Genome-Wide Scope Is Not Uniform Coverage
Whole-genome sequencing (WGS) sequences DNA from across the genome rather than selecting only a predefined set of regions for capture. Its broad scope can reveal variation in coding and noncoding regions, including regions not selected by an exome design. That breadth is valuable, but it does not mean that every base is read equally often, or that every genomic structure can be reconstructed equally well.
Coverage, introduced in the earlier material on read depth, describes how many sequencing reads cover a genomic position. In an idealized library, reads would sample all positions evenly. In practice, DNA composition, library preparation, sequencing, and genomic structure can make some regions more or less represented. WGS therefore changes the range of regions available for analysis; it does not remove technical biases.
It is useful to separate three questions. First, was a region within the assay’s intended scope? Second, did reads provide adequate depth there? Third, does the sequence’s composition or structure make the resulting data difficult to interpret? A region can be within WGS scope but have low depth, and a region can have substantial depth while still being difficult to represent correctly.
An Idealized Model for Uncovered Positions
A simple probability model shows why even broadly sampled genomes may contain positions with no reads. Suppose the depth \(C\) at a particular base follows a Poisson distribution with mean \(\lambda\), where \(\lambda\) is the average depth. This approximation assumes reads are sampled independently and without position-specific bias, with a common average depth across the positions being considered. Real genomic data often depart from these assumptions, so the model is a baseline rather than a complete description.
For \(N\) positions with the same mean depth, let \(I_j=1\) if position \(j\) has zero reads and \(I_j=0\) otherwise. The number of uncovered positions is \(U=\sum_{j=1}^{N} I_j\). Since \(E(I_j)=P(C_j=0)=e^{-\lambda}\), linearity of expectation gives the expected number of uncovered positions:
This expectation does not require the zero-coverage indicators to be independent, but it does require the stated zero-coverage probability to apply to each position. Independence would be needed for a simple binomial distribution for \(U\), not for the expectation above.
Worked Example: Expected Uncovered Bases at Mean Depth Eight
Consider a hypothetical region of 100,000 bases. Assume, just for this calculation, that each base has Poisson depth with mean \(\lambda=8\), and that every position has the same depth distribution.
For one base, the probability of zero reads is \(e^{-8}=0.0003354626\) to the shown precision. Thus:
The model therefore predicts about 33.55 uncovered bases on average in a region of 100,000 bases under these idealized assumptions. This is an expectation, not a claim that a particular sequencing run must contain a fractional number of uncovered bases or exactly 34 such bases. Position-specific biases can make the actual count differ, sometimes substantially.
GC Content and Coverage Bias
GC content is the proportion of bases in a sequence that are guanine or cytosine. Regions with unusually high or low GC content can be underrepresented relative to the rest of a library. Depending on the protocol and sequencing platform, effects can arise during amplification, library preparation, or sequencing. The result is a systematic relationship between local GC content and observed depth, rather than just random variation around a common average.
A useful diagnostic is to compare depth across GC-content bins, ideally accounting for differences in the number of bases in each bin. A trough in depth at high or low GC content suggests that those sequences are less well represented. The size and direction of the effect are protocol-dependent; high-GC regions are not invariably the only regions affected.
Worked Example: Comparing Depth Across GC Bins
In a hypothetical WGS run, suppose the median depth across ordinary-composition regions is 30 reads per base. Two equally sized sets of genomic windows have median depths of 12 reads per base for a high-GC bin and 27 reads per base for a moderate-GC bin. Assume the windows were measured using the same depth calculation and are comparable in length.
Relative to the 30-read baseline, the high-GC depth ratio is \(12/30=0.40\), or 40% of baseline. The moderate-GC ratio is \(27/30=0.90\), or 90% of baseline. In absolute terms, the high-GC windows are 18 reads per base below the baseline, while the moderate-GC windows are 3 reads per base below it.
This pattern is consistent with a high-GC coverage deficit in this hypothetical run. It does not establish that every high-GC locus is undercovered, or identify a single cause. A proper check would inspect depth across many windows, confirm that the pattern is not driven by a few regions, and compare against relevant library and sequencing quality information.
GC-associated coverage differences matter because depth-based summaries may make a systematic deficit look like a set of isolated random gaps. They can also complicate comparisons across samples: if two libraries have different GC-depth profiles, the same genomic region may have different observed depth even when the underlying DNA copy number is the same. This tutorial’s focus is on recognizing that measurement limitation, not on making copy-number inferences from depth.
Repetitive and Structurally Complex Regions
A repetitive region contains sequence patterns that occur multiple times, either close together or at different locations in the genome. Repeats can be short and tandemly repeated, or part of longer duplicated blocks. If a read or read pair does not include enough distinctive sequence to identify a repeat copy, the data may not resolve which copy was sampled. Repeats can also be hard to assemble into a single representation, particularly when the repeat length exceeds the span of the available fragments.
Structurally complex regions include loci with features such as long insertions, deletions, inversions, duplicated segments, or combinations of these. A short-read library may sample sequence on either side of a structural change without providing a fragment that spans enough informative sequence to determine the full arrangement. Consequently, depth can be uneven, and the precise structure or copy count can remain uncertain even when nearby sequence has adequate coverage.
These limitations differ from a simple lack of sequencing. Low depth is a shortage of observations. Repetitive or complex sequence can instead create a resolution problem: the observations may not distinguish among multiple plausible representations. More reads can help when the issue is sampling, but increasing depth alone does not guarantee that a fragment spans a long repeat or clarifies a complex arrangement.
Worked Example: A Repeat Longer Than the Fragment Span
Suppose a hypothetical genomic locus contains a 600-base repeated segment between two distinctive flanking sequences. The library produces paired-end reads of 150 bases from fragments whose typical total span is about 350 bases. Assume that no fragment in this example spans the entire repeated segment together with informative sequence from both flanks.
The maximum typical fragment span, 350 bases, is shorter than the 600-base repeat. A pair can therefore provide local sequence evidence without directly connecting both distinctive flanks across the full repeated segment. If the data also show median depth of 28 reads per base in nearby unique sequence but 7 reads per base across the repeat, the repeat-to-flank depth ratio is \(7/28=0.25\). The repeat has one quarter of the nearby depth in this hypothetical example.
The low ratio flags a coverage deficit, while the fragment-span comparison explains why the structure may remain unresolved. Neither calculation establishes the true number of repeat copies: depth can vary for technical reasons, and the available fragments do not bridge the full structure. Longer fragments or other complementary measurements may be needed for that question.
Interpreting Coverage Diagnostics
Coverage summaries should be read as diagnostics of representation, not as proof that every well-covered base is error-free or that every low-depth base is biologically unusual. Useful summaries include the distribution of depth across the genome, the fraction of positions meeting a chosen depth threshold, depth by GC-content bin, and depth in annotated repetitive or structurally complex regions. These summaries are most informative when compared across samples prepared and sequenced in a similar way.
Summarize depth across broad genomic regions and note the typical level and spread. A genome-wide average alone can conceal local deficits.
Compare depth across GC bins and between unique, repetitive, and structurally complex regions. This helps distinguish broad sampling variation from systematic patterns.
Look for similar deficits across many windows or samples and review relevant library and sequencing summaries. A single unusual interval is weaker evidence of a general bias.
Describe whether the concern is low depth, nonuniform coverage, or insufficient resolution of a complex sequence. These are related but not interchangeable problems.
Diagnostics can also mislead. A high overall mean depth does not guarantee that difficult regions have adequate depth. Conversely, a few low-depth windows do not show that the entire library is poor. GC bins can differ in genomic composition, and repetitive annotations may not capture every difficult sequence. A useful interpretation therefore connects the metric to the region and the specific question.
How WGS Differs From Targeted Assays
Exome capture is designed to enrich selected coding regions, so its scope is restricted by the target design and capture performance. WGS avoids that particular target-selection boundary and samples broadly outside exons as well. However, it still has uneven depth and can have pronounced deficits in compositionally biased or structurally complex regions. A genotyping array measures a predetermined set of loci; its coverage is limited to those probes and their design, whereas WGS can sample far more sequence. Neither contrast means that one technology is universally superior: the useful assay depends on the study question, required regions, and relevant limitations.
| Feature | Whole-genome sequencing | Exome capture | Genotyping array |
|---|---|---|---|
| Intended scope | Broad sampling across the genome | Selected exonic targets and nearby designed regions | Predetermined assayed loci |
| Characteristic coverage concern | Nonuniform depth, including GC and repeat-associated deficits | Target-dependent capture and uneven enrichment | Variants outside the probe set are not directly measured |
| Complex sequence | May remain poorly covered or unresolved with short fragments | May be missed if not captured or difficult to sequence | May be absent from the assay or poorly measured by a probe |
The comparison is about scope and measurement behavior, not a claim that WGS resolves every region. The earlier material on exome capture explains why enrichment can create target-specific bias; the corresponding WGS lesson is that removing capture targets does not remove sequence-dependent limitations.
Common Mistakes and Limitations
- Equating genome-wide scope with complete coverage. WGS is designed to sample broadly, but some bases can receive little or no coverage.
- Treating average depth as a guarantee for every locus. A high genome-wide mean can coexist with GC-associated gaps or poorly represented repeats.
- Calling every depth deficit random. A consistent relationship with GC content or repeat annotation suggests a systematic pattern that a uniform Poisson model does not capture.
- Assuming more reads always solve structural complexity. Additional reads can reduce sampling gaps, but short fragments may still fail to span a long repeat or identify the full arrangement.
- Interpreting a diagnostic as a biological conclusion. A low-depth region is evidence about data representation, not by itself evidence of a biological deletion or other genomic change.
WGS is most useful when its breadth matches the question and when coverage limitations are assessed at the regions that matter. Its genome-wide design reduces dependence on a fixed capture target list, but it does not guarantee even representation, complete sequence recovery, or structural resolution. Recognizing which limitation is present is the first step toward choosing an appropriate follow-up or complementary measurement.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- Under the idealized Poisson model with mean depth 6, what is the probability that a specified base has zero reads?
- A region has 20 reads per base compared with a genome-wide baseline of 25. What is its depth ratio, and what can that ratio establish?
- Why can a high average depth fail to resolve a long repeated segment?
- Name one difference in scope between WGS and exome capture, and one coverage limitation they can both have.
Solutions. (1) \(P(C=0)=e^{-6}\approx0.002479\), or about 0.248% for one base under the model. (2) The ratio is \(20/25=0.80\), so the region has 80% of baseline depth; this indicates a relative depth deficit but does not establish its cause or biological meaning. (3) Average depth counts observations but does not ensure that any fragment spans enough distinctive sequence to determine the repeat’s structure. (4) WGS samples broadly across the genome, whereas exome capture enriches selected exonic targets; both can have uneven coverage, including deficits associated with sequence composition or difficult regions.