NGS Variant Confirmation Interpreter
NGS Variant Confirmation Interpreter
Whether a next-generation sequencing call needs orthogonal confirmation before it is reported, from its depth, allele fraction, strand support, genomic context and what it is being reported as. The two published threshold sets disagree with each other, and the page shows you where you are against both.
Does this call need Sanger?
Call metrics + context → confirm or notA missense single-nucleotide variant in unique sequence, being reported as likely pathogenic. Read depth 180, variant allele fraction 47%, supporting reads balanced across both strands
The two published threshold sets, and the one document that governs both
the conservative set: depth > 100 AND allele fraction > 40%
- depth ≥ 20, fraction ≥ 20%
- from a study of 1,109 variants in 825 clinical exomes: 1,079 variants meeting FILTER=PASS, QUAL at least 100, depth at least 20 and fraction at least 20% gave 100% concordance with Sanger, with zero false-positive single-nucleotide variants (866) and zero false-positive indels (213). Homopolymer regions, pseudogene regions and variants with nomenclature uncertainty were excluded from that claim
- depth over 100, fraction over 40%
- from a study of 7,845 non-polymorphic variants in which 1.3% were NGS false positives. False positives had a mean allele fraction of 16% and a mean coverage of 223; true variants had 46% and 415. Dropping confirmation entirely would have reduced sensitivity from 100% to 97.8%, missing 176 confirmed variants
- QUAL
- not an input on this page, because the scale is not comparable between callers and a QUAL of 100 from one pipeline is not a QUAL of 100 from another. If your pipeline emits one, apply your own validated threshold alongside the two above
- the governing rule
- neither threshold set is a standard. The AMP and NSGC joint report states that “read depth, variant type, variant allele fraction, genomic context, and many current quality score calculations have been shown to be inadequate alone”, requires combinations of criteria validated on the laboratory’s own data using a dataset separate from the one they were derived on, and requires the policy to be written, approved by the laboratory director and stated on the report
Worked example
A missense single-nucleotide variant in unique sequence, being reported as likely pathogenic. Read depth 180, variant allele fraction 47%, supporting reads balanced across both strands
It is being reported as clinically actionable, so the question is live rather than moot
Unique sequence: no paralogue or pseudogene, no homopolymer or repeat. Neither of the two "always confirm" contexts applies
Strand support balanced: no artefact signature
A single-nucleotide variant, not an indel, so the nomenclature ambiguity argument does not apply either
Depth 180 is above 100 and the allele fraction of 47% is above 40% — this clears the more conservative of the two published threshold sets, not merely the floor
So the technical case for omitting confirmation is as strong as the literature gets. It is still not permission. Whether this laboratory may report it without Sanger depends on whether this laboratory has validated omission at these metrics, on this assay, and written it into a policy the director has approved
Now change one thing: make it an insertion. The answer flips to confirm, and not because the call looks any worse — because the exact representation of a reportable indel is itself something confirmation settles
What the two studies actually found
| Exome study, 1,109 variants | Panel study, 7,845 variants | |
|---|---|---|
| Threshold proposed | FILTER=PASS, QUAL ≥ 100, depth ≥ 20, fraction ≥ 20% | depth > 100 and fraction > 40% |
| False positives found | Zero, among 1,079 high-quality calls — 866 single-nucleotide variants and 213 indels | 1.3% of variants were NGS false positives |
| What false positives looked like | — | mean allele fraction 16%, mean coverage 223, against 46% and 415 for true variants |
| Cost of dropping confirmation entirely | Not assessed; 3 of 789 discrepancies were Sanger false negatives rather than NGS false positives | Sensitivity would fall from 100% to 97.8%, missing 176 confirmed variants |
| Explicitly excluded | homopolymer regions, pseudogene regions, variants with nomenclature uncertainty, prenatal cases | — |
| Conclusion drawn | Sanger is no longer always necessary | Sanger confirmation is required to achieve optimal sensitivity and specificity |
When confirmation is not optional
| Situation | Why |
|---|---|
| Homopolymers, short tandem repeats, low-complexity sequence | Named in the AMP and NSGC report as regions to “always” confirm. Polymerase slippage produces systematic, reproducible indel errors, so seeing the same call in other samples is not evidence it is real |
| Genes with a pseudogene or close paralogue | The failure is mispping, not miscalling, so every quality metric looks normal. Confirmation must be gene-specific — generic Sanger primers amplify both copies |
| Strongly strand-biased support | The signature of oxidative or fixation damage introduced before sequencing. Depth does not rescue it |
| A variant allele fraction far below 50% | Could be an artefact, a mosaic, clonal haematopoiesis or contamination. Note that Sanger’s own limit of detection is around 15 to 20%, so it may not be the right confirmatory method |
| Reportable insertions and deletions | Confirmation settles the representation as well as the existence, and the representation is what cascade testing and database submission depend on |
| Any doubt about sample identity | The AMP and NSGC report notes that a confirmation assay doubles as a check against sample and data mix-ups. A laboratory that stops confirming needs another way to catch them — a SNP identity assay or spiked-in controls |
The argument for dropping Sanger, and what it does not cover
For the first decade of clinical next-generation sequencing, every reportable variant was confirmed by Sanger sequencing. That was a reasonable default when NGS was new and its error modes were poorly characterised, and it is expensive: confirmation adds days to turnaround and a substantial fraction of the cost of a panel or exome, for a step that, on most calls, changes nothing. The case for dropping it is therefore real and the literature behind it is serious.
The trouble is that the literature disagrees with itself, and understanding why is more useful than picking a side. One study of 1,109 variants in 825 clinical exomes found that 1,079 calls meeting a straightforward quality filter — the caller’s own PASS, a quality score of at least 100, at least 20 reads and a variant fraction of at least 20% — matched Sanger perfectly, with zero false positives among 866 single-nucleotide variants and 213 indels. It also found that three of the discrepancies it did see were Sanger false negatives rather than NGS false positives, which is a point worth sitting with: the confirmatory method is not automatically the truth. Another study, of 7,845 variants on targeted panels, found 1.3% of calls were NGS false positives, characterised them (mean allele fraction 16% and mean coverage 223, against 46% and 415 for the true ones), and concluded that confirmation should be retained for anything below 100 reads and a 40% fraction, because abandoning it entirely would have cost 176 confirmed variants and taken sensitivity from 100% to 97.8%.
Both are right about their own assay. False positive behaviour is a property of a particular capture design, chemistry, aligner and variant caller, and it does not transfer. That is exactly the position the 2023 AMP and NSGC joint report takes: it sets no threshold, states that read depth, variant type, allele fraction, genomic context and quality scores “have been shown to be inadequate alone”, and requires laboratories to derive combinations of criteria and validate them on their own data — using a separate dataset from the one the criteria were derived on, which is the step that prevents a policy from simply describing the data it was fitted to. It also requires the policy to be written down, approved by the laboratory director, made available to ordering clinicians and summarised on the report.
Three things the depth-and-fraction argument does not reach at all, and they are the reason this page asks about context before it asks about metrics. In a region with a close paralogue or a pseudogene, the error is in the mapping rather than the base calling, so a beautifully covered call with a perfect allele fraction can be in the wrong gene; PMS2, SMN1, CYP21A2, GBA, NEB and STRC are the standing examples, and generic Sanger primers do not resolve them either. In a homopolymer or short tandem repeat, polymerase slippage generates systematic errors that recur across samples, which is why the AMP and NSGC report warns specifically against treating repeated observation of a variant in other samples as confirmation. And an indel’s problem is often its description rather than its existence: two pipelines will write the same deletion at different coordinates, and confirmation settles which representation goes into the report and into every cascade test that follows.
One last thing that is easy to lose when confirmation is dropped. A Sanger confirmation is also, incidentally, a sample identity check — it catches a swap or a data mix-up that no amount of sequencing depth would reveal, because the wrong sample sequences perfectly well. A laboratory that stops confirming universally needs another mechanism for that, and the AMP and NSGC report names the options: an independent SNP identity assay, or spiked-in tracking oligonucleotides. Related pages: the variant allele frequency calculator for an allele fraction that does not look germline, the NGS run quality interpreter for whether the run the call came from passed QC at all, the NGS coverage depth calculator for the depth the assay was designed to deliver, and the ACMG variant classification interpreter for what the variant means once it is confirmed.
Frequently asked questions
Does every NGS variant need Sanger confirmation?
No, and no professional body requires it. The 2023 AMP and NSGC joint report allows confirmation to be omitted for variants that meet criteria “rigorously demonstrated to ensure high positive predictive value”, provided the laboratory has validated those criteria on its own data, written them into a director-approved policy and stated the policy on its reports. What it does not do is tell you what the criteria should be, because false positive behaviour differs between assays. Certain situations remain exceptions whatever the metrics: low-complexity and repeat regions, genes with close paralogues or pseudogenes, strand-biased calls, low allele fractions and reportable indels.
What read depth and variant allele fraction allow Sanger to be skipped?
The two published answers differ by roughly fivefold, and both are defensible for the assay they were derived on. A study of clinical exomes found zero false positives among 1,079 variants meeting FILTER=PASS, QUAL at least 100, depth at least 20 and variant fraction at least 20%, excluding homopolymer and pseudogene regions. A study of targeted panels found 1.3% false positives and recommended depth above 100 with a fraction above 40% before confirmation could be dropped. This page treats 20 reads and 20% as a floor nobody argues below, and 100 reads and 40% as the point above which even the conservative study agreed — with your own laboratory’s validation deciding everything in between.
Why do homopolymer regions always need confirmation?
Because the errors there are systematic rather than random. Polymerase slippage during library amplification inserts or deletes a base in a run of identical nucleotides at a rate that climbs with run length, and because the mechanism is reproducible, the same false call appears in sample after sample. That defeats the usual quality heuristics twice over: the call can have excellent depth and a convincing allele fraction, and seeing it in other samples looks like corroboration when it is really the same artefact recurring. The AMP and NSGC report names low-complexity repeats among the regions to confirm always, and the largest study supporting reduced confirmation excluded them from its own claim.
Can Sanger sequencing confirm a low-level mosaic variant?
Often not, and this is a real trap. Sanger’s limit of detection is around 15 to 20% variant allele fraction, so a genuine mosaic below that can come back negative and be wrongly dismissed as an NGS artefact. If the NGS allele fraction is low and mosaicism is plausible, the right response is a more sensitive orthogonal method — droplet digital PCR, deep amplicon sequencing, or a second independent NGS assay — and, where the clinical question allows it, testing a second tissue such as skin fibroblasts or buccal cells, since a mosaic may be absent from blood altogether.
If a variant was confirmed in another patient, does that count as confirmation?
No, and the AMP and NSGC report specifically discourages it. Systematic errors recur: an artefact generated by the chemistry or the alignment at a particular position will appear in every sample that covers that position, so seeing the same call confirmed previously can be evidence that the artefact is reproducible rather than that the variant is real. Repeated confirmation of the same variant across samples can support a laboratory’s validated policy as part of a formal analysis, but it is not a substitute for confirming the call in front of you.
What else does a confirmation assay do besides confirm the variant?
It checks that the sample is the sample. A specimen swap, a plate-position error or a data mix-up produces sequencing data of perfect quality belonging to the wrong person, and no depth or allele fraction threshold can detect that. Orthogonal confirmation catches it incidentally because the confirmatory assay is run on the original specimen. The AMP and NSGC report makes the point explicitly and says that a laboratory which does not confirm universally needs an alternative — an independent SNP identity assay, or spiked-in tracking oligonucleotides — rather than nothing.
Related calculators
References
- Crooks KR, et al. Recommendations for Next-Generation Sequencing Germline Variant Confirmation: A Joint Report of the Association for Molecular Pathology and National Society of Genetic Counselors. J Mol Diagn. 2023;25(7):411–427. doi:10.1016/j.jmoldx.2023.03.012. Confirmation policies must be written and director-approved; single metrics “have been shown to be inadequate alone”; “low-complexity repeats, such as homopolymers and short tandem repeats, mobile elements, and genes with highly similar paralogs or pseudogenes” should “always be subject to orthogonal confirmation”; criteria must be validated on a separate dataset; and confirmation doubles as a check for “sample or data mix-ups”.
- Mu W, et al. Sanger Confirmation Is Required to Achieve Optimal Sensitivity and Specificity in Next-Generation Sequencing Panel Testing. J Mol Diagn. 2016. 7,845 non-polymorphic variants, 1.3% NGS false positives; false positives mean heterozygous ratio 16% (range 10–40) and mean coverage 223 (range 10–689) against 46% and 415 for confirmed variants; recommended “minimal read depth coverage of >100 and a variant allele frequency (or heterozygous ratio) of >40%”; dropping confirmation would reduce sensitivity from 100% to 97.8%, missing 176 confirmed variants.
- Sanger sequencing is no longer always necessary based on a single-center validation of 1109 NGS variants in 825 clinical exomes. Sci Rep. 2021. doi:10.1038/s41598-021-85182-w. 1,079 variants meeting FILTER=PASS, QUAL at least 100, depth at least 20X and variant fraction at least 20% gave 100% concordance with Sanger and zero false positives among 866 single-nucleotide variants and 213 indels, excluding homopolymer regions, pseudogene regions and variants with nomenclature uncertainty; 3 of 789 discrepancies were Sanger false negatives.
- Lincoln SE, et al. A Rigorous Interlaboratory Examination of the Need to Confirm Next-Generation Sequencing-Detected Variants with an Orthogonal Method in Clinical Genetic Testing. J Mol Diagn. 2019. The interlaboratory study on which much of the case for selective confirmation rests; cited here for its existence and conclusions, as its full text could not be retrieved in this session and no numeric threshold on this page is attributed to it.
- 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 framework within which a confirmation policy is validated.
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.