplicate interpretation, but a label mismatch is not proof that the samples are different people.

Verify the comparison.
Confirm that the two records refer to the same markers with compatible allele coding. Record the number of jointly called markers and the filters used.

Review relatedness.
Use the estimator’s documented scale. Note the estimate, uncertainty where available, and whether the result is compatible with a duplicate or a close-relative candidate.

Review genotype concordance.
Calculate the matching fraction on eligible markers, inspect discordant calls, and consider whether poor call quality, missingness, or batch differences could explain disagreement.

Check identity records and classify cautiously.
Compare sample labels and provenance. Record a likely duplicate, likely distinct relatives, or unresolved candidate, along with the evidence and any follow-up needed.

For an unresolved pair, useful follow-up can include checking the original sample manifest, plate maps, collection dates, and laboratory records; reviewing call rates and discordant markers by batch; and, if material is available, repeating genotyping from a retained aliquot. A pattern of discordance concentrated in one batch suggests a possible technical issue, while widespread agreement with a few isolated mismatches may be consistent with measurement error. Neither pattern should be assumed without examining the data.

Worked Example: Documenting an Ambiguous Candidate

Two hypothetical records have kinship \(0.460\) and PI_HAT \(0.910\), but only 4,000 markers are jointly called. Of these, 3,800 calls agree. Their sample labels differ, and the records show that the samples were processed in different batches.

Concordance calculation.

C=\frac{3{,}800}{4{,}000}=0.950, \qquad 4{,}000-3{,}800=200\text{ discordant calls}.

The relatedness estimates are high, but the 95.0% concordance is not near-complete agreement. The overlap is also much smaller than in the earlier examples, and batch differences could contribute to the discordance. As a rough calculation only, if marker outcomes were independent Bernoulli trials, the standard error of the observed concordance would be

\sqrt{\frac{C(1-C)}{M}} = \sqrt{\frac{0.950(0.050)}{4{,}000}} \approx 0.00345.

This calculation does not resolve the case: linkage means marker outcomes are not generally independent, and the standard error does not account for batch effects or genotype-calling errors. The appropriate record is unresolved high-relatedness candidate, not “duplicate.” The next steps are to check the batch-specific quality patterns and sample manifest, and to seek an independent genotype measurement if one is available.

Common Mistakes

Calling every high-kinship pair a duplicate. Close relatives can have high relatedness estimates. Check exact genotype agreement and sample records before assigning a duplicate label.
Treating high concordance as conclusive. Common genotypes can match frequently even among unrelated people. Interpret concordance using the marker set and its allele-frequency properties.
Comparing estimator scales as if they were the same. Kinship and PI_HAT have different usual reference values. Confirm the definition and scale of the reported column before interpreting it.
Ignoring the denominator. A percentage based on a small number of jointly called markers can be unstable or unrepresentative. Report the number of eligible markers alongside concordance.
Forcing a decision when evidence conflicts. A high relatedness estimate, appreciable genotype discordance, and inconsistent records warrant follow-up, not an unsupported categorical conclusion.

Key takeaway: A duplicate interpretation is strongest when near-duplicate relatedness estimates, near-complete high-quality genotype concordance, and sample-provenance evidence agree. When these sources conflict or overlap is limited, retain the pair as ambiguous and document the checks needed to resolve it.

Check Your Understanding

Use the definitions and decision principles in this tutorial to answer the following questions.

Under the stated conventions, what kinship and PI_HAT values are expected approximately for a duplicate pair? How do the usual first-degree reference values differ?
A pair matches at 9,600 of 10,000 jointly called markers. Calculate its concordance and name two other kinds of evidence needed before calling it a duplicate.
Why can two unrelated individuals have high genotype concordance at a marker where one allele is very common?
A pair has high relatedness estimates but only 2,000 jointly called markers and substantial discordance. What should be documented before reaching a classification?
Why can genotype data alone fail to distinguish duplicate records from genetically indistinguishable twins?