How SNP Arrays Produce Genotype Calls

From Probe Signal to Genotype

A single-nucleotide polymorphism (SNP) array does not directly observe a genotype. It measures how strongly labeled DNA from a sample binds to probes designed for alternative alleles at many genomic positions. The observed fluorescence intensities provide evidence about the alleles present; software uses the pattern of those intensities to assign a genotype call.

For a diploid sample, the possible genotypes at a biallelic SNP are homozygous for one allele, heterozygous, or homozygous for the other allele. In the notation used here, call the two assay alleles A and B. These labels describe the assay’s two allele channels; they do not, by themselves, say which allele is the reference allele. If B is the allele whose dosage is being counted, the corresponding diploid dosages are 0 for AA, 1 for AB, and 2 for BB.

In a typical assay, sample DNA is prepared and labeled, then allowed to hybridize to many short probes. Probes are designed so that binding differs according to the allele at a target SNP. After unbound material is washed away, a scanner measures fluorescence. An allele-specific signal is usually stronger when the sample DNA matches the corresponding probe, but the observed intensity is also affected by background, probe behavior, sample preparation, and measurement variation. Thus, a signal is evidence about genotype, not a direct count of DNA molecules or a genotype label on its own.

For a given SNP and sample, represent the two measured, nonnegative allele-channel intensities by \(I_A\) and \(I_B\). In an idealized case, an AA sample has stronger A-channel signal, an AB sample has appreciable signal in both channels, and a BB sample has stronger B-channel signal. Real measurements vary, so software examines many samples at the SNP together and looks for groups, or clusters, of similar signal patterns.

Signal Balance and Cluster Positions

A useful way to understand a cluster plot is to separate two questions: which allele channel dominates? and how much total signal is present? Relative balance helps distinguish AA, AB, and BB. Total intensity helps show whether the measurements are strong or unusually weak. Plot axes differ by platform and software, so a cluster plot must be read using its axis labels and documentation.

For illustration, define a balance coordinate from the two nonnegative intensities as follows. The two-argument function \(\operatorname{atan2}(I_B,I_A)\) gives the angle of the point \((I_A,I_B)\), using the signs of both coordinates to determine the angle. Because these signals are nonnegative, its angle is between 0 and \(\pi/2\), inclusive. The balance coordinate below rescales that angle to the interval from 0 to 1. It is an illustrative transformation, not a claim that every array platform uses this exact formula.

When \(I_A>0\), the same expression can be written as \(\theta=(2/\pi)\arctan(I_B/I_A)\). The \(\operatorname{atan2}\) form also handles a zero intensity in one channel: if \(I_B=0\) and \(I_A>0\), then \(\theta=0\); if \(I_A=0\) and \(I_B>0\), then \(\theta=1\). If both intensities are zero, the angle and balance are undefined, because there is no signal direction to measure. That case should not be assigned a balance value by this formula.

For positive, equal signals, \(\operatorname{atan2}(I_B,I_A)=\pi/4\), so \(\theta=0.5\). The coordinate therefore moves from the A-dominated end toward the B-dominated end, while equal channel signals lie at the midpoint. This does not mean every heterozygote has exactly equal measured intensities; it describes the idealized center of a heterozygote cluster.

A second illustrative coordinate is total intensity, for example \(T=I_A+I_B\), or a normalized version of that sum. A normalized total signal can be plotted vertically against balance horizontally. In such a plot, three well-separated bands of points may appear near the AA, AB, and BB balance regions, while their vertical spread reflects differences in total signal. Other platforms use different transformations, such as coordinates derived from normalized allele intensities. The interpretation depends on the specific axis definitions, but the general idea is that relative signal distinguishes allele composition while total signal can reveal weak or atypical measurements.

Why Three Clusters Are Expected

The basic cluster pattern follows from the three possible diploid genotypes. Suppose, only as a simplified signal model, that the B allele contributes in proportion to its copy count \(g\), and the A allele contributes in proportion to the remaining copy count \(2-g\), with a common positive scale factor \(k\). Then \(I_A=k(2-g)\) and \(I_B=kg\). This is not a precise physical model of hybridization; it provides a compact justification for the expected balance ordering.

The endpoints are evaluated with \(\operatorname{atan2}\), so neither requires division by zero. Since \(k>0\), none of these three idealized signal pairs is the both-zero case. Real samples do not fall exactly at these ideal points: intensities vary, clusters have width, and some samples can fall between or outside the expected groups.

Software typically processes and normalizes signal measurements before assigning calls. Depending on the platform and calling method, the software may estimate cluster centers and their spread from a reference set or from the data, then assess how well each sample fits the genotype clusters. A call is made when the signal pattern supports one genotype sufficiently well under the method’s criteria. If support is weak, a sample may be marked as a no-call instead of being forced into the nearest group. The exact transformations, model, and confidence thresholds are platform-specific.

How to Interpret a Cluster Plot

A cluster plot is useful because it shows both the genotype groupings and individual samples that do not fit them well. When interpreting one, first identify the axes and their scales. Then locate the expected homozygous and heterozygous regions, and assess whether the plotted points form distinct, reasonably compact groups. Finally, examine individual points for weak total signal, separation from their likely cluster, or placement between groups. A visual inspection helps explain the call pattern, but does not replace the software’s documented calling procedure.

One reason low total signal matters is that a ratio or angle is driven by relative channel values. When both measurements are small, modest background or measurement variation can substantially change their relative balance. A sample may therefore appear to have a particular balance while providing little reliable evidence for any genotype. Conversely, an unusual point is not automatically an incorrect genotype: it should be evaluated using the assay’s calling criteria and appropriate follow-up information.

Common Mistakes and Practical Meaning

• Treating intensity as a direct genotype readout. A higher signal in one channel is evidence favoring that allele, not a direct observation of allele copies. Calls depend on the joint signal pattern and the calling method.
• Assuming all platforms use the same axes or formula. Balance and total intensity are useful concepts, but software may transform and normalize signals differently. Read the plot labels and assay documentation before interpreting coordinates.
• Calling every midpoint sample heterozygous. A balance near the middle can be compatible with AB, but a weak or distant point may not meet the caller’s confidence criteria. Consider cluster fit and total intensity together.
• Using a ratio when its denominator is zero. The expression \(I_B/I_A\) is undefined when \(I_A=0\), and the balance is undefined if both channels are zero. The \(\operatorname{atan2}\) form handles a single zero channel for nonnegative intensities, while correctly leaving the both-zero case undefined.
• Forcing a genotype for every sample. A no-call preserves uncertainty when the measured pattern does not support a sufficiently confident assignment. It is not evidence that the sample has a fourth genotype.

Reliable cluster separation matters because genotype calls are the inputs to many later genetic analyses. An incorrect or uncertain call can affect a sample’s allele dosage and downstream estimates. The essential interpretation is modest: a cluster position is an intensity-based assignment supported to some degree by the assay and software, not a guarantee that the underlying genotype has been observed without error.