Exome Capture and Its Sources of Bias

Why Exome Capture Is Needed

An exome assay sequences selected regions of the genome, usually regions intended to cover protein-coding exons and sometimes nearby sequence. Sequencing an entire prepared library without enrichment would spend many reads on regions outside that intended target. Exome capture uses complementary DNA probes to enrich library fragments from selected genomic regions before sequencing. Enrichment makes it possible to direct more sequencing effort toward those regions, but it does not guarantee that every targeted base will be represented equally.

The distinction between a target and an observed read matters. A target is a region specified by the assay design; a read is a sequenced fragment from the enriched library. A targeted base can receive few or no reads, while fragments from outside the target can also be sequenced. Capture performance therefore affects which regions are well measured, not just the total number of reads produced.

How Hybridization Capture Works

The sequencing library consists of genomic fragments prepared for sequencing, as introduced in the earlier material on sequencing libraries and short reads. In a typical solution-based capture, the library is mixed with short, single-stranded DNA or RNA probes designed to be complementary to selected reference-genome regions. Under suitable hybridization conditions, fragments that match a probe bind to it. The probes are often biotinylated, allowing probe–fragment complexes to be collected with streptavidin-coated magnetic beads. Washing removes many unbound fragments; the enriched fragments are then recovered, amplified as needed, and sequenced.

Probe sets may cover target regions with overlapping probes, a design known as tiling. The probes do not have to begin and end at every exon boundary: a design can include padding into flanking intronic sequence, or can target only specified portions of an exon. Thus, “the exome” is an assay-defined target, not a single universal list of bases. Different kits can target different transcripts, exon boundaries, and flanking regions.

Capture enriches fragments, rather than directly selecting isolated bases. A fragment can overlap a target and extend beyond it. Consequently, some reads may include flanking sequence, and whether such a read counts as on-target depends on the target intervals and the analysis definition. Comparisons between assays are meaningful only when the target definition and read-counting rules are clear.

Why Capture Is Uneven

The amount of sequencing depth at a target region depends partly on how efficiently its fragments hybridize and are retained. Probe binding is affected by sequence composition, probe length and design, hybridization conditions, and the availability of fragments containing the target. Regions with extreme GC content can be difficult to capture or amplify consistently. Repetitive or highly similar sequence can also make probe design less specific. Differences in probe tiling, gaps between probes, and fragment length contribute to differences across regions.

These effects vary by locus. A single sample can have strong depth at one exon and weak depth at another, even though both are included in the assay design. More total reads may improve depth overall, but cannot be assumed to correct every local deficit: if a region captures inefficiently, it can remain underrepresented relative to other regions as sequencing effort increases.

A useful way to describe the effect is to let \(q_j\) be the probability that a sequenced read covers target base \(j\), under a fixed library and sequencing setup. If \(N\) reads are sampled, expected depth at that base is \(Nq_j\). Capture efficiency changes the probabilities \(q_j\) across positions; uneven values produce uneven expected depth. This is a description of read allocation, not a claim that every base has a fixed biological property independent of assay conditions.

For a simple sampling approximation, suppose reads covering a base arrive independently and its depth follows a Poisson distribution with mean \(\lambda\). Then the probability of zero reads is \(P(X=0)=e^{-\lambda}\), so the probability of at least one read is \(1-e^{-\lambda}\). The calculation assumes independent sampling and a fixed mean depth. It can help illustrate why low-depth bases are more likely to be uncovered, but real capture libraries can depart from this model because fragments are not independent and amplification can create duplicates.

Off-Target Reads and Exon Boundaries

Off-target reads arise because hybridization is not a perfect all-or-none selection. Fragments can bind nonspecifically, partially match a probe, or be retained despite not overlapping the intended target. Some off-target reads also result from fragment overlap with a target: a fragment selected because part of it binds may extend into nearby sequence. A nonzero off-target rate is therefore expected in many capture assays. Its practical cost is that a portion of sequencing effort is spent outside the intended intervals.

Exon boundaries deserve particular attention. If probes stop at a defined exon edge, fragments near the edge may have less opportunity to hybridize than fragments in a well-tiled interior. Coverage can also depend on whether the design includes flanking sequence and on whether library fragments span the boundary. As a result, bases near an exon’s start or end may have lower depth than bases farther inside the exon. This is not inevitable: padding and tiling can improve boundary representation, and some boundaries may have good coverage. The target coordinates and the observed coverage profile must be checked rather than inferred from the word “exome.”

A boundary-related coverage gap is a measurement limitation. If a base is absent or weakly covered, the data provide less evidence about that base; lack of observed reads is not evidence that the base has no variant. The same caution applies to any poorly captured region. A report of average depth across an assay can mask such local gaps.

Diagnostics, Alternatives, and Limitations

Capture quality should be assessed with several summaries rather than a single number. The on-target rate measures read allocation to the target, while mean depth and the fraction of bases above a stated depth threshold describe different aspects of coverage. A per-base or per-exon coverage profile can reveal gaps hidden by averages. It is also useful to examine coverage by GC content and by position within exons, particularly near boundaries. Comparisons across samples can show whether a low-coverage region recurs consistently or appears mainly in particular libraries or batches.

When a region has unexpectedly weak coverage, check the target definition first: the region may not have been included, or only part of it may have been targeted. Then inspect the local depth profile, the surrounding exon, and relevant design information, including probe tiling and flanking sequence. A recurring deficit in the same region is consistent with a systematic design or sequence-context limitation; a deficit isolated to one sample can also reflect that sample’s library or capture performance. Neither pattern should be diagnosed from mean depth alone.

Hybridization capture is one way to enrich selected regions. A main alternative for targeted sequencing is amplicon enrichment, in which selected regions are amplified with primer pairs. Amplicon methods can be efficient for a fixed, relatively small set of targets, but their coverage depends on successful primer amplification and primer placement. Hybrid capture can target many dispersed regions through probe sets and can include fragments that extend beyond a precise primer-defined amplicon, but it has its own unevenness and off-target costs. Neither method guarantees uniform coverage; the appropriate comparison is performance on the regions and samples relevant to the study.

Capture results are limited by the assay’s target list and the sample-specific library. A well-performing assay can still have poorly represented regions, and a high on-target rate does not guarantee adequate depth at every target base. Moreover, depth is only a measure of read representation; it does not, by itself, establish that the reads are accurate or that a variant call is correct. Those are separate quality questions.

Common Pitfall

Wrong reasoning: “The assay has a high average depth, so every exon boundary is adequately measured.” The average can be high because many bases have abundant reads while a smaller set has little or none.

Correction: Evaluate coverage across the target at base or exon level, state the depth threshold, and report the proportion of target bases that meet it. Inspect boundary zones separately when those bases matter to the study. Average depth and on-target rate are useful summaries, but neither substitutes for local coverage assessment.