Why Probe Design Matters
Infinium methylation arrays estimate methylation at many CpG sites by measuring fluorescent signal from bisulfite-converted DNA. The resulting methylated and unmethylated intensities are often summarized as a beta value, as covered in the earlier material on beta values and M values for CpG sites. But those intensities are not generated in exactly the same way at every locus. Infinium Type I and Type II designs encode methylation state using different probe and color-channel arrangements. This distinction can affect signal strength and the shape of observed distributions, even when the biological quantity of interest is the same.
The central point is that a beta value is a ratio of measured signals, not a direct count of methylated molecules. Its interpretation depends on how well the assay captures the two signal components. Comparing probes therefore requires attention to both the numerical summary and the design that produced it.
From Bisulfite Conversion to Array Signal
In the standard bisulfite-based assay, an unmethylated cytosine is converted to a uracil and is read as a thymine after amplification, while a methylated cytosine is generally retained as cytosine. The array probes are designed to recognize the sequence patterns that result. Fluorescent intensities then provide measurements associated with methylated and unmethylated DNA at each locus. In this setting, those two components are commonly denoted \(I^M\) and \(I^U\), respectively.
For Type I, the methylated-specific and unmethylated-specific probe sequences provide separate measurements. The two probes use the same color channel at that locus, although which channel is assigned can differ across loci. For Type II, a single probe binds the locus, and a single-base extension step produces different color-channel signals for the two sequence states. The exact mapping from methylation state to color depends on the probe and assay design; “red” and “green” do not mean methylated and unmethylated universally.
Both designs therefore produce a pair of methylated and unmethylated measurements, but the pair is assembled differently. In Type I, the distinction is primarily between two probe sequences. In Type II, it is primarily between two color signals generated from one probe. This is the design difference to keep in mind when inspecting intensities.
Worked Example: Identifying the Signal Components
Suppose a hypothetical Type I locus has two bead types. The methylated-specific bead produces an intensity of 720 units, and the unmethylated-specific bead produces 240 units. Both measurements use the same color channel for this locus. A hypothetical Type II locus instead has one bead type, with the two color channels yielding 480 units and 160 units, assigned by its design to methylated and unmethylated signal.
For the Type I locus, \(I^M=720\) and \(I^U=240\), because the two probe types identify the two sequence states. For the Type II locus, \(I^M=480\) and \(I^U=160\), because the design assigns the two color signals to those states. The channel labels alone do not establish which signal is methylated; that assignment comes from the locus-specific design.
The sums differ: the Type I signals total \(720+240=960\) units, while the Type II signals total \(480+160=640\) units. Those totals do not, by themselves, show that one locus has more DNA or greater methylation. Signal scale can reflect probe, channel, and assay efficiencies as well as the underlying sample.
How the Design Enters a Beta Value
A common summary of methylation at one locus is the beta value, a ratio between zero and one. One standard form includes a positive offset \(c\), chosen by the processing convention, to stabilize the ratio when intensities are small:
When the offset is negligible relative to the signals, the ratio is approximately the fraction of the measured methylated-plus-unmethylated signal attributed to the methylated component. It is not automatically the exact fraction of DNA molecules that are methylated: that interpretation also depends on assay response. With \(c=0\), multiplying both signal components by the same positive constant leaves the ratio unchanged. With a nonzero offset, this scale invariance is only approximate when the signals are large relative to \(c\).
A short measurement model shows why probe design can matter. Let \(\theta\) be the underlying methylated fraction, and suppose the methylated and unmethylated signals are proportional to their respective fractions but have response factors \(a_M\) and \(a_U\):
where \(q\) represents a shared scale factor such as the amount of available target DNA. If \(c=0\), substituting these expressions into the beta ratio gives
If \(a_M=a_U\), then \(\beta=\theta\) under this simplified model. If the response factors differ, the measured ratio need not equal the underlying fraction. For Type I, response differences can arise between the two probe sequences; for Type II, they can arise between color-channel responses and the sequence-extension signals. The model is a justification for caution, not a complete description of array chemistry: real measurements can also include background and other technical variation.
Worked Example: A Shared Scale Change and an Offset
Consider hypothetical intensities \(I^M=600\) and \(I^U=200\), with \(c=0\). The beta value is
If both intensities are multiplied by 2, the beta value remains \(1200/(1200+400)=0.750\). This follows because the common factor cancels from the ratio. Thus, without an offset, a shared increase in both signals changes total intensity but not beta.
Now use \(c=100\). The original signals give \(600/(600+200+100)=0.6667\). After doubling both signals, the result is \(1200/(1200+400+100)=0.7059\). The offset does not double, so the ratio changes. The two beta values differ even though the methylated-to-unmethylated intensity ratio is unchanged. This is why the offset convention and overall signal scale matter, especially for weak signals.
Comparing Type I and Type II Measurements
| Feature | Type I | Type II |
|---|---|---|
| Probe types per CpG | Two: methylated-specific and unmethylated-specific | One probe |
| How the states are distinguished | By the two probe sequences | By two color-channel signals after extension |
| Color arrangement | Both state signals use the same channel for that locus | The two state signals use different channels |
| Potential source of unequal response | Differences between the two probe sequences | Differences between channels or extension responses |
| Implication for comparison | Signal reflects two probe responses | Signal reflects one probe and two channel responses |
Neither design is universally “better” based on the design label alone. Type I spends two probe types on each locus and reads both through the same channel; Type II uses one probe and two channels. Each arrangement has different routes by which sequence binding, extension, or channel response can affect the measured pair. Consequently, intensity distributions can differ by probe type, and beta values can show type-associated patterns even when the loci have similar underlying methylation.
Such patterns are not proof of a particular technical failure, nor are they evidence that methylation biology must differ by design. They are a reason to examine distributions with probe type in view. For instance, if one design tends to have a different total intensity range or a different spread of beta values, pooling all probes into one distribution can obscure that structure. Conversely, a difference between the two designs does not establish that every probe of one type is biased in the same direction.
Worked Example: Similar Ratios, Different Signal Scales
Suppose a hypothetical Type I probe has \(I^M=900\) and \(I^U=300\), while a hypothetical Type II probe has \(I^M=300\) and \(I^U=100\). Use \(c=0\) for this illustration.
The Type I beta value is \(900/(900+300)=0.750\). The Type II beta value is \(300/(300+100)=0.750\). The total intensities are 1200 and 400 units, respectively, despite the matching ratios.
This arithmetic shows that a common multiplicative signal-scale difference need not change beta when there is no offset. It does not show that all Type I and Type II probes are interchangeable: the hypothetical values assume the same methylated-to-unmethylated response ratio. If that ratio differs because the two signal components have different efficiencies, the beta values can differ even at the same underlying methylation fraction. The example therefore distinguishes one kind of scale difference from a response imbalance.
Reading Distributions Without Overinterpreting Them
When viewing array data, compare intensity and beta-value distributions by probe type as well as across the full set of loci. This descriptive check can reveal whether the two designs occupy different ranges or have different spreads. It does not identify the cause of a difference on its own. Loci differ in sequence context and biological methylation, and design is not assigned randomly across all genomic features. A difference between type-specific distributions can therefore reflect a mix of design response and which loci belong to each group.
At an individual locus, inspect the methylated and unmethylated components alongside the beta value. A beta value near 0.5 formed from two strong signals is not equivalent, in measurement strength, to the same ratio formed from two very weak signals. The latter is more affected by offsets and background. Likewise, a large total intensity does not imply high methylation: methylation is encoded by the relative signals, not their sum alone.
For comparisons across loci, treat beta values as useful summaries, not as guarantees of identical measurement properties. Probe type is one relevant design feature, but it is not a complete explanation of a locus’s behavior. More detailed handling of probe-level problems and normalization belongs to separate analysis steps; the design label is a cue for interpretation, not a substitute for those procedures.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions below.
- How many probe types are used for one Type I CpG, and how many for one Type II CpG?
- A Type I locus has \(I^M=450\), \(I^U=150\), and \(c=0\). Calculate its beta value.
- Why does doubling both signal components leave beta unchanged when \(c=0\), but not necessarily when \(c\) is positive?
- Can a difference in total intensity between a Type I and a Type II locus establish that their methylation levels differ? Explain.
- Why should a researcher avoid assuming that “red” always means methylated across loci?
Solutions. (1) Type I uses two probe types per CpG; Type II uses one probe. (2) \(\beta=450/(450+150)=0.750\). (3) With \(c=0\), a shared multiplier cancels from numerator and denominator. With positive \(c\), the offset does not scale with the signals, so the ratio can change. (4) No. Total intensity is the sum of measured components and can differ because of signal scale or response, independently of the methylated-to-unmethylated ratio. (5) The channel-to-state assignment depends on probe design; color is not a universal methylation label.