Genotyping Arrays, Exomes, and Whole-Genome Sequencing

Why Platform Choice Shapes a GWAS

A GWAS can test only genetic variation that its data measure adequately. A genotyping array measures a selected set of markers; exome sequencing reads targeted coding regions; whole-genome sequencing (WGS) reads across the genome. These are not interchangeable ways to produce the same data. Their differences affect which variants can be examined, how well common variation is represented, and how many participants a budget can support.

The relevant choice depends on the study question. A large study focused on common variants may value an affordable platform that supports many participants. A study interested in coding changes may need exome data, while a study that needs broad access to both coding and noncoding variation may consider WGS. No platform automatically guarantees complete or equally accurate measurement of all variants: assay design, laboratory performance, calling and quality-control procedures also matter.

Definition: A genotyping array measures a predefined set of genetic markers using probes. Exome sequencing reads targeted regions intended primarily to cover protein-coding exons. Whole-genome sequencing (WGS) reads DNA across the genome, including coding and noncoding regions. A variant is directly assayed when the platform measures that variant itself, rather than relying on it being represented by another marker.

Coverage and Variant Classes

Arrays are designed around selected markers, often common single-nucleotide variants (SNVs) chosen to support common-variant association analyses. The number and identity of markers depend on the array. A marker count is therefore not the same as a count of all variants in the genome, or even all common variants. A common variant absent from the array is not directly measured by that array. Arrays may also include selected variants for particular populations or purposes, and some designs include variants that are not common.

Exome sequencing focuses on coding regions. It can measure many coding SNVs and small insertions or deletions (indels), including variants not included on a particular array. Its coverage is not uniform: targeted regions can be missed or read at insufficient depth, and the assay generally does not provide comprehensive measurements of intronic, intergenic, or other noncoding variation. WGS has the broadest genomic scope of these three approaches. It can provide data on coding and noncoding SNVs and small indels, and may support analysis of larger variant classes when the sequencing and analysis pipeline are suitable. “Whole genome” describes the intended scope; it does not mean every base or variant class is measured perfectly.

For common-variant GWAS, the comparison is not simply “few variants versus many variants.” An array can be an efficient way to measure a large, selected marker set in many participants. Sequencing can directly observe a broader set of variants, but its usable coverage depends on read quality, depth, and calling. The variants available for analysis must also meet the study’s quality criteria. In particular, a platform’s advertised marker count or genomic scope should not be treated as the number of usable variants in the final analysis.

Platform	Typical scope	Common-variant implications	Important limitation
Genotyping array	Preselected markers across the genome	Can directly measure many selected common markers efficiently	Variants outside the marker set are not directly measured; marker usefulness can differ by ancestry
Exome sequencing	Targeted coding regions	Measures coding variants in adequately covered regions, including some not on an array	Does not comprehensively cover noncoding regions; capture and coverage are uneven
Whole-genome sequencing	Genome-wide sequence, coding and noncoding	Offers broad direct access to common variants throughout the genome when adequately called	Usually has higher per-sample cost and still depends on sequencing and analysis quality

Ancestry and Array Content

An array’s usefulness for a population depends partly on which markers were selected and on how genetic variation is distributed in that population. Array content has often been informed by particular reference samples and discovery efforts. A marker chosen because it tracks nearby variants well in one ancestry group may be a weaker proxy in another: allele frequencies and linkage disequilibrium patterns can differ. Thus, the same array may offer different representation of common variation across ancestry groups.

This limitation is not a claim that arrays are unusable outside the groups represented in their design. Rather, it means that marker content should be evaluated for the intended study population. Sequencing reduces dependence on a preselected set of array markers because it observes sequence variants directly in the regions adequately read. It does not remove all ancestry-related limitations: variant calling, reference resources, quality filters, and the frequency of a variant in the study population still affect what can be analyzed. Exome sequencing also remains restricted by its targeted coding scope.

A useful planning distinction is between direct measurement and broader representation. For an array, a target variant absent from the array is not directly measured. Whether other markers can provide information about it is a separate analysis question; it should not be silently counted as direct coverage. For sequencing, a target variant must be in the sequenced region and reliably called. In either case, “covered” should be defined in terms of the study’s actual target variants and quality requirements.

Worked Example: Comparing Direct Coverage in Two Groups

Suppose a hypothetical study defines 12 common target variants for a planned analysis in each of two ancestry groups. A candidate array directly assays 8 of the 12 targets that are common in Group A, and 6 of the 12 targets that are common in Group B. Compare direct coverage, without counting proxy markers.

For Group A, the direct-coverage proportion is

\frac{8}{12}=0.667

So the array directly measures 66.7% of the specified targets in Group A. For Group B,

\frac{6}{12}=0.500

It directly measures 50.0% of the specified targets in Group B. The difference is $0.667-0.500=0.167$, or 16.7 percentage points. These figures describe only the hypothetical target set and the array’s direct assays; they do not establish how well unassayed variants might be represented by other markers. Before selecting the array, investigators would check its actual marker content against the variants and population relevant to their study.

Cost, Sample Size, and the Whole-Study Budget

Per-sample platform costs can differ substantially, but a comparison is meaningful only if the costs include comparable items. Depending on the study, the relevant total may include laboratory processing, sample handling, sequencing or array reagents, data storage, and analysis. A quoted assay price alone is not necessarily the total cost of generating usable study data. Costs can also vary with scale, laboratory, and required data quality, so budget examples should be treated as planning assumptions rather than universal prices.

A basic budget calculation translates a fixed budget into the largest feasible whole number of samples. Let $B$ be the available budget and $c$ the positive total cost per sample, with no separate fixed cost. Since a sample cannot be fractional, the maximum sample count is the greatest integer not exceeding $B/c$. If there is a fixed cost $F$ in addition to a positive per-sample cost $c$, the same reasoning applies to the remaining budget $B-F$, provided $B\geq F$.

Result: With a positive per-sample cost $c$, budget $B$, and no separate fixed cost, the maximum whole-number sample size is $n_{\max}=\lfloor B/c\rfloor$. With fixed cost $F$ and $B\geq F$, it is $n_{\max}=\lfloor(B-F)/c\rfloor$. The floor function means rounding down to the greatest integer not exceeding its argument.

To justify the formula, a feasible integer sample size $n$ must satisfy $nc\leq B$, or $n\leq B/c$. The largest integer satisfying that inequality is $\lfloor B/c\rfloor$. With fixed cost, feasibility requires $F+nc\leq B$, so $n\leq(B-F)/c$; again, the largest feasible integer is the floor. Rounding to the nearest integer would be wrong whenever it rounds upward, because that could exceed the budget.

Worked Example: Comparing Sample Sizes Under One Budget

A hypothetical study has a total genotyping budget of $B=\$600{,}000$, with no separate fixed cost. Assume the stated per-sample costs include all costs being compared: $\$180$ for an array, $\$600$ for exome sequencing, and $\$1{,}200$ for WGS. Calculate the largest whole-number sample size for each option.

For the array,

n_{\max}=\left\lfloor\frac{\$600{,}000}{\$180\text{ per sample}}\right\rfloor =\lfloor 3333.33\ldots\rfloor=3333\text{ samples}.

For exome sequencing,

n_{\max}=\left\lfloor\frac{\$600{,}000}{\$600\text{ per sample}}\right\rfloor =\lfloor 1000\rfloor=1000\text{ samples}.

For WGS,

n_{\max}=\left\lfloor\frac{\$600{,}000}{\$1{,}200\text{ per sample}}\right\rfloor =\lfloor 500\rfloor=500\text{ samples}.

These calculations compare sample counts only under the stated hypothetical costs. They do not show that one platform is best: each provides different variant coverage. If costs were incomplete or quality requirements differed, the sample-size comparison would need to be recalculated using the relevant total costs.

Worked Example: Including a Fixed Study Cost

Suppose a hypothetical WGS project has a budget of $\$500{,}000$, a fixed setup and data-management cost of $\$35{,}000$, and a total variable cost of $\$1{,}150$ per sample. The remaining budget is $\$500{,}000-\$35{,}000=\$465{,}000$. Therefore,

n_{\max}=\left\lfloor\frac{\$500{,}000-\$35{,}000}{\$1{,}150\text{ per sample}}\right\rfloor =\lfloor 404.348\ldots\rfloor=404\text{ samples}.

Checking the boundary confirms the result: $404(\$1{,}150)+\$35{,}000=\$499{,}600$, within budget, whereas $405(\$1{,}150)+\$35{,}000=\$500{,}750$, which exceeds the budget. Thus, 404—not 405—is the maximum feasible whole-number sample size.

Choosing by the Variant Classes Needed

The platform should follow the scientific target rather than a general preference for more data. If the primary aim is a common-variant GWAS and the array’s content is suitable for the target population and variants, an array may allow a larger sample for a given budget. If the main interest is coding variation, exome sequencing offers direct access to many coding variants, subject to capture and call quality. If the study needs broad access to noncoding as well as coding variants, WGS offers wider sequence scope, at a potentially higher cost.

Worked Example: Matching Platform to Variant Class

A hypothetical team wants to evaluate 10 prespecified variants: 4 coding SNVs, 2 coding indels, and 4 noncoding SNVs. Assume the candidate array contains only 3 of these targets, all coding SNVs. Assume the exome assay adequately covers all 6 coding targets but none of the 4 noncoding targets. Assume the WGS assay adequately covers and calls all 10 targets. Under these explicit assumptions, compare the directly available targets.

The array directly assays $3$ of $10$ targets, so its direct coverage is $3/10=0.300$, or 30%. Exome sequencing directly covers $6$ of $10$, giving $6/10=0.600$, or 60%; its six targets are the four coding SNVs and two coding indels. WGS directly covers $10/10=1.00$, or 100%, under the stated coverage and calling assumptions. Thus, if all 10 targets are essential, WGS is the only one of these hypothetical options that directly covers the full set. If the question were restricted to the six coding targets, exome sequencing would cover all six in this example. These proportions follow from the assumed target content and do not imply that every real assay performs identically.

Common Mistakes and Practical Checks

Equating marker count with genome-wide coverage. An array with many markers still measures a predefined subset. Check whether the variants relevant to the study are actually present.
Assuming an array works equally well across ancestries. Marker selection and the relationship between markers and nearby variation can differ across populations. Evaluate the intended array for the study population rather than treating its marker list as universally representative.
Calling an exome a small whole genome. Exome sequencing targets coding regions. It is not a substitute for broad noncoding coverage when noncoding variants are central to the question.
Assuming WGS means perfect measurement. Sequencing scope does not guarantee adequate read depth, successful calling, or analysis of every variant class. Specify the required coverage and quality criteria.
Rounding a budget quotient to the nearest sample. The sample count must be an integer and must not exceed the budget. Use the floor, and check the cost of $n_{\max}$ and $n_{\max}+1$ when useful.

Key takeaway: Compare platforms using the study’s target variants, intended population, required variant classes, and complete per-sample costs. Arrays offer selected markers, exomes target coding regions, and WGS has broader sequence scope; usable coverage and feasible sample size depend on the specific design.

Check Your Understanding

Answer each question using the definitions and assumptions in this tutorial.

Why is an array’s number of markers not the same as its number of directly measured common variants in every study population?
A study has a budget of $\$72{,}000$, a fixed cost of $\$6{,}000$, and a per-sample cost of $\$250$. What is the maximum whole-number sample size?
Which platform has the broadest intended scope for coding and noncoding sequence variation, and what does that scope not guarantee?
A team is interested only in coding variants. Name one reason exome sequencing may suit that target and one coverage limitation it should check.
Why should unassayed array variants not be described as directly measured, even if other array markers may contain information about them?