Estimating Close Relationships From Genome-Wide Genotypes

Why Use Kinship and IBD Together?

A genome-wide estimate of relatedness can indicate that two people are genetically close, but a single number does not always identify the relationship. First-degree pairs, such as parent–offspring and full siblings, have the same expected kinship coefficient, even though their identity-by-descent (IBD) patterns differ. Second-degree relationships, such as half siblings or grandparent–grandchild, have a lower expected kinship. Comparing kinship with estimated IBD-state proportions therefore provides a more informative classification than interpreting either alone.

The goal is to classify the genetic evidence, not to prove a pedigree. A pair with a pattern typical of second-degree relatives may have several possible genealogical relationships. Estimates also vary across the genome and can be affected by genotype quality and the allele-frequency and population-structure handling used by the estimator. The discussion assumes that those choices have been made appropriately for the data, as in robust kinship estimation.

Expected Patterns for Common Relationship Classes

These values describe pedigree expectations across the genome, not exact requirements for every observed pair. For unrelated individuals in the reference population, the expected pattern is \(k_0=1\), \(k_1=k_2=0\), with \(\phi=0\) and \(r=0\). For a second-degree relationship such as half siblings, the expected IBD proportions are \(k_0=0.5\), \(k_1=0.5\), and \(k_2=0\). Thus \(\phi=0.5/4=0.125\) and \(r=2\phi=0.25\). The second-degree relationship coefficient is 0.25, not 0.125; 0.125 is the corresponding kinship coefficient.

A parent–offspring pair has one allele IBD at essentially every autosomal locus, so its expected pattern is \(k_0=0\), \(k_1=1\), \(k_2=0\). This gives \(\phi=0.25\) and \(r=0.5\). Full siblings have expected proportions \(k_0=0.25\), \(k_1=0.5\), and \(k_2=0.25\), which also give \(\phi=0.25\) and \(r=0.5\). Unlike parent–offspring pairs, however, full siblings can share zero or two alleles IBD in different genomic regions.

The table summarizes the expected patterns. A measured pair need not match every entry exactly, particularly when estimates are based on a finite number of markers or genomic segments.

From Genome-Wide Estimates to a Classification

Let \(\hat{\phi}\) be the estimated kinship coefficient for a pair, and let \(\hat{k}_0,\hat{k}_1,\hat{k}_2\) be estimated IBD-state proportions. If the estimator provides all three IBD proportions on a compatible scale, calculate \(\hat{\phi}=\hat{k}_1/4+\hat{k}_2/2\) and \(\hat{r}=2\hat{\phi}\). If a method estimates kinship directly, use that estimate as a complementary summary rather than assuming it is numerically interchangeable with an IBD fraction. Always check the method's scale and definition before comparing values.

A practical interpretation looks for agreement between the kinship estimate and the IBD pattern. A near-zero kinship estimate and a pattern dominated by \(k_0\) are consistent with an unrelated pair. An estimate near the second-degree expectation, with appreciable \(k_0\) and \(k_1\) but little \(k_2\), supports a second-degree classification. A higher kinship estimate supports first-degree relatedness; a pattern with \(k_1\) near one is especially consistent with parent–offspring, whereas appreciable \(k_0\) and \(k_2\) support full siblings.

The assumptions behind this interpretation matter. The genotypes should be sufficiently accurate and the markers should provide useful genome-wide information; the estimates should not be dominated by missingness or a small set of problematic regions. The kinship method must be appropriate to the population structure and allele-frequency choices in the data. Also, the expected patterns presume the specified relationship class and ordinary autosomal inheritance; unusual pedigrees or technical artifacts can produce departures. In practice, classification is evidence-based and may remain uncertain.

Why the IBD Pattern Separates First-Degree Pairs

Kinship alone cannot distinguish parent–offspring from full siblings because both have expected \(\phi=0.25\). The IBD-state proportions provide the relevant contrast: parent–offspring pairs are expected to have \(k_1\) near one, whereas full siblings have genomic regions in both the zero-IBD and two-IBD states. The distinction follows directly from inheritance. A child receives one allele at each autosomal locus from each parent, so a parent and child share one transmitted allele IBD at that locus. Two siblings inherit independently from their parents; at a locus they can share neither parental allele, one, or both.

The expected full-sibling proportions are averages over loci and across possible inheritance outcomes. They do not mean that every full-sibling pair has exactly those proportions. Recombination creates long runs of shared IBD states, so the number and lengths of observed segments affect how close a realized pair is to the expectation.

Diagnostics, Alternatives, and Limitations

Before assigning a relationship class, check that \(\hat{k}_0+\hat{k}_1+\hat{k}_2\) is close to one and that no estimate is materially outside the range from zero to one. Small deviations can arise from rounding or estimation; large deviations suggest incompatible scales, a reporting issue, or a method that does not estimate proportions as assumed. Also check whether the kinship value and IBD pattern tell a coherent story. For example, an apparently first-degree kinship estimate paired with a pattern dominated by \(k_0\) should prompt investigation rather than an automatic first-degree label.

A direct kinship estimate is compact and useful for screening many pairs, but it can hide distinctions between relationship types with the same expected kinship. IBD-state estimates reveal more of that structure, though they can be noisier and can depend on how genomic regions are modeled. Pedigree records, when reliable, offer a different source of evidence but are not a substitute for checking the genetic data. These approaches answer related but not identical questions; agreement strengthens a classification, while disagreement calls for review.

No single numerical cutoff is universal. Thresholds depend on the estimator, its scale, marker density, population composition, and the desired balance between false-positive and false-negative classifications. Close pairs also share large genomic segments, so markers are not equivalent to independent observations. A precise-looking estimate should not be mistaken for certainty, and an estimated second-degree class does not identify which second-degree pedigree applies.

Common Mistakes

• Confusing kinship with relationship coefficient. For a second-degree pair, the expected kinship coefficient is 0.125 and the expected relationship coefficient is 0.25. State which quantity is being reported.
• Classifying from kinship alone. First-degree parent–offspring and full-sibling pairs have the same expected kinship; use the IBD pattern to distinguish their typical profiles.
• Treating expected IBD proportions as exact values. Realized proportions vary across the genome. Compare the overall pattern and uncertainty, not exact equality to a pedigree expectation.
• Claiming a specific pedigree from a broad class. A second-degree pattern can fit several relationships. The genetic classification alone may not identify which one applies.
• Ignoring a mismatch between estimates. A disagreement is a diagnostic signal to check scale, markers, and assumptions, not a reason to choose one result without explanation.