NGS Run Quality Interpreter
NGS Run Quality Interpreter
Whether a sequencing run or a sample within it is fit to report, from Q30 against the instrument’s own read-length-specific specification, the coverage distribution across the target, the duplication rate and the on-target fraction. No professional body publishes universal pass marks, and this page does not invent any.
Does this run or sample pass QC?
Q30, duplication, on-target, uniformityA 2 × 150 bp germline panel run. 88% of bases at Q30, clusters passing filter in the usual range, the capture performing normally, 18% duplicates, and 97% of target bases covered at 20× or more
What each metric is, and what governs its threshold
everything else is compared against the laboratory’s own validated baseline
- Q30
- the proportion of called bases with a Phred quality of 30 or more, meaning a one-in-a-thousand chance of being wrong. The instrument specification falls as read length rises: above 85% for 2 × 75, above 80% for 2 × 150, above 75% for 2 × 250 and above 70% for 2 × 300, measured on a PhiX control library at the supported cluster density
- clusters passing filter
- the proportion of clusters the instrument’s chastity filter accepts. No clinical threshold is published, because the acceptable figure depends on library type and loading concentration, and an over-clustered flow cell can pass a percentage target while producing poor quality. Read it with cluster density and with phasing and pre-phasing
- on-target fraction
- reads mapping inside the designed target. Entirely a property of the capture design — routinely above 95% for amplicon assays, 70 to 80% for a small hybrid-capture panel, 50 to 60% for an exome. There is no cross-assay pass mark and this page does not print one
- coverage distribution
- the percentage of target bases at or above the calling floor. This is the metric that decides whether a negative result means anything, and it is the one a mean depth hides. The ACMG standards’ exome example pairs a mean of 100× with 90 to 95% of bases at 10× or more, for exactly that reason. For the expected MEAN depth of a planned run, use the coverage depth calculator instead
- duplication rate
- reads that are copies of the same original molecule. It converts raw depth into a smaller effective depth, and its normal range depends on input quantity, amplification cycles and specimen type — heavily elevated as a matter of course in formalin-fixed material
- the governing rule
- the ACMG clinical laboratory standards for next-generation sequencing: “it is not possible to recommend a specific minimum threshold for coverage, and laboratories will need to choose minimum coverage thresholds in accordance with total metrics for analytical validation”. Its worked examples — 10 to 20× minimum for a targeted panel, a mean of 100× with 90 to 95% of bases at 10× for an exome, a mean of 30× for a genome — are examples and are labelled as such
Worked example
A 2 × 150 bp germline panel run. 88% of bases at Q30, clusters passing filter in the usual range, the capture performing normally, 18% duplicates, and 97% of target bases covered at 20× or more
The 2 × 150 specification is above 80% of bases at Q30. The run gives 88%, so the run-level quality gate is passed
Clustering is in its usual range and the capture is performing normally, so neither run-level nor library-level failure applies
97% of target bases are at 20× or above — at or above the 95% line this page uses for a clean pass, and well above the 90% line below which a negative result stops meaning anything
18% duplicates is unremarkable for a fresh high-input germline library. Effective depth is a little below raw depth, as always, which is why the coverage figure is the one to quote
Result: within specification. The report should still list whatever small proportion of the target sits below 20×, because those bases were not tested
Now change one thing: make it a 2 × 300 run at the same 88% Q30. It still passes, and more comfortably, because the specification for 2 × 300 is above 70%. Change it the other way, to 2 × 75, and 88% is still a pass but only just — the specification there is above 85%. The same Q30 number means three different things
Q30 specification by read length — one number is not enough
| Read length | Specification, % of bases above Q30 | A run at 88% Q30 |
|---|---|---|
| 2 × 75 bp | above 85% | Passes, narrowly |
| 2 × 150 bp | above 80% | Passes |
| 2 × 250 bp | above 75% | Passes comfortably |
| 2 × 300 bp | above 70% | Passes very comfortably |
What each metric can and cannot tell you
| Metric | What a bad value means | What a good value does not prove |
|---|---|---|
| Q30 | A run-level sequencing problem — over-clustering, reagents, phasing. No sample on the run escapes it | That any individual base was covered. Q30 is about the quality of the bases that were read, not about which ones were |
| Clusters passing filter | The flow cell was loaded wrongly or the library was mis-quantified | That the reads went to the right place, or that the library was complex |
| On-target fraction | The hybridisation or the library preparation has failed. More sequencing will not fix the proportion | That coverage is even inside the target. A high on-target fraction can still be piled onto a few regions |
| Duplication rate | Too little input, too many cycles, or a degraded specimen. Effective depth is well below raw depth | That the library was deep enough. A low duplication rate on a shallow run is just a shallow run |
| Target at the calling floor | Part of the target was not tested, so a negative result there is not a negative result | That the variants found are real — that is a per-variant question, not a per-run one |
| Mean depth | Very little on its own | Almost nothing about the distribution. This page deliberately does not ask for it |
Why there are no universal pass marks, and what to use instead
People arriving at this subject usually want a table of numbers: the Q30 a run must reach, the duplication rate that fails, the on-target percentage that is acceptable. No professional body publishes one, and that is a deliberate position rather than an omission. The ACMG clinical laboratory standards for next-generation sequencing state it directly — “it is not possible to recommend a specific minimum threshold for coverage, and laboratories will need to choose minimum coverage thresholds in accordance with total metrics for analytical validation” — and the AMP and CAP bioinformatics pipeline validation standards take the same line for every other metric. The reason is that these numbers describe an assay, not a technology. An amplicon panel routinely runs above 95% on target; an exome runs at 50 to 60%, and 60% is excellent for one and a catastrophe for the other.
What can be checked against a published number is Q30, and even there the number is not a constant. The instrument’s install specification depends on read length, because sequencing quality decays along a read: the same chemistry is specified at above 85% of bases at Q30 for 2 × 75 and only above 70% for 2 × 300. A laboratory carrying a single remembered Q30 pass mark will fail long runs that are performing exactly as designed and pass short runs that are not. That is why the read length is the first thing this page asks for.
The metric that decides whether a result means anything is the coverage distribution, and it is the one most often replaced by a mean. A mean depth of 300× across a target is entirely compatible with a tenth of that target sitting at 5×, because capture efficiency varies enormously with GC content, repeat structure and probe design — and it is those shallow bases, not the deep ones, that generate false negatives. The ACMG standards’ own worked example for exome sequencing pairs a mean of 100× with a separate requirement that 90 to 95% of target bases reach at least 10×, precisely because the mean does not imply the tail. This page therefore asks for the percentage of target at or above the calling floor and does not ask for the mean at all. If what you want is the expected mean depth of a run you have not done yet, from read counts and target size, that is the NGS coverage depth calculator and it is a different calculation.
Two consequences follow for the report rather than for the run. First, whatever percentage of the target fell below the calling floor should be listed by region on the report, every time. A negative result over an untested base is not a negative result, and a clinician who is not told which regions were not covered will read the report as an exclusion. Second, a run that passes at the run level says nothing about any individual call within it: whether a particular variant needs orthogonal confirmation is a separate question with its own metrics, and it is on the NGS variant confirmation interpreter. Run quality is necessary and not sufficient.
Finally, treat all of these as trends rather than as events. A single slightly low on-target fraction is noise; the same fraction falling across five runs is ageing probes or a hybridisation step drifting, and it is far easier to fix while samples are still passing than after one fails. The value of a QC metric is mostly in its history, and a laboratory that records these figures run on run gets an early warning that no single-run threshold can give it. For the variant-level question that follows a clean run, see the NGS variant confirmation interpreter and the ACMG variant classification interpreter; for the expected depth of a planned run, the NGS coverage depth calculator.
Frequently asked questions
What Q30 percentage is acceptable for a sequencing run?
It depends on the read length, and that is the part most often missed. The instrument specification falls as reads get longer because quality decays along a read: for the MiSeq the figures are above 85% of bases at Q30 for 2 × 75, above 80% for 2 × 150, above 75% for 2 × 250 and above 70% for 2 × 300, measured on a PhiX control library at the supported cluster density. A single remembered pass mark will therefore over-fail long runs and under-fail short ones. Use your own instrument’s specification for the run length you actually performed, and validate your own acceptance criterion against it.
Is there a standard duplication rate or on-target percentage for NGS?
No, and no professional body publishes one. Both are properties of the assay rather than of the technology. On-target fraction is set mainly by capture design — routinely above 95% for amplicon assays, 70 to 80% for a small hybrid-capture panel, 50 to 60% for an exome — so a figure that is excellent for one design is a failure for another. Duplication rate depends on input quantity, amplification cycles and specimen type, and formalin-fixed material duplicates far more than fresh material as a matter of course. The meaningful comparison in both cases is against what this assay gives in this laboratory when it is working, which is why this page asks for that comparison rather than for a number.
Why is mean coverage depth a poor QC metric on its own?
Because it says nothing about the distribution, and the distribution is what determines whether a negative result means anything. Capture efficiency varies with GC content, repeat structure and probe design, so a target with a mean depth of 300× can contain exons at 5× that cannot be called at all. The ACMG standards’ worked example for exome sequencing makes the point by pairing a minimum mean coverage of 100× with a separate requirement that 90 to 95% of target bases reach at least 10×. This page asks for the percentage of the target at or above the calling floor for that reason and does not ask for the mean.
Can a sample be reported if part of the target was not covered?
Frequently yes, and the requirement is not that coverage be complete but that the gaps be declared. Whatever proportion of the target fell below the calling floor should be listed by region on the report, because a clinician reading a negative result over an untested base will take it as an exclusion. Where the same regions drop out repeatedly, that is an assay design finding rather than a run finding — repeating the same capture will reproduce the same gap — and the answer is a redesigned probe set or a targeted fill-in by another method.
Does a run that passes QC mean every variant in it can be reported?
No. Run-level and sample-level quality are necessary and not sufficient. An individual call within a clean run can still be an artefact: a homopolymer indel from polymerase slippage, a strand-biased damage artefact, or a mismapped read pile in a region with a close paralogue, none of which a run-level Q30 or a coverage figure can see. Whether a particular call needs orthogonal confirmation is a separate judgement with its own metrics, and it has its own page.
What causes a high duplication rate and does it matter?
It nearly always comes from the front of the process: too little input DNA, too many PCR cycles, or a degraded specimen. It matters because duplicate reads are re-readings of the same original molecule, so they inflate the raw read count without adding information — the effective depth is lower than the raw depth, and adding more sequencing from the same library does not help. Trust the post-deduplication coverage figure over the read count. Where high duplication is unavoidable, as in low-input or circulating-tumour-DNA work, unique molecular identifiers change the picture by allowing duplicates to be collapsed into consensus reads rather than discarded.
Related calculators
References
- Rehm HL, et al.; ACMG Laboratory Quality Assurance Committee. ACMG clinical laboratory standards for next-generation sequencing. Genet Med. 2013;15(9):733–747, and the 2021 revision doi:10.1016/j.gim.2021.08.005. “It is not possible to recommend a specific minimum threshold for coverage, and laboratories will need to choose minimum coverage thresholds in accordance with total metrics for analytical validation”; worked examples of “10–20X as a minimum for covering all bases of a targeted panel”, “a minimum mean coverage of 100X for the proband and 90–95% of bases in the laboratory’s defined target reach at least 10X coverage” for exome sequencing, and “a minimum mean coverage of 30X” for genome sequencing.
- Roy S, Coldren C, et al. Standards and Guidelines for Validating Next-Generation Sequencing Bioinformatics Pipelines: A Joint Recommendation of the Association for Molecular Pathology and the College of American Pathologists. J Mol Diagn. 2018. The metric list this page is built around, and the requirement that acceptance criteria be established during validation rather than borrowed.
- Jennings LJ, et al. Guidelines for Validation of Next-Generation Sequencing-Based Oncology Panels: A Joint Consensus Recommendation of the Association for Molecular Pathology and College of American Pathologists. J Mol Diagn. 2017.
- Illumina. MiSeq System Specifications. Quality scores by read length: “> 90% bases higher than Q30” at 2 × 25 (v2), “> 85%” at 2 × 75 (v3), “> 80%” at 2 × 150 (v2), “> 75%” at 2 × 250 (v2) and “> 70%” at 2 × 300 (v3); “Install specifications based on Illumina PhiX control library at supported cluster densities between 1000-1200 k/mm² for v2 chemistry and 1200–1400 k/mm² for v3 chemistry.”
- Gargis AS, et al. Assuring the quality of next-generation sequencing in clinical laboratory practice. Nat Biotechnol. 2012. The original statement that NGS quality metrics are assay-specific and must be established per assay.
Medical Disclaimer: The tools and content provided here are for educational and reference purposes only. They are not intended to substitute for professional medical advice, diagnosis, or treatment. Clinical decisions should always be based on the comprehensive assessment of a qualified healthcare professional.