What an Intensity Cluster Represents
A genotyping array measures signal from probes designed to distinguish two alleles at a locus. The earlier material on how genotyping arrays measure alleles establishes the basic idea: a sample produces signal in two allele-specific channels. A two-dimensional intensity plot places one channel on each axis, so each sample appears as a point. The pattern of points across samples can reveal whether the two allele signals form groups consistent with the three possible genotypes.
For clarity, call the two alleles A and B, and let \(I_A\) and \(I_B\) denote the corresponding plotted intensities. The axes may be labelled differently in a particular plot, and signal scales depend on how the array data were processed. The interpretations below assume that both channels are measured on comparable scales and that plotted intensities have had appropriate background handling. The units are often arbitrary signal units rather than physical measurements.
In an idealized signal pattern, an A/A sample has relatively strong A signal and weak B signal, an A/B sample has appreciable signal in both channels, and a B/B sample has relatively weak A signal and strong B signal. Thus A/A tends toward the horizontal axis, A/B toward the diagonal, and B/B toward the vertical axis. “Weak” does not mean exactly zero: background and imperfect allele discrimination can leave signal in the other channel.
These are qualitative expectations, not universal numeric cutoffs. Real clusters have spread, their centers need not lie at exact idealized coordinates, and their positions can vary from locus to locus. A plot is most informative when the allele labels, axes, and sample set are clear and all points being compared were measured and processed comparably.
Reading Position in Two Dimensions
The first question is whether the plot shows three distinguishable groups in the expected arrangement. Look for one group with higher \(I_A\) than \(I_B\), one with substantial signal in both channels, and one with higher \(I_B\) than \(I_A\). Then assess how tightly points gather around each group and whether the groups are separated. Do not infer a genotype from one coordinate alone: a point can have moderate A signal because its B signal is also high, or because both signals are weak.
A useful descriptive summary is allele balance, the fraction of the combined signal in the B channel, together with total signal:
When \(I_A+I_B>0\), the B-signal fraction is near 0 for an A-dominant point, near 1 for a B-dominant point, and often intermediate for a point with both signals. In an ideal equal-response example, an A/B point has a B-signal fraction near 0.5. This is a description of measured signal, not a guarantee that the biological genotype has that exact signal ratio. The fraction also discards total signal: points with very different combined intensities can have the same fraction. The original two-dimensional plot retains both pieces of information.
Worked Example: Interpreting Three Well-Separated Clusters
Suppose a hypothetical locus has three compact groups, with approximate centers \((I_A,I_B)=(900,100)\), \((500,500)\), and \((100,900)\), in arbitrary signal units. A sample has intensities \((480,520)\). Describe its position and the evidence it provides.
Its B-signal fraction is
Its total signal is \(480+520=1000\) units. The point is close to the group centered at \((500,500)\), with substantial signal in both channels, and its B-signal fraction is close to 0.5. If the plotted groups are compact and separated, this position supports an A/B interpretation. It is evidence from the cluster pattern, not proof based only on the calculated fraction.
Overlap, Spread, and Displacement
Cluster quality matters as much as cluster location. If two genotype groups overlap, a point in the shared region is not clearly distinguished by its intensities. If a group is widely spread, points far from its center may be less confidently interpreted than points near the center. If a group is displaced from its expected relative position, its members may have unusual signal balance. These patterns are reasons to flag an interpretation as uncertain; they do not, on their own, establish why the pattern occurred.
A cluster plot is a visual diagnostic, not a rule that assigns every point to the nearest group. Automated calling methods may use more information and explicit models, but their assignments also depend on assumptions and data quality. If a plot does not display clearly separated groups, forcing every point into one of three classes can conceal uncertainty rather than resolve it.
Worked Example: A Point in an Overlap Region
In a hypothetical plot, the A/A group centers near \((800,150)\), while the A/B group centers near \((550,450)\). Their point clouds overlap around \((650,300)\). A sample has intensities \((640,310)\). Determine what can reasonably be concluded.
For this sample, the B-signal fraction is
The total signal is \(640+310=950\) units. The fraction is between the strongly A-dominant pattern expected for A/A and the more balanced pattern expected for A/B. More importantly, the specified plot shows that the sample lies in a region occupied by both groups. The intensities therefore do not clearly distinguish those two interpretations. It would be unjustified to choose A/A merely because \(I_A>I_B\), or to choose A/B merely because both signals are present. The sample should be treated as uncertain based on this plot.
Displacement differs from overlap. A group may be compact but shifted away from the position expected relative to the other groups. To describe displacement, compare the group’s location with the other groups in the same plot and check whether the shift is evident in the points rather than driven by a single outlier. When intensity scales differ or are transformed, raw coordinate distances can be misleading; visual comparison should respect the axes and their scales.
Worked Example: A Compact but Displaced Heterozygous Group
Consider a hypothetical plot whose A/A and B/B groups center near \((900,100)\) and \((100,900)\). A/B samples would be expected to form a group with substantial signal in both channels, roughly between those homozygous groups. Instead, the observed A/B-like group is compact around \((650,380)\). Compare its balance with the equal-signal midpoint \((500,500)\).
At the observed center, the B-signal fraction and total signal are
At the equal-signal midpoint, the B-signal fraction would be \(500/(500+500)=0.500\), with total signal \(1000\) units. The observed group is shifted toward the A-dominant side and has a lower B-signal fraction. If its points are genuinely compact, the group is not diffuse, but its unusual position is still a reason to review its genotype interpretation. The comparison describes the geometry; it does not establish the cause of the shift or prove that each member is miscalled.
Assumptions, Checks, and Common Pitfalls
Reading these plots relies on several practical conditions. The two channels must correspond to the intended alleles; the axes must be understood; and the samples being compared should be measured on a sufficiently comparable scale. The three-cluster expectation also assumes a diploid, biallelic setting in which the relevant samples can be represented by A/A, A/B, or B/B. If these conditions are not reasonable, the standard three-group picture may not apply. Even when they are reasonable, intensity patterns are noisy measurements rather than direct observations of genotype.
Before interpreting a suspicious point or group, check whether the axes use the same scale throughout, whether points are plotted for the same locus, and whether one isolated point is distorting the apparent cluster. Examine the whole configuration: a balanced signal ratio at very low total intensity is not equivalent to a balanced ratio at a strong signal level. Where sample groups, plates, or batches are shown separately, check that the apparent shift is not simply a consequence of comparing differently processed or differently scaled plots.
A common mistake is to reason, “A/B has signal in both channels, and this sample has both signals, so the call is certainly A/B.” The correction is that both-channel signal is expected for A/B, but overlap, weak or unusual signal, and cluster displacement can make the assignment uncertain. Another mistake is to treat the B-signal fraction as a complete substitute for the two-dimensional plot. It cannot show total signal, and different points can have the same fraction but very different combined intensity.
| Plot feature | What it supports | What it does not establish |
|---|---|---|
| Three compact, separated groups in the expected arrangement | The intensities distinguish the three genotype classes relatively clearly | That every individual call is certainly correct |
| Two groups overlapping | Points in the shared region have ambiguous intensity evidence | Which genotype is correct for each ambiguous point |
| A broad or scattered group | Members vary substantially in their measured signal pattern | A particular cause of that variation |
| A compact but displaced group | The group’s signal balance differs from the expected relative pattern | That the group is wrong, or why it is displaced |
The main alternative to visual inspection is a model-based or algorithmic genotype-calling procedure. Such procedures can apply consistent criteria across many samples, while a plot makes overlap, spread, and relative displacement visible to a reviewer. Neither approach removes the need to check whether the underlying data and assumptions are appropriate. A one-dimensional allele-balance display can make ordering along the A-to-B direction easier to see, but it suppresses total intensity and therefore cannot replace the two-channel view when signal strength is relevant.
Check Your Understanding
Answer each question using the stated intensity patterns. The solutions follow the questions.
- A sample has intensities \((I_A,I_B)=(240,760)\). Calculate its B-signal fraction and describe the pattern it suggests, without claiming certainty.
- Why might two samples with B-signal fractions of 0.50 provide different quality of intensity evidence?
- A point lies halfway between the A/A and A/B clusters in a plot where those groups overlap. What is the appropriate interpretation?
- A compact group has both channels appreciably high but is shifted toward the A/A group compared with the other clusters. What can be concluded, and what cannot?
Solutions. For question 1, the fraction is \(760/(240+760)=760/1000=0.760\). The point is B-dominant and is consistent with the B/B side of the plot, but its location relative to the actual clusters determines how strongly that interpretation is supported. For question 2, the total intensities may differ: for example, \((50,50)\) and \((500,500)\) both have fraction 0.50, but their total signals are 100 and 1000 units. The fraction alone omits that difference. For question 3, the point is ambiguous because its position is shared by two clusters; the plot does not clearly distinguish the corresponding genotypes. For question 4, the group’s compactness and shifted position can be described, and the shift is evidence that its intensity pattern is unusual. The plot alone cannot identify the cause or establish that the group’s genotype assignments are wrong.