Variant Call Rate

Why Measure Call Rate at a Variant?

A genotype data set may contain one row for each sample and one column for each variant. At a particular variant, some samples have a recorded genotype and others have no usable genotype call. The variant call rate summarizes the completeness of that variant’s genotypes across the samples being considered. It is a simple proportion, but interpreting it correctly depends on defining both what counts as a call and which samples belong in the denominator.

For example, a genotype recorded as reference homozygous is a call just as a heterozygous or alternate homozygous genotype is. A missing-value code, or a genotype that has been designated a no-call under the analysis rules, is not a call. The rate describes whether a genotype is available; it does not establish whether that genotype is correct.

Definition: For variant $v$, let $S_v$ be the set of samples eligible to have a genotype recorded at that variant, and let $N_v=|S_v|$. For sample $i\in S_v$, define $I_{iv}=1$ if the genotype at $v$ is called and $I_{iv}=0$ if it is missing or designated a no-call. The variant call rate is $$ \text{Call rate}_v=\frac{\sum_{i\in S_v} I_{iv}}{N_v}. $$ It may be reported as a proportion or as a percentage.

The definition applies to a sample-by-variant genotype cell: each eligible sample contributes at most one indicator for the variant. For a diploid genotype, the cell represents the sample’s genotype at that locus, not two separate observations because there are two allele copies. The genotype notation and missing-value conventions established when reading genotype data determine which cells are called.

The denominator is the number of eligible samples, not automatically the number of rows in a file. Usually, all samples retained for the analysis are eligible for an autosomal variant, so the denominator is the number of those samples. If a sample was excluded from the analysis or the variant was not applicable to that sample under the data protocol, that decision should be handled before calculating the rate. The denominator must follow a stated, consistent rule; it should not be changed simply because a sample has a missing genotype at this particular variant.

Calculation and Interpretation

To calculate the rate, count the eligible samples with a called genotype, then divide by the number of eligible samples. Multiplying by 100 expresses the result as a percentage. Equivalently, if $M_v$ is the number of eligible samples with a missing or no-call genotype, then the number called is $N_v-M_v$, and

\text{Call rate}_v=\frac{N_v-M_v}{N_v} =1-\frac{M_v}{N_v}.

Thus the missing-call proportion and call rate are complements, provided they use the same eligible sample set and the same rule for classifying cells. A call rate of 0.90 means that 90% of the eligible samples have a called genotype at that variant in the data being summarized. It does not mean that 90% of those genotypes are correct, nor does it describe the proportion of allele copies that are accurate.

Worked Example: Counting Calls at One Variant

Suppose a hypothetical data set contains 8 eligible samples at variant $v_1$. The genotype cells are: AA, AG, missing, GG, AG, missing, AA, and AG. Assume the three genotype labels are recorded calls and “missing” is the only no-call code.

There are 6 called genotypes and 2 missing genotypes. Each of the 8 samples is eligible, so $N_{v_1}=8$. Therefore,

\text{Call rate}_{v_1} =\frac{6}{8} =0.75 =75\%.

The missing-call proportion is $2/8=0.25$, or 25%, and $1-0.25=0.75$. In this hypothetical data set, three quarters of the eligible samples have a recorded genotype at $v_1$. This calculation does not assess whether the six recorded genotypes match the samples’ true genotypes.

The call rate is a summary of a particular variant across a specified set of samples. It can be calculated even if the called genotypes differ in type; the calculation only distinguishes called from not called. Two variants can have the same call rate but have missing calls in different samples. The rate alone does not retain that pattern information.

Worked Example: Same Call Rate, Different Missing Samples

Consider 12 hypothetical eligible samples and two variants, $v_2$ and $v_3$. At $v_2$, samples 1–9 have calls and samples 10–12 are missing. At $v_3$, samples 1–3 are missing and samples 4–12 have calls. All other genotype cells are recorded calls.

For $v_2$, the number of calls is 9, so

\text{Call rate}_{v_2}=\frac{9}{12}=0.75=75\%.

For $v_3$, the number of calls is also 9, so

\text{Call rate}_{v_3}=\frac{9}{12}=0.75=75\%.

The rates are equal, but the missingness patterns are not: $v_2$ is missing in samples 10–12, whereas $v_3$ is missing in samples 1–3. The call-rate calculation is correct for each variant, but the two percentages do not show which samples are affected. This distinction matters when a summary is used to decide what further checks are needed.

Denominators and Comparisons

Call rates can be compared only with their denominators and eligibility rules in view. The same number of calls can produce different rates when the eligible sample counts differ. Conversely, the same percentage can arise from different counts. Reporting the numerator and denominator alongside the rate makes its scale transparent and helps prevent accidental comparisons of unlike quantities.

Suppose two hypothetical variants have rates calculated in different eligible sample sets because a documented sample exclusion was made before the calculation. The rates remain valid proportions for those sets, but their comparison is not automatically a comparison on a common population of samples. If samples are excluded inconsistently, the difference in rates may reflect the changing denominator as well as variation in call availability. State the sample set or count used for each rate.

Worked Example: Different Sample Counts

In a hypothetical analysis, variant $v_4$ has 45 calls among 50 eligible samples. Variant $v_5$ has 54 calls among 60 eligible samples. Assume each denominator was defined before counting missing genotype cells and that all listed samples are eligible at the corresponding variant.

For $v_4$,

\text{Call rate}_{v_4} =\frac{45}{50} =0.90 =90\%.

For $v_5$,

\text{Call rate}_{v_5} =\frac{54}{60} =0.90 =90\%.

Both variants have a 90% call rate, though the first has 45 called genotypes and the second has 54. Each has a missing-call proportion of 10%: $5/50=0.10$ for $v_4$, and $6/60=0.10$ for $v_5$. The equal rates describe equal observed fractions in their respective eligible sets. They do not imply that the same number of samples was called or that the missing calls occurred in comparable samples.

Justification and Practical Checks

The complement relationship follows directly from counting the eligible genotype cells. If every eligible cell is classified exactly once as either called or missing/no-call, then the number called plus the number missing is the total number eligible:

$$ C_v+M_v=N_v. $$

Dividing both sides by $N_v$, where $N_v>0$, gives

\frac{C_v}{N_v}+\frac{M_v}{N_v}=1.

Since $C_v/N_v$ is the call rate, it follows that the missing-call proportion is $1-\text{Call rate}_v$. This also explains why the formula requires a positive denominator: if no samples are eligible, the rate is undefined, not zero.

Before reporting a rate, check the data representation and the counting rule. A missing genotype might be stored as a dedicated token, a blank cell, or a special code. A code used to represent a genotype must not be mistaken for missingness merely because it looks unusual, and a missing-value code must not be counted as a genotype. If low-quality calls have already been recoded as no-calls, they contribute zero under this definition. If they have not, the rate based on recorded calls will count them as called, even though their reliability may be questionable.

The variant call rate is distinct from a per-sample summary: it fixes one variant and counts across samples. A per-sample call-rate calculation instead fixes one sample and counts across variants. These summaries answer different questions and have different denominators. This tutorial’s calculation is also distinct from genotype quality or validation accuracy: a genotype may be present but wrong, or missing despite evidence that could potentially support a call.

Common Mistakes and Limitations

Using the wrong denominator. Dividing by the number of called genotypes returns 1, not the call rate. Divide by all eligible samples, including those with missing calls.
Dropping missing samples before calculating. Removing no-call cells from both numerator and denominator hides the missingness. Keep eligible samples in the denominator even when their genotype is missing.
Counting alleles instead of genotype cells. At a diploid variant, a sample contributes one genotype cell to this rate, not two separate allele observations.
Treating the rate as accuracy. A high rate means calls are available for many eligible samples. It does not demonstrate that the calls are correct.
Comparing percentages without denominators or eligibility rules. Report the number called, the number eligible, and the rate; clarify when sample sets differ.

A call rate is intentionally a compact completeness summary. It does not show the identities of samples with missing calls, diagnose why calls are absent, or establish whether missingness is related to sample characteristics or study groups. Those questions require examining patterns beyond this single proportion. Likewise, the rate alone does not determine a universal accept-or-reject decision; the interpretation depends on the analysis context and the stated quality-control plan.

Key takeaway: For one variant, count called genotype cells across the eligible samples and divide by the number eligible. State the numerator and denominator, use a consistent missing-call rule, and interpret the result as genotype availability—not genotype accuracy.

Check Your Understanding

Work each item by identifying the eligible denominator before calculating. Solutions follow the questions.

A hypothetical variant has 18 called genotypes and 2 missing genotypes among 20 eligible samples. What is its call rate and missing-call proportion?
There are 30 eligible samples at a variant. A file uses “.” for missing and records 24 genotype labels other than “.”. What is the call rate? Explain what would go wrong if the six “.” entries were removed from the denominator first.
Two variants each have a call rate of 80%. Must they have missing genotypes in the same samples? Explain.
A variant has 0 called genotypes among 0 eligible samples. Is its call rate 0%? Explain.
In one sentence, explain why a 98% call rate does not establish 98% genotype accuracy.

Solutions. (1) The rate is $18/20=0.90$, or 90%; the missing-call proportion is $2/20=0.10$, or 10%. (2) The denominator is 30 eligible samples, so the call rate is $24/30=0.80$, or 80%. Removing the missing entries first would give $24/24=1$, incorrectly concealing that 6 eligible samples lack calls. (3) No. Equal rates specify equal fractions called, not which samples are missing; the missing sets can differ. (4) No. With zero eligible samples the denominator is zero, so the rate is undefined. (5) Call rate measures the proportion with a recorded genotype, whereas accuracy requires knowing whether recorded genotypes are correct.

What You'll Learn