What a Sample Call Rate Measures
A genotype data set is often represented as a matrix: rows are variants and columns are samples. A sample call rate summarizes one column. It answers a narrow but useful question: for what proportion of the variants being assessed does this sample have an observed genotype call? A high rate means relatively few genotypes are missing in that sample on the assessed variant set. It does not, by itself, establish that the observed genotypes are accurate.
The earlier material on genotype calls and missing-value encoding distinguishes an observed genotype from an entry recorded as missing. For calculating a sample call rate, the genotype category is not the focus: an observed genotype counts as called, whether it is homozygous or heterozygous. A missing entry does not count as called.
The denominator \(M\) is the number of variants included in the assessment, not the number of nonmissing calls and not the number of samples. To compare samples fairly, calculate rates on the same defined variant set. That set should be stated clearly: for example, all variants in the current data matrix, or a specified set of variants retained for an analysis. The count is descriptive; no probability model is required to calculate it.
Calculation and Interpretation
The calculation is the count of observed genotypes divided by the number of eligible variant positions. The earlier tutorial on variant call rate uses the complementary orientation: it summarizes calls across samples for a variant. The sample call rate instead summarizes calls down the variants for one sample. These are related views of the same matrix, but they answer different questions.
Worked Example: Calculating One Sample’s Call Rate
Suppose a hypothetical data set assesses five variants in sample S1. Its entries, in variant order, are call, call, missing, call, call. Thus four of the five expected genotypes are observed. The sample call rate is $$ \mathrm{CR}_{\mathrm{S1}}=\frac{4}{5}=0.800=80.0\%. $$ The missingness proportion is \(1-0.800=0.200\), or \(20.0\%\). In context, S1 has observed genotype calls at 80.0% of these five assessed variants. This does not say whether the four observed calls are correct.
A sample call rate assigns equal weight to each included variant. It is not weighted by allele frequency, sequencing depth, or the importance of a locus. If variants have been excluded from the assessed set, they are not in the denominator. If different samples use different denominators, their percentages may not be directly comparable: a rate from 9 calls out of 10 is based on a different set size from a rate of 90 calls out of 100, and the assessed variants may differ as well.
Worked Example: Equal Rates, Different Missingness Patterns
Consider two hypothetical samples assessed at ten ordered, nearby variants. “Call” means an observed genotype and “Missing” means no genotype call. In each sample, three genotypes are missing.
| Variant position in order | Sample A | Sample B |
|---|---|---|
| 1 | Missing | Call |
| 2 | Call | Missing |
| 3 | Call | Missing |
| 4 | Call | Missing |
| 5 | Missing | Call |
| 6 | Call | Call |
| 7 | Call | Call |
| 8 | Call | Call |
| 9 | Missing | Call |
| 10 | Call | Call |
Sample A has \(10-3=7\) calls, so \(\mathrm{CR}_{\mathrm{A}}=7/10=70.0\%\). Sample B also has \(10-3=7\) calls, so \(\mathrm{CR}_{\mathrm{B}}=7/10=70.0\%\). The overall rates are identical, but their patterns differ: A’s missing calls are dispersed across the ordered positions, whereas B’s are concentrated at positions 2–4. If the positions are genomic and neighboring, B’s pattern is localized in this small example. The rate alone cannot represent this spatial arrangement.
Why the Missingness Pattern Adds Information
A single percentage is a compression of a whole column of genotype records. It preserves the total number of calls but discards their locations. As the example shows, two samples can have the same total and therefore the same call rate while differing in whether missing calls are spread out or concentrated in a region. Such a pattern may be relevant when deciding what to investigate, but the call-rate calculation alone does not identify its cause.
A localized pattern could reflect properties of particular variants or genomic regions, while broadly distributed missing calls could have a different explanation. These are possibilities, not conclusions implied by the percentage. A sample call rate is a summary of call availability, not a diagnostic test for a specific technical failure, a measure of genotype accuracy, or evidence that missingness is random. Interpretation requires the surrounding data and appropriate follow-up.
Worked Example: The Denominator Must Match the Question
Suppose a hypothetical assessment includes six eligible variants for sample S3. Five have observed calls and one is missing. Using all six eligible variants gives $$ \mathrm{CR}_{\mathrm{S3}}=\frac{5}{6}=0.8333=83.33\%. $$ Now suppose the question is explicitly changed to a four-variant subset, and S3 has calls at all four variants in that subset. The rate on that subset is \(4/4=100\%\). This is not a correction to the original rate; it is a rate for a different denominator and a different question.
If instead the one missing variant were simply dropped from the original six-variant denominator, S3 would appear to have \(5/5=100\%\) call rate. That result describes the reduced set, not the original assessment. A rate can rise when variants with missing entries are removed, so comparisons should specify the variant set and use it consistently across samples.
Calculation Checks and Common Pitfalls
State which \(M\) variants are included. Use the same set when comparing samples unless the different sets are intentional and clearly reported.
For the sample of interest, count entries recorded as observed genotypes; do not count missing-value codes as calls.
Compute calls divided by \(M\), then report the rate or percentage with its denominator and sample context.
If the distribution of missing calls matters, retain their variant positions when comparing samples; the total rate does not describe their arrangement.
A frequent error is to divide the number of missing genotypes by the number of observed genotypes. That is an odds-like ratio, not the call rate: with four calls and one missing entry, \(4/1=4\) is not a rate. The correct denominator is all five assessed variants, giving \(4/5=80.0\%\). Another error is to count each distinct genotype category or each sample as the denominator; neither matches the definition.
A second pitfall is to treat a high call rate as proof of high-quality genotypes. Call rate concerns presence versus absence only. A sample could have a call at nearly every variant and still have erroneous observed genotypes. Conversely, a lower rate flags more missing entries on the assessed set but does not explain why they are missing. Keep the measurement and its interpretation separate.
Assumptions, Comparisons, and Limits
The arithmetic relies on a clear, consistent definition of an eligible variant and a reliable distinction between observed and missing genotype entries. It also assumes that each included variant is counted once for the sample. If the data contain duplicate rows, inconsistent missing-value encodings, or different variant sets across samples, the rate can be computed mechanically but may not answer the intended comparison. Resolve or document those issues before interpreting the result.
Sample and variant call rates are complementary summaries. A variant call rate asks how often a particular variant is called across samples; a sample call rate asks how often a particular sample is called across variants. Neither summary locates all missing entries unless the underlying matrix is also inspected. When all samples use the same \(M\) variants, the unweighted mean of their sample call rates equals the overall fraction of called matrix entries: $$ \frac{1}{N}\sum_{j=1}^{N}\mathrm{CR}_j =\frac{1}{N}\sum_{j=1}^{N}\frac{\sum_{i=1}^{M}R_{ij}}{M} =\frac{\sum_{j=1}^{N}\sum_{i=1}^{M}R_{ij}}{MN}. $$ This identity follows by substituting the definition and collecting the total number of calls over the \(MN\) matrix entries. It does not imply that every sample has a rate close to the mean.
The rate is descriptive for the chosen data and variant set. It does not estimate a universal property of a person, and it does not reveal the mechanism generating missingness. It also does not prescribe a pass/fail cutoff; choosing such thresholds is a separate decision that must account for study aims and data characteristics. The practical value of the rate is that it provides a transparent first summary of sample-level call availability, while retaining the missing-call locations helps avoid losing important structure.
Check Your Understanding
Answer each question, then compare your reasoning with the solutions.
- A sample has calls at 18 of 20 assessed variants. What is its call rate and missingness proportion?
- Two samples each have 92 calls among 100 variants. What can be concluded from their call rates, and what cannot be concluded about the locations of missing calls?
- A sample has 8 calls and 2 missing entries among 10 variants. A student reports a call rate of \(8/2=400\%\). Identify the error and give the correct rate.
- Why should samples be compared on the same defined variant set?
- Does a 99% sample call rate establish that the observed genotype calls are accurate? Explain.
Solutions. (1) The rate is \(18/20=0.900=90.0\%\); the missingness proportion is \(2/20=0.100=10.0\%\). (2) Each has a call rate of \(92/100=92.0\%\), so they have the same fraction called on that set. The rates alone do not show whether missing calls are dispersed or concentrated at particular variants. (3) The student used the number missing rather than all assessed variants as the denominator. The correct rate is \(8/(8+2)=8/10=80.0\%\). (4) A changed variant set changes what is counted in the denominator and may change the percentage; consistent sets make the comparison interpretable. (5) No. Call rate records whether genotypes are present, not whether observed genotypes are correct.