Quality Control for Chromatin Sequencing Libraries

Why Library Quality Control Is Multidimensional

A chromatin sequencing library can produce many reads yet provide little useful information. Reads may fail to align, map ambiguously, repeat already observed fragments, or show little enrichment for the biological signal of interest. Quality control (QC) asks whether the library and its sequencing yield are adequate for the intended assay—not whether a sample is biologically “good” or whether a result is true.

The relevant evidence depends on the assay, reference genome, read layout, sequencing depth, and library preparation. A narrow enrichment pattern and a broad enrichment pattern need not have the same expected fragment profile. Therefore, QC should combine complementary metrics and compare samples with appropriate assay-specific expectations, rather than rely on a single pass/fail cutoff.

Before interpreting the metrics, state the counting rules: whether the denominator is read pairs or individual reads, whether multimapping alignments are retained, how duplicate fragments are defined, and whether enrichment uses duplicate-marked or nonduplicate fragments. These choices can change the numerical values, so comparisons are meaningful only when the rules are consistent.

Mapping, Duplication, and Complexity

Mapping summarizes how much sequenced material can be placed on the chosen reference. One useful quantity is the overall mapping rate, the number of fragments with an accepted alignment divided by all sequenced fragments. The unique mapping rate instead uses uniquely mapped fragments in the numerator. Reporting both helps distinguish poor overall alignment from a high rate of ambiguous placement. The reference, read quality, contamination, and repetitive sequence can all affect these rates.

A duplicate-marking procedure identifies fragments that share the same genomic endpoints under its operational rules. The duplicate rate is the number of marked duplicate fragments divided by the mapped fragments examined. It is a warning about redundancy, not a direct count of PCR cycles: genuinely abundant biological fragments can also share endpoints, particularly in a strongly enriched library or a deeply sequenced sample.

Library complexity concerns how many distinct fragment patterns are represented, and whether additional sequencing is likely to recover new ones. A simple summary is the non-redundant fraction (NRF): the number of distinct mapped fragment endpoint patterns divided by the number of mapped fragments. Under identical endpoint and denominator rules, NRF is approximately one minus the duplicate rate. Other summaries, such as PBC1, compare the number of genomic fragment patterns observed exactly once with the number observed at least once. Such summaries depend on the counting definition and do not by themselves estimate a universal library size.

A useful justification for examining complexity is a simple occupancy model. Suppose there are \(M\) possible fragment patterns and each of \(n\) sequenced fragments independently draws one pattern with equal probability. For any one pattern, the probability it is never observed is \((1-1/M)^n\). Its probability of appearing at least once is therefore \(1-(1-1/M)^n\). Summing this probability over the \(M\) patterns gives the expected number of distinct patterns observed:

As \(n\) increases, the number of distinct patterns still rises, but each additional fragment is increasingly likely to repeat one already observed. Real libraries violate the model’s equal-probability assumption: fragments have unequal probabilities because of biological abundance, assay enrichment, and technical effects. The model explains why duplication tends to rise with sequencing depth; it is not a literal complexity estimator for a real library.

Fragment Distributions and Enrichment

For paired-end data, the inferred fragment length is the genomic span between the aligned read ends. A histogram of these lengths can reveal an unexpected concentration of very short or very long fragments, a restricted range, or a pattern inconsistent with the library-preparation protocol. The expected profile is assay-dependent; there is no single ideal histogram for all chromatin assays. Check whether the distribution reflects the intended preparation, and investigate unexpected modes or abrupt truncation rather than automatically rejecting a sample. For single-end data, a fragment-size profile requires an estimate beyond the read length and should be labelled accordingly.

Enrichment metrics ask whether reads concentrate in regions expected to carry signal. One commonly used summary is the fraction of reads in peaks (FRiP). Define the peak set and denominator explicitly. For example, FRiP can be calculated as the number of uniquely mapped, nonduplicate fragments overlapping called peaks divided by all uniquely mapped, nonduplicate fragments. Because peak calling and filtering choices affect the result, FRiP is not comparable across analyses unless these choices are aligned. A low value may reflect weak enrichment, a mismatched peak set, low depth, or an unsuitable assay-specific expectation.

Another descriptive check compares signal in target regions with background. If counts are normalized to the same library-depth scale, the ratio of target to background in the assay library can be compared with the same ratio in a matched control. This ratio-of-ratios is an interpretable enrichment summary, but it is not a substitute for the assay’s full analysis. It depends on the target and background definitions, control quality, and normalization.

Practical Interpretation and Failure Modes

A defensible QC review begins with measurement conditions, then interprets metrics jointly. Verify that the reference and mapping settings suit the organism and assay; check read quality and contamination indicators; confirm duplicate handling and denominators; and compare fragment and enrichment summaries with libraries prepared and processed similarly. Where several libraries are available, a single outlier metric is a reason to investigate, not an automatic reason to discard a sample.

A common mistake is to reason, “The duplicate rate is high, so this library failed,” or the reverse, “The mapping rate is high, so this library passed.” Neither conclusion follows from that metric alone. A high duplicate fraction may arise after deep sequencing or from strong enrichment; a high mapping rate does not guarantee unique, informative, or enriched fragments. The correction is to report the counting rules, inspect the other metrics, and judge the combined pattern against appropriate assay-specific expectations.

QC metrics have further limits. They cannot by themselves establish biological validity, prove that a control is appropriate, or show that a downstream result is reproducible. Thresholds are useful only when they have been justified for the assay and workflow; applying a universal cutoff across different assays can misclassify libraries. Retain the underlying summaries and document decisions so that a later analysis can distinguish a technical concern from an expected assay feature.