Epigenomic Tables: Methylation Fractions and Accessibility Signals

Why Epigenomic Tables Need Careful Reading

Epigenomic tables often place several kinds of measurements side by side, but their entries do not all mean the same thing. A DNA methylation entry may estimate the fraction of observed molecules that are methylated at a locus. A chromatin-accessibility entry may be a fragment count or a processed signal for a genomic region. The first is bounded between zero and one; the second depends on the assay and processing and may have no such bound.

That distinction matters before comparing samples or relating the measurements to other molecular layers. A methylation value of 0.7 has a proportion-based interpretation, subject to how it was measured. An accessibility value of 0.7 usually does not mean that 70% of a region is accessible. Its meaning depends on the signal’s definition and units.

The earlier material on RNA sequencing tables introduced sample-by-feature measurements for sequencing assays. Here the focus is narrower: how to read methylation fractions and accessibility measurements without treating them as interchangeable quantities.

Methylation Fractions at a Locus

For example, if 42 of 60 informative molecules are classified as methylated, the observed fraction is \(42/60=0.70\). The value describes the observed molecules at that locus in that sample; it does not establish that every cell has the same methylation state. In a bulk tissue sample, the fraction can reflect a mixture of cell types, alleles, and molecular states.

Coverage is part of the interpretation. Two fractions can be numerically equal but based on very different amounts of information. A fraction of \(8/10=0.80\) is based on ten observations, whereas \(800/1000=0.80\) is based on one thousand. Neither value alone says how representative the sampled molecules are of the underlying biological material.

A common simple model makes this dependence explicit. Suppose, as an idealization, that the \(c_{ij}\) observations are independent Bernoulli trials with the same probability \(p_{ij}\) of being classified as methylated. Then \(m_{ij}\) has a binomial distribution with size \(c_{ij}\) and probability \(p_{ij}\), and \(\hat{\beta}_{ij}\) estimates \(p_{ij}\). Under those assumptions, its standard error is estimated by \(\sqrt{\hat{\beta}_{ij}(1-\hat{\beta}_{ij})/c_{ij}}\). This is a model-based summary of sampling variability, not a guarantee that read observations are truly independent.

Chromatin-Accessibility Signals

For example, ATAC-seq records DNA fragments associated with transposase access, while DNase-based assays record evidence related to nuclease cleavage. A table may report counts assigned to a region, or a processed signal derived from those counts. Raw counts are nonnegative integers; after normalization or transformation, values may be decimal and need not be integers. Such processing does not turn the measurement into a proportion.

An accessibility signal is evidence generated by a particular assay and analysis pipeline, not a direct, universal measurement of how “open” a region is. A higher signal can indicate more observed assay evidence under compatible measurement conditions. It does not, by itself, give the percentage of cells with accessible chromatin, the probability that a base is accessible, or a causal effect on gene activity.

Reading the Table Together

A sample-by-locus methylation table has one entry for each sample–locus combination, and a corresponding accessibility table may summarize regions rather than individual methylation sites. The locus definitions matter: a single CpG site and a wider genomic region are not necessarily the same feature. A table can therefore place values from related genomic locations together without making them identical measurements.

Methylation and accessibility may be biologically related in some contexts, but their relationship is not a universal rule. A high methylation fraction at one site does not automatically imply low accessibility across a nearby region. The relationship can depend on genomic context, cell mixture, assay design, and how regions are defined. Neither measurement alone establishes a change in gene expression or a causal mechanism.

Assumptions, Diagnostics, and Failure Modes

Before interpreting values, check how each measurement was produced. The binomial standard-error calculation for methylation assumes independent observations with a common methylation probability at the locus. In practice, dependence among reads, amplification, mapping uncertainty, conversion or classification error, and biological mixtures can undermine that idealization. The fraction remains a useful summary of observed calls, but its uncertainty should not be treated as fully described by the simple formula when those complications are substantial.

For accessibility comparisons, a meaningful numerical contrast requires compatible feature definitions, assay units, and processing. Raw counts from samples with different sequencing depths are not directly comparable as though the total measurement effort were equal. Even processed values can be misleading if pipelines or region boundaries differ, or if technical effects are aligned with the sample groups being compared.

• Check methylation coverage. Confirm that the denominator is positive and inspect coverage alongside the fraction. Low coverage means the estimate rests on few observations.
• Check the range and encoding. A sequencing-based fraction should be between 0 and 1. A value outside that range suggests a different encoding, a processing issue, or a data error; do not silently interpret it as a fraction.
• Check signal units. Determine whether accessibility values are raw counts, normalized counts, or transformed scores. The same number can mean different things under different definitions.
• Check comparability. Confirm that samples use compatible assays, processing, and region definitions before interpreting between-sample differences.
• Check the biological unit. Read-level observations are not the same as independent biological samples. Many reads from one sample do not substitute for measurements on multiple individuals.

An absent accessibility peak or a zero count also needs context: it may mean no fragments were observed under the assay and processing, not that a region is physically inaccessible in every cell. Similarly, a methylation fraction of zero means no methylated observations were recorded among the informative observations used for that entry; it does not prove that methylation is impossible in the underlying population.

Common Mistakes and Alternatives

Different summaries answer different questions. A methylation fraction uses methylated observations divided by total informative observations; it preserves the proportion-based interpretation. A methylated count alone gives the numerator but omits the denominator. Accessibility may be represented by raw counts when describing observed fragments, or by a suitable normalized signal when comparing compatible samples. Those are alternatives for summarizing assay evidence, not interchangeable scales. The right choice depends on the measurement process and the intended comparison.

The central limitation is that each entry is a summary of assay evidence at a defined feature. Methylation fractions are bounded but can be imprecise at low coverage; accessibility signals can be useful for comparisons but lack a universal unit. Neither table alone resolves cell-type mixture, technical bias, or causal direction. Interpret values alongside their denominators, units, feature definitions, and study design.