Skip to main content
AACR Project GENIE v19.0 · 21,017 myeloid samples Panel-adjusted Fisher's exact with Benjamini-Hochberg FDR N=1 case study · Not clinical guidance

Statistical Rarity

Bayesian rare-event estimation; 0 of 8,655 patients, exact 95% bound below 1 in 2,889

Jeffreys Median Ultra-rare
2.63e-5
Posterior median under Jeffreys Beta(0.5, 0.5) prior i
Patients for 50%
25,965
From the zero count alone (Jeffreys); ignores gene frequencies i
Gene Level (fitted)
about 1 in 25,000
All five genes mutated, fitted pairwise maximum-entropy (Ising) model; the exact variants have no model value i
Combined Search Zero
0 matches
8,655 GENIE patients, all five genes on panel

Jeffreys Posterior Estimation

With 0 observed quintuples in 8,655 myeloid patients, the Jeffreys non-informative prior Beta(0.5, 0.5) yields a posterior Beta(0.5, 8,655.5). The posterior median (2.63e-5) represents the best point estimate of the true population frequency. The 95% credible interval spans 5.67e-8 to 2.90e-4, providing bounds on the plausible frequency range.
Statistic Value Interpretation
Posterior median 2.63e-5 Best point estimate of true frequency
Posterior mean 5.78e-5 Mean of Beta(0.5, 8,655.5)
95% CI lower 5.67e-8 Lower 2.5th percentile
95% CI upper 2.90e-4 Upper 97.5th percentile
One-sided 95% upper 2.22e-4 Conservative upper bound
Source: bayesian_rarity_results.json. Jeffreys prior: non-informative conjugate prior for binomial data.

Rule of Three (Frequentist)

The Rule of Three (Hanley and Lippman-Hand, 1983) provides a quick frequentist upper bound: with 0 events in n trials, the 95% upper confidence limit is approximately 3/n. For 8,655 patients, this gives an upper bound of 3.47e-4. The Clopper-Pearson exact interval (3.46e-4) and Poisson exact 99% bound (5.32e-4) provide alternative frequentist estimates, all consistent with the Bayesian credible interval. Patients are the unit (several samples from one patient are merged), so the Clopper-Pearson bound is the figure used elsewhere on this portal: below 1 in 2,889.
Method Upper Bound Confidence Level
Rule of Three 3.47e-4 95%
Clopper-Pearson exact 3.46e-4 95%
Poisson exact 5.32e-4 99%
Source: bayesian_rarity_results.json. Hanley JA, Lippman-Hand A. If nothing goes wrong, is everything all right? JAMA (1983).

Gene-Level vs Variant-Level Rarity

The five genes (DNMT3A, IDH2, SETBP1, PTPN11, EZH2) are individually common in myeloid malignancies. At the gene level, with frequencies from the REVOLVER input (7,956 patients, all carrying at least one target-gene mutation, so enriched), the expected co-occurrence count is 8.99e-3 (0.0034 with the frequencies of the 18,625 panel-eligible samples), placing this combination at the 77.8th percentile among all 278,256 possible five-gene combinations from 34 myeloid genes. A fitted pairwise maximum-entropy (Ising) model puts mutations in all five genes at about 1 in 25,000 myeloid patients (1 in 13,000 to 1 in 65,000).

The exact five-variant combination has no estimable population frequency. Zero in 8,655 patients bounds it below 1 in 2,889 (95%, exact binomial). An earlier "variant specificity" multiplier was withdrawn on 3 November 2026: it gave EZH2 V662A, which has 0 carriers, an arbitrary 1/N value.
Gene-Level Expected
8.99e-3
REVOLVER-input frequencies (enriched); 77.8th percentile among 278,256 combinations
Quintuple Bound
< 1 in 2,889
Exact 95% bound from 0 of 8,655 patients; no point estimate
Source: bayesian_rarity_results.json. Extreme Value Theory: Coles S. Statistical Modeling of Extreme Values. Springer (2001).

Posterior Predictive: Patients Needed

Using the Jeffreys posterior, the probability of observing at least one quintuple match in a future cohort of M patients is computed via the posterior predictive distribution. 25,965 patients are needed for a 50% chance of finding one match. 3,453,444 patients are needed for 95% confidence. These figures use only the observation 0 of 8,655 patients and ignore the gene frequencies. At gene level, a fitted pairwise maximum-entropy (Ising) model puts mutations in all five genes at about 1 in 25,000 patients; for the exact five variants no model value is defensible, because EZH2 V662A has 0 carriers.
Future Cohort Size P(at least 1 match)
1,000 5.3%
5,000 20.4%
10,000 31.9%
50,000 61.6%
100,000 71.8%
500,000 87.0%
1,000,000 90.7%
50% Detection
25,965 patients
Future cohort needed for 50% chance of 1 match
95% Detection
3,453,444 patients
Future cohort needed for 95% chance of 1 match
Source: bayesian_rarity_results.json. Posterior predictive under Jeffreys conjugate Beta prior.

Extreme Value Theory

All 278,256 possible five-gene combinations from the 34 target myeloid genes were enumerated and their expected counts computed under independence. A Generalized Extreme Value (GEV) distribution was fitted to the log-transformed expected counts. The patient quintuple ranks 216,373 out of 278,256 at the gene level (77.8th percentile), confirming that these five genes are individually common enough in myeloid malignancies that their gene-level co-occurrence is unremarkable. The exact variant combination has no estimable frequency beyond the exact bound (below 1 in 2,889).
EVT Metric Value Interpretation
Gene-level percentile 77.8th Genes common, combination not extreme
Gene-level expected count 8.99e-3 Expected quintuples at gene level (REVOLVER-input frequencies, enriched)
Fitted Ising, all five genes about 1 in 25,000 Pairwise maximum-entropy model, patient level (1 in 13,000 to 1 in 65,000)
GEV shape parameter 0.2278 scipy c > 0, i.e. ξ < 0: bounded (Weibull-type) tail, not heavy-tailed
The fitted GEV shape is c = 0.23 in scipy's sign convention, which is ξ = -0.23 in the usual one: a bounded (Weibull-type) tail, not a heavy tail, so the rarest gene combinations do not stretch out without limit. The patient quintuple sits in the body of this distribution at the gene level.
Source: bayesian_rarity_results.json. Coles S. Statistical Modeling of Extreme Values. Springer (2001). Babur et al. Genome Biol (2015). Canisius et al. Genome Biol (2016).
References
  1. AACR Project GENIE Consortium. AACR Project GENIE: Powering Precision Medicine through an International Consortium. Cancer Discov (2017). DOI
  2. Coles S. An Introduction to Statistical Modeling of Extreme Values. Springer Series in Statistics (2001). DOI
  3. Babur O, Gonen M, Aksoy BA, et al. Systematic identification of cancer driving signaling pathways based on mutual exclusivity of genomic alterations. Genome Biol (2015). DOI
  4. Canisius S, Martens JW, Wessels LF. A novel independence test for somatic alterations in cancer. Genome Biol (2016). DOI
  5. Hanley JA, Lippman-Hand A. If nothing goes wrong, is everything all right? Interpreting zero numerators. JAMA (1983). PubMed