Sample Call Rate and Missingness Diagnostics

Why Summarize Missingness by Sample?

A genotype data set can contain missing entries for many reasons: a sample may have produced weak signals, a genotype may not have met a calling rule, or data may have been lost during processing. Looking at one entry does not reveal whether missingness is isolated or widespread. A sample-level call rate condenses the genotype records for one sample into a summary: the fraction of eligible genotypes that were successfully called.

This summary is useful for locating samples that warrant closer review. It is not a direct measure of whether the recorded genotypes are correct. A sample can have a high call rate and still contain incorrect calls, while a low call rate signals missing data rather than identifying its cause. The aim is to use the calculation and the distribution across samples as diagnostics, not to declare quality from one number alone.

Defining and Calculating the Summary

First decide which genotype entries count as eligible. For a simple data set of diploid genotypes at a defined set of variants, each sample may have one eligible entry per variant. The denominator is then the number of variants in that set. The missing count is the number of those entries represented as missing according to the data format, such as a VCF missing genotype token. A genotype coded \(0/0\) is a called homozygous genotype, not a missing value; the code must be interpreted according to the file format rather than by whether it contains a zero.

The set of variants and the rules for eligibility should be stated and applied consistently. If some entries are not applicable or are outside the intended analysis set, they should not silently be counted as missing or as called. Different denominators can make sample rates less directly comparable, so a reported rate should be understood together with its denominator and the rule that defined it.

Call rate and missingness are complementary: increasing one by a given amount decreases the other by the same amount. Reporting both is usually redundant, but either can be convenient. A call rate of \(0.975\), for example, means 97.5% of the eligible genotype entries were called and 2.5% were missing.

Summarizing a Distribution Across Samples

A table of individual rates shows exactly which samples contribute to a pattern, while summaries and plots help describe the pattern across the full set. Common summaries include the mean, median, range, and selected quantiles. A histogram or dot plot can reveal whether most samples have similar rates, whether there is a low-rate tail, or whether the distribution has several groups.

The mean call rate is the average of the sample-specific call rates. It gives each sample equal weight. If every sample has the same denominator, this mean also equals the pooled fraction of called entries across all samples. If denominators differ, the pooled fraction gives more weight to samples with more eligible entries and generally differs from the unweighted mean. Neither calculation is inherently wrong; they answer different questions and should be labelled clearly.

Why the Denominator and the Pattern Matter

The denominator defines the scope of the claim. “A 98% call rate” is incomplete unless it is clear which entries were eligible. If a calculation uses one variant set and a later calculation uses a different set, the sample rate can change even when the sample’s raw genotype record has not changed. Comparisons are most interpretable when the same eligibility rules and variant set are used for all samples being compared.

Rates also hide the locations of missing entries. Two samples can have the same missingness proportion while missing calls at entirely different variants. A sample summary therefore does not describe the distribution of missingness across variants, and it cannot show whether missingness is concentrated in a region or associated with a particular data-processing pattern. Those require additional diagnostics.

A pooled rate can also conceal a low-rate sample. For instance, if most samples have nearly all entries called but one has many missing entries, an overall average may still appear high. Inspecting individual rates and a distribution plot together reduces this risk. Conversely, an isolated low value should not be interpreted without checking whether its denominator is unusually small or whether its entries were defined differently.

A Short Derivation: Why the Two Rates Are Complements

Among \(N_i\) eligible entries for sample \(i\), exactly \(M_i\) are missing. Therefore \(N_i-M_i\) are called. Dividing the missing count and the called count by the same positive denominator gives

This identity depends on the entries being classified consistently as either called or missing within the eligible set. It also shows why the two summaries contain the same numerical information: one can be recovered exactly from the other.

Common Mistakes and Practical Interpretation

• Treating one cutoff as universal. A low call rate is a reason to review a sample, not proof that it must be excluded. What counts as unusually low depends on the data set and the analysis purpose.
• Confusing a called genotype with a missing one. A genotype such as \(0/0\) is a call. Missingness is determined by the file’s missing-value representation, not by a genotype containing the digit zero.
• Reporting a rate without its scope. State the eligible variant set or other denominator definition. Otherwise, two rates may not be comparable.
• Using an overall average as the only diagnostic. A high pooled or mean rate can obscure a low-rate sample. Inspect sample-specific values and their distribution.
• Equating call rate with accuracy. Call rate measures how many eligible entries are called, not whether those calls are correct. A high value does not rule out genotype errors.
• Interpreting the flag as a cause. The proportion missing identifies a pattern, not its mechanism. Follow-up should consider the data and processing context rather than attributing missingness to a specific cause from the rate alone.