Reading the Report as a Set of Connected Signals
A genotyping QC report is most useful when its summaries are read together. A low call rate, unusual allele frequency, or high heterozygosity can each have several explanations; the pattern across metrics helps decide whether to investigate a sample, a variant, a batch, or the underlying data definitions. A flag is a reason to check, not by itself proof that a genotype is wrong or that a sample must be removed.
The examples below use a hypothetical dataset of 600 samples and 100,000 autosomal variants. The report distinguishes sample-level summaries from variant-level summaries and gives results by laboratory batch. The numerical thresholds are illustrative: suitable thresholds depend on the assay, study design, sample composition, and the quality expected for the platform.
The denominator matters. A sample call rate should use the variants eligible for that sample after the report’s stated exclusions; a variant call rate should use the eligible samples. A report that uses different masks or filters across summaries may produce numbers that cannot be compared directly. Before interpreting a flag, check that the sample set, variant set, genotype filters, and allele coding are clear. For heterozygosity and allele balance, also check which chromosomes, genotype classes, and read-depth requirements were included.
A Hypothetical Report at a Glance
| Report section | Hypothetical summary | Initial signal |
|---|---|---|
| Sample call rate | Batch A mean 99.6%; Batch B mean 98.7%; S017 is 93.8% | Broadly lower completeness in B and a more extreme sample flag |
| Variant call rate | Median 99.4% in A and 96.2% in B | Many variants are less completely called in B |
| Allele frequency | At rsMock1, reported frequency is 0.18 in A and 0.29 in B | Check missingness, batch, coding, and sample composition |
| Heterozygosity | Across the report’s autosomal loci, median is 29.8%; S017 is 34.1% | S017 is high relative to this comparison set |
| Allele balance | For S017, 68% of reviewed heterozygous calls fall outside 0.35–0.65 | Check whether the pattern is widespread and technically supported |
These entries do not establish that Batch B caused the frequency difference at rsMock1, or that S017 is contaminated. They indicate where to look next. The comparison set matters too: allele frequencies and heterozygosity can vary with ancestry and sample composition, while call rate and balance can vary with laboratory procedures and sequencing quality.
Worked Example: Interpreting a Flagged Sample
S017 has a reported sample call rate of 93.8% across 100,000 eligible autosomal genotype entries. Its heterozygosity rate is 34.1%, compared with a report median of 29.8%, and its median allele balance among reviewed heterozygous calls is 0.43. Of 1,000 such calls, 680 have balance outside the report’s 0.35–0.65 range. What should be concluded?
Check the call-rate arithmetic. The report implies 93,800 called entries and 6,200 missing entries:
The sample is less complete than a sample near the report’s batch means. Its high heterozygosity and the large fraction of reviewed calls with atypical balance are additional signals, not independent proof of a particular cause. Together they justify checking sample-level read quality, depth and genotype-quality distributions, the allele-balance pattern by variant, and sample identity or contamination indicators. If the balance issue occurs mainly at a subset of loci, those variants deserve separate review. Do not infer contamination from the heterozygosity value alone.
Compare the Scale and Pattern of Each Signal
Start by asking whether the signal is isolated or shared. A single sample with low call rate calls for sample-focused checks; a set of samples in one batch with lower call rates suggests a batch-level problem or difference. A single low-call-rate variant may need variant-level review, while a broad shift across many variants in one batch may reflect a technical difference. These are priorities for investigation, not automatic diagnoses.
The distinction between a pooled summary and its parts can be made explicit. If batches have equal numbers of samples and each sample contributes the same number of eligible entries, the pooled call rate is the average of the batch call rates. Here the hypothetical batch means imply:
A pooled mean of 99.15% can conceal the fact that Batch B is less complete than Batch A. If batch sizes or denominators differ, a simple average of batch rates is not generally the pooled rate; the pooled rate must weight each rate by its number of eligible entries. Thus, inspect both the overall summary and the stratified summaries.
Worked Example: Distinguishing a Batch Signal From a Single Variant
Suppose the variant call-rate medians are 99.4% in Batch A and 96.2% in Batch B. At rsMock1, the report gives a call rate of 99.7% in A (299 of 300 samples) and 96.0% in B (288 of 300 samples). Is the concern limited to rsMock1?
Compare the variant-specific rates. At rsMock1, 1 sample is missing in A and 12 are missing in B:
The 3.67 percentage-point difference at this variant is consistent with the broader report pattern: the median variant call rate is also lower in B. The report therefore points to both a batch-wide completeness concern and an additional reason to inspect rsMock1. Follow up by checking missingness across other variants and samples in each batch, then review rsMock1’s genotype-quality and depth summaries by batch. The summary alone does not identify whether the cause is a laboratory issue, sample composition, or another technical difference.
Read Allele Frequency With Its Context
A reported allele frequency is conditional on the samples and called genotypes included in its calculation. A difference between batches can reflect a genuine difference in sample composition, a technical problem that changes which genotypes are observed, allele coding or strand issues, or chance variation. Interpret it alongside call rates and the report’s batch definitions. If missingness differs by batch, the frequency comparison may be based on different subsets of samples.
For rsMock1, the reported frequencies are 0.18 in A and 0.29 in B, a difference of 0.11 on the report’s frequency scale. This is a diagnostic discrepancy, not evidence on its own that either estimate is correct or that a batch effect caused it. Check that the same allele is being reported in both batches; confirm the variant’s reference and alternate allele coding; inspect batch-specific missingness and genotype-quality summaries; and check whether the batches differ in ancestry or other relevant sample composition. Only after those checks should the frequency difference be interpreted further.
Worked Example: Prioritizing an Allele-Frequency Flag
A QC report lists rsMock1 at frequency 0.18 in Batch A and 0.29 in Batch B. Batch B also has a lower median variant call rate than A, and rsMock1 itself has 12 missing calls in B versus 1 in A, with 300 eligible samples per batch. What is the appropriate next step?
First, describe the observed difference:
Next, do not treat 0.11 as a causal or statistical conclusion. The report indicates that the frequency summaries accompany unequal completeness, especially at this variant. Confirm the allele labels and coding, review which samples contributed called genotypes, and inspect genotype quality and depth at rsMock1 by batch. Then assess whether the batch sample sets differ in composition. If the frequency difference remains after these checks, it is a difference in the reported data that may merit further analysis; this QC summary alone does not determine its explanation.
Heterozygosity and Allele Balance Are Different Diagnostics
Heterozygosity summarizes how often called genotypes are heterozygous across a specified set of loci. A high or low value is a relative signal whose meaning depends on the comparison group, the loci included, ancestry, and relatedness. Technical problems can also affect it. Compare like with like where possible and check whether the value is unusual across the whole set or driven by a subset of loci.
Allele balance instead uses read support within heterozygous sequencing calls. For a diploid autosomal heterozygote without substantial technical bias, the two alleles are generally expected to have similar read support, so a balance near 0.5 is a useful reference—not a universal pass/fail rule. Low depth, mapping bias, allele-specific amplification, or incorrect calls can shift balance. A useful report presents the distribution and denominator, not only a single average. In particular, balance calculated on a small number of calls is less informative than a similar pattern across many well-supported calls.
For S017, 680 of 1,000 reviewed calls are outside the stated range, so the fraction is:
That is a substantial flag under this hypothetical report’s criterion. It should prompt inspection of read depth, genotype quality, and whether the imbalance clusters by variant or is broadly distributed. It does not by itself establish the presence or amount of contamination.
A Practical Order for Follow-Up
Confirm the sample and variant sets, filters, allele labels, and summary denominators. Check that batch comparisons use compatible definitions.
Determine whether the flag affects one sample, one variant, many samples, many variants, or a whole batch. Compare distributions as well as overall summaries.
Read call rate with missingness; read allele frequency with call rate, allele coding, and sample composition; read heterozygosity with its comparison set; and read allele balance with its call count and read quality.
Use the pattern to select sample-, variant-, or batch-level follow-up. Document the evidence and uncertainty before applying any exclusion or correction.
Common Mistakes
- Reading a high pooled call rate as proof that every batch is sound. An overall mean can hide a lower-quality subgroup. Inspect batch-stratified summaries.
- Treating an allele-frequency difference as proof of a technical error. First check allele coding, missingness, and sample composition; the report alone does not identify a cause.
- Calling an unusual heterozygosity value contamination. Heterozygosity is a screening signal with both biological and technical explanations. Interpret it with other metrics and follow-up checks.
- Applying one threshold without checking its definition or denominator. A percentage is interpretable only when the included samples, variants, and genotype filters are known.
- Removing every flagged sample or variant immediately. A flag prioritizes investigation. Decisions should follow review of the relevant evidence and the study’s prespecified QC plan.
Check Your Understanding
For each question, explain what the report supports and what it does not establish.
- A sample has a low call rate but ordinary heterozygosity and allele balance. Which level of follow-up is the most direct starting point, and why?
- Why can an overall call rate conceal a batch-specific completeness problem?
- A variant has different reported allele frequencies in two batches, and one batch has more missing calls at that variant. Name two checks to make before interpreting the frequency difference.
- What is the distinction between heterozygosity and allele balance?
- Why is a QC flag not, by itself, sufficient reason to conclude that a sample is contaminated or should be removed?