Sample-Level Diagnostics For Structural-Variant Cohorts

Why Diagnose Samples Before Association Analysis?

Variant-level filters assess whether individual structural-variant calls are credible. They do not establish that every sample was measured equally well. A sample with low callability may have many missing calls, while a sample processed in a different batch may show an unusual number of events or an atypical distribution of copy-number states. If these patterns align with phenotype or study group, downstream comparisons can reflect technical differences as well as biological variation.

Sample-level diagnostics are summaries for detecting patterns that merit investigation. They are not, by themselves, proof that a sample is erroneous or that an observed event is biological. The comparison must use an appropriate denominator, and interpretation should account for assay design, callable regions, ploidy, and batch composition.

Call Rate: The Denominator Matters

For a sample, call rate is the fraction of eligible sample–locus observations for which a call was successfully made. A locus belongs in the denominator only if it was intended to be assessed and was eligible for calling in that sample. Thus, the denominator should reflect the assay’s targets and the sample’s callable opportunities, rather than an arbitrary count of all possible genomic positions.

The definition of “locus” must match the analysis. It might refer to targeted intervals, a harmonized set of candidate regions, or another prespecified unit. If two samples have different target sets, dividing both by the same nominal number of loci can make their rates incomparable. Likewise, a locus that is not callable because of known assay limitations should not be treated as a failed call without stating that convention.

Call rate is a completeness measure, not an accuracy measure. A sample can have a high call rate and still have inaccurate calls. A low rate can reflect poor DNA quality, coverage, mapping difficulty, or an overly broad denominator. Inspect call rate alongside the technical records and the variant-level quality criteria already applied.

Event Counts and Callable Opportunity

A total event count is the number of retained structural-variant events attributed to a sample under a specified event representation and filtering procedure. State exactly what is counted: for example, whether the count includes deletions and duplications, how multiallelic or overlapping calls are represented, and whether only calls passing the chosen variant-level filters are included. The count is useful for identifying unusually high or low values, but it is not automatically a measure of biological burden.

Observed event counts can differ because samples have different amounts of callable sequence or different detection sensitivity. When possible, compare samples assessed over the same regions and with comparable callability. If callable sequence differs, report it and consider an event rate per callable unit, such as events per callable megabase, only when the callable bases are defined consistently and the rate is scientifically interpretable.

An event-rate adjustment does not solve every comparability problem. If event detection sensitivity differs between samples, or if callable regions differ in their propensity to contain events, a simple rate may still be misleading. It is often preferable to restrict the comparison to a common callable region and inspect call rate and event count together.

Copy-Number Distributions Across Callable Regions

A sample’s copy-number distribution summarizes how often each copy-number state is observed across the regions assessed for that sample. Earlier material introduced copy-number states and dosage coding; here the purpose is quality review, not modeling an association. Summaries should be restricted to regions where a copy-number call was eligible and callable, and should respect the expected baseline ploidy. For example, interpreting a state relative to diploidy requires care for sex chromosomes or other regions where the expected baseline differs.

For sample \(i\), let \(n_{ic}\) be the number of callable regions assigned copy-number state \(c\), and let \(N_i\) be the total number of callable regions summarized. The within-sample proportion in state \(c\) is:

These proportions make samples with different numbers of callable regions easier to compare, provided the regions represent comparable opportunities. A copy-number histogram can reveal a sample with an unexpected concentration of apparent losses or gains, a distorted baseline peak, or an unusual fraction of unresolved states. Such a pattern can result from technical issues, but can also reflect real sample composition or region selection. Review the underlying calls and denominators before interpreting it.

Batch-Stratified Summaries

A cohort-wide summary can conceal technical heterogeneity. At minimum, inspect call rate, event count, and copy-number summaries by sequencing or calling batch, alongside the corresponding sample-level distributions. A batch summary may use a median and range or interquartile range for continuous diagnostics, and proportions for copy-number categories. Report the number of samples and the denominator behind each summary; otherwise, a difference can be difficult to assess.

Stratification helps distinguish an isolated sample outlier from a batch-wide shift. If many samples in one batch have lower call rates or higher event counts, a pipeline, reagent, sequencing, or target-design difference becomes a plausible explanation to investigate. If only one sample is unusual, sample-specific quality or biology may be more relevant. Neither pattern proves a cause.

Batch and phenotype can be associated by design or chance. If nearly all cases are in one batch and nearly all controls in another, a technical difference may be entangled with the group comparison. The diagnostic summaries can expose this risk, but they do not correct it. Preserve batch labels, document the imbalance, and resolve comparability before drawing association conclusions.

A Practical Diagnostic Sequence

Common Mistakes and Interpretation

• Using incomparable denominators. A call rate based on all nominal targets may unfairly penalize samples for regions that were not eligible or callable. Define the denominator and compare like with like.
• Treating a high event count as proof of poor quality. A high count can be biological, technical, or both. Check callability, filters, and batch patterns before deciding.
• Comparing raw copy-number counts across unequal numbers of regions. Use proportions among callable regions and confirm that the region sets are comparable.
• Assuming batch differences are harmless because calls passed variant-level filters. Variant-level filters do not guarantee equivalent sample-level sensitivity. Review sample and batch summaries as well.
• Automatically removing every flagged sample. A descriptive fence or unusual histogram is a prompt for investigation, not a universal exclusion threshold. Use prespecified criteria and retain an audit trail.