Tumour Mutational Burden (TMB) Calculator
Tumour Mutational Burden (TMB) Calculator
Countable mutations divided by the megabases of coding sequence the assay actually interrogated — and the reason a TMB of 10 on one panel is not a TMB of 10 on another.
Tumour Mutational Burden (TMB)
Mutations, Mb → mut/Mb13 countable mutations over a 1.1 Mb interrogated coding region
Formula
TMB-high, as approved in June 2020: ≥ 10 mut/Mb on FoundationOne CDx
counting noise: coefficient of variation ≈ 1 ÷ √(mutations counted)
- countable mutations
- the somatic variants the assay's rules admit. FoundationOne CDx counts synonymous as well as non-synonymous substitutions and short insertions and deletions, at 5% allele frequency or above; other assays count only protein-changing variants. Two assays counting different things do not produce the same number from the same tumour, and neither is wrong
- interrogated region
- the megabases of coding sequence in the denominator — about 0.8 Mb for FoundationOne CDx, tens of megabases for a whole exome. This is the quantity that makes the units meaningful, and it is also the quantity that sets how many mutations there are to count, and therefore the precision
- germline filtering
- a tumour-only assay must remove the patient's inherited variants without a matched normal sample, usually by subtracting variants present in population databases. Those databases under-represent some ancestries, so the method systematically overestimates burden in people whose ancestry is poorly represented — a matched normal sample is more robust
- why the count is small
- THE POINT OF THE PAGE: at 10 mut/Mb a 1 Mb panel counts about ten mutations. Mutation counts are Poisson, so the coefficient of variation is roughly 1 over the square root of the count — about 30% at ten mutations, and about 5% at the three hundred and forty a whole exome would count for the same tumour
- what it is not
- a measure of how many neoantigens are presented, or of whether they are recognised. Burden is a proxy several steps removed from the immunology, which is why the same threshold performs very differently in different cancer types
Worked example
13 countable mutations over a 1.1 Mb interrogated coding region
TMB = 13 ÷ 1.1 = 11.8 mut/Mb, above the 10 mut/Mb threshold
Now take one mutation away. 12 ÷ 1.1 = 10.9. Take another: 11 ÷ 1.1 = 10.0, exactly on the threshold. Two variant calls separate a clearly positive result from a borderline one, on the same assay and the same specimen
The counting noise is quantifiable. Mutation counts are Poisson, so the coefficient of variation is about 1 ÷ √13 = 28% — not a rounding error, a third of the value
Sequence the same tumour on a 0.5 Mb panel and you would expect about 6 mutations rather than 13. One mutation either way now moves the result by 2.0 mut/Mb, and 6 ÷ 0.5 = 12.0 against 5 ÷ 0.5 = 10.0
Sequence it by whole exome instead — 34 Mb — and you would expect about 340 mutations. One either way moves the result by 0.03 mut/Mb, which does not show at one decimal place: 340 ÷ 34 and 341 ÷ 34 both print 10.0
None of that is measurement error in the usual sense. It is the arithmetic consequence of dividing a small integer by a small number, and it is why a mutational burden near the threshold should be read as an estimate with a wide interval rather than as a value
The same true burden of 10 mut/Mb, measured five ways
| Interrogated region | Mutations expected | One mutation moves TMB by | Counting noise (1 ÷ √n) |
|---|---|---|---|
| 0.5 Mb | 5 | 2.0 mut/Mb | 45% |
| 0.8 Mb — the FoundationOne CDx footprint | 8 | 1.25 mut/Mb | 35% |
| 1.1 Mb | 11 | 0.9 mut/Mb | 30% |
| 1.5 Mb | 15 | 0.67 mut/Mb | 26% |
| 34 Mb — whole exome | 340 | 0.03 mut/Mb | 5.4% |
Why one mutational burden is not another
| Difference between assays | What it does to the number | What has been measured |
|---|---|---|
| Footprint | Sets the precision, not the average. A small panel scatters around the true value; it does not systematically raise or lower it | Below 0.5 Mb the variance rises sharply; panels under about 667 kb cannot hold adequate agreement with whole exome at the 10 mut/Mb cutoff, and under 577 kb positive percent agreement falls below 85% |
| Which variants count | Systematic. Counting synonymous substitutions as well as protein-changing ones raises every value from that assay relative to one that does not | FoundationOne CDx counts synonymous and non-synonymous substitutions plus short insertions and deletions at 5% allele frequency or above; other assays count only protein-changing variants |
| Cancer-gene enrichment | Systematic and upward, because a panel built of cancer genes is enriched for mutated sequence | Against whole exome at the 10 mut/Mb cutoff, the proportion of patients overestimated by a cancer-gene panel ran from 0 to 17% by cancer type on one panel and 0 to 27% on another, and reached 17.7% (62 of 350) in lung cancer |
| Germline filtering | Systematic and upward where filtering is weak, and unequally so | Filtering at any population minor allele frequency above 0% overestimates burden relative to whole exome; tumour-only filtering against population databases inflates burden most in people whose ancestry those databases under-represent |
| Pathogenic-variant handling | Systematic and upward if driver mutations are left in the count | Failure to filter known pathogenic variants overestimated burden relative to whole exome for every assay examined in the harmonisation study |
| Tumour content of the specimen | Downward. Variants below the detection limit are simply not counted | Low tumour purity causes underestimation, because variants from tumour DNA become harder to detect against the normal DNA in the block |
Mutational burden and microsatellite instability overlap; they are not the same test
| Question | Answer in a series of more than 100,000 profiled cancers |
|---|---|
| Do MSI-high tumours have a high mutational burden? | Almost always — 97% of MSI-high samples had a burden of 10 mut/Mb or more |
| Are high-burden tumours microsatellite unstable? | Usually not — only 16% of high-burden samples were MSI-high |
| What accounts for the rest? | Ultraviolet and tobacco damage, POLE and POLD1 proofreading defects, and therapy-related hypermutation, none of which destabilise microsatellites |
| What is the median burden overall? | 3.6 mut/Mb, across a range from 0 to 1,241 |
A ratio whose denominator belongs to the assay
Tumour mutational burden is the number of somatic mutations found, divided by the amount of coding sequence the assay looked at. The arithmetic is a single division. Everything that makes the figure hard sits in the two quantities being divided, and both of them are properties of the assay rather than of the tumour. The numerator depends on which variants an assay's rules admit to the count — synonymous substitutions or not, short insertions and deletions or not, above what allele frequency, with which germline variants subtracted and by what method. The denominator depends on how much coding sequence the panel covers. Change either and the same tumour yields a different number.
That would be an academic complaint if the number were not used against a fixed threshold. In June 2020 pembrolizumab was granted accelerated approval for previously treated, unresectable or metastatic solid tumours with a burden of at least 10 mutations per megabase and no satisfactory alternative treatment, determined by an approved test. The evidence was the biomarker analysis of the KEYNOTE-158 trial: among 790 evaluable patients, 102 met the threshold and 29% of them responded, against 6% of the 688 who did not — a genuine and large separation, with 4% of responses complete and about half of them lasting two years or more. But the threshold was defined on one assay, FoundationOne CDx, which interrogates roughly 0.8 megabases and counts synonymous variants as well as protein-changing ones. Read on a different assay, 10 mut/Mb is a different quantity. The companion diagnostic's own labelling says as much, warning that its calculation may differ from others depending on the amount of genome interrogated, the percentage of tumour and the assay's limit of detection.
There is a second problem, and it is arithmetic rather than policy. At a burden of 10 mutations per megabase, a one-megabase panel counts about ten mutations. Mutation counts behave as counts do: the relative uncertainty falls as one over the square root of the number counted, which is about 30% at ten mutations. A single variant call — included or excluded by a filtering decision that could reasonably have gone either way — moves the result by a full unit. Whole-exome sequencing of the same tumour would count several hundred mutations and the same relative uncertainty would be about 5%. This is why panels below about 0.5 megabases become unusable for the purpose, why a minimum of one megabase is recommended for clinical work, and why a borderline burden deserves to be reported as an estimate with an interval rather than as a value with a decimal place.
None of this makes the biomarker useless, but it does bound the claim. Burden correlates with neoantigen load, and neoantigen load is one input to whether a tumour is recognised — which is several steps removed from response. A re-analysis across cancer types found that in those where CD8 T-cell infiltration tracks neoantigen load — lung, bladder, melanoma — a high burden was associated with a response rate of 39.8% against a substantially lower rate in low-burden tumours. In cancer types where that relationship does not hold, including breast, prostate and glioma, high-burden tumours responded at 15.3%, which was worse than low-burden tumours of the same types. The authors' conclusion was that a single threshold cannot identify who benefits across all cancers. A mutational burden figure is therefore best read as one line of a report: with the assay named, the count and the footprint stated, the microsatellite status beside it, and the histology in mind.
Frequently asked questions
How is tumour mutational burden calculated?
Divide the number of countable somatic mutations by the size of the coding region the assay interrogated, in megabases. Thirteen mutations over a 1.1 Mb footprint gives 11.8 mut/Mb. What counts as a countable mutation, and how large the footprint is, are decided by the assay — which is why the same tumour gives different values on different assays.
What is the TMB-high threshold and where does it come from?
At least 10 mutations per megabase. It comes from the June 2020 accelerated approval of pembrolizumab for previously treated unresectable or metastatic solid tumours with no satisfactory alternative, based on the KEYNOTE-158 biomarker analysis, and it was defined on the FoundationOne CDx companion diagnostic rather than on sequencing in general.
Can a TMB from one assay be compared with a TMB from another?
Not directly. Assays differ in footprint, in whether synonymous variants and short insertions and deletions are counted, in the allele-frequency cutoff, and in how germline variants are removed. Cancer-gene panels overestimate relative to whole exome, reclassifying up to 17% or 27% of patients at the 10 mut/Mb cutoff depending on the panel. Comparability requires deliberate calibration to a common reference.
Why are small panels less reliable for TMB?
Because they count fewer mutations. The relative uncertainty of a count is roughly one over its square root, so a 0.5 Mb panel counting five mutations carries about 45% uncertainty where a whole exome counting 340 carries about 5%. Below about 0.5 Mb the variance rises sharply, and a minimum of 1 Mb is recommended for clinical use.
Is a high TMB the same as microsatellite instability?
No, and the overlap is one-way. In a series of more than 100,000 cancers, 97% of MSI-high tumours had a burden of at least 10 mut/Mb, but only 16% of high-burden tumours were MSI-high. Ultraviolet and tobacco damage and polymerase proofreading defects produce high burdens without destabilising microsatellites, so a high burden does not answer a question about mismatch repair.
Does a high TMB predict response to immunotherapy in every cancer type?
No. In cancers where CD8 T-cell infiltration tracks neoantigen load — lung, bladder, melanoma — high burden was associated with a 39.8% response rate. In breast, prostate and glioma, high-burden tumours responded at 15.3%, worse than low-burden tumours of the same types. The authors concluded that one threshold cannot identify who benefits across all cancers.
Related calculators
References
- Marabelle A, Fakih M, Lopez J, et al. Association of tumour mutational burden with outcomes in patients with advanced solid tumours treated with pembrolizumab: prospective biomarker analysis of the multicohort, open-label, phase 2 KEYNOTE-158 study. Lancet Oncol. 2020;21(10):1353–1365.
- US Food and Drug Administration. FDA approves pembrolizumab for adults and children with TMB-H solid tumors. 16 June 2020 — the accelerated approval, the ≥10 mut/Mb threshold and the FoundationOne CDx companion diagnostic.
- Chalmers ZR, Connelly CF, Fabrizio D, et al. Analysis of 100,000 human cancer genomes reveals the landscape of tumor mutational burden. Genome Med. 2017;9(1):34.
- Vega DM, Yee LM, McShane LM, et al. Aligning tumor mutational burden (TMB) quantification across diagnostic platforms: phase II of the Friends of Cancer Research TMB Harmonization Project. Ann Oncol. 2021;32(12):1626–1636.
- Fang H, Bertl J, Zhu X, et al. Tumour mutational burden is overestimated by target cancer gene panels. J Natl Cancer Cent. 2023;3(1):56–64.
- Doig KD, Fellowes A, Scott P, Fox SB. Tumour mutational burden: an overview for pathologists. Pathology. 2022;54(3):249–253.
- Golkaram M, Zhao C, Kruglyak K, et al. The interplay between cancer type, panel size and tumor mutational burden threshold in patient selection for cancer immunotherapy. PLoS Comput Biol. 2020;16(11):e1008332.
- McGrail DJ, Pilié PG, Rashid NU, et al. High tumor mutation burden fails to predict immune checkpoint blockade response across all cancer types. Ann Oncol. 2021;32(5):661–672.
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.