Enigma somatic classification
How somatic oncogenicity and clinical significance are computed · 1.0 (25 Oct 2026)
Overview
Somatic variant pages show two complementary results: an oncogenicity classification from the SVIG-UK points framework and a clinical significance tier from AMP/ASCO/CAP evidence. Results depend on the selected tumour type and available public evidence. These are research annotations and require expert review.
- Tier I-A
- Strong clinical significance, level A
- Tier I-B
- Strong clinical significance, level B
- Tier II-C
- Potential clinical significance, level C
- Tier II-D
- Potential clinical significance, level D
- Tier III
- Unknown clinical significance
- Tier IV
- Benign or likely benign
Oncogenicity points
- Oncogenic
- At least 10
- Likely oncogenic
- 6 to 9
- VUS
- 0 to 5
- Likely benign
- -6 to -1
- Benign
- At most -7
- Each code counts once; a strength chosen by hand replaces the automatic one for that code.
- O4 (cancer databases) counts only when O3 (absent or rare in the population) also applies.
- O4 is not counted when B1 (common in the population) applies.
- O10 (functional studies) is not counted together with O2 at Very strong.
- O3 is dropped when two or more benign codes apply.
- Stand-alone codes decide the class: O1 gives the canonical list class, B1 at >1% gives Benign, B2 gives VUS.
- Two-line rule: a (likely) oncogenic total needs at least two oncogenic codes, a (likely) benign total at least two benign codes; otherwise the class is VUS.
Evidence criteria
O1 · Canonical oncogenic variant (stand-alone)
Applied when the gene and protein change are on the SVIG-UK canonical list (Supplementary Table 3); the class is the list class (Oncogenic or Likely oncogenic).
O2 · Null variant in a tumour suppressor gene
Stop-gain, frameshift, canonical splice-site and start-loss changes in a gene whose role is tumour suppressor (or both roles) in the IntOGen Compendium of Cancer Genes, or with the UniProt "Tumor suppressor" keyword when IntOGen does not list the gene. Very strong when NMD is expected; Strong in the last exon when more than 10% of the protein is lost, otherwise Moderate; Moderate for a stop in the first 33 codons of exon 1; Strong at a canonical splice site (splice outcome not predicted); Supporting for start loss.
O3 · Absent or extremely rare in the general population
gnomAD v4.1 exomes and genomes. Moderate when the allele is absent; Supporting when its frequency is at most 0.001% in every genetic-ancestry group with more than 100,000 alleles and it is seen at most 1 time in smaller groups. Multi-base substitutions absent from gnomAD are not assessed (gnomAD splits them into single-base parts).
O4 · Recurrent in cancer cohorts
Samples with the allele in public cBioPortal studies (a stand-in for AACR GENIE, which the guideline names). Missense: Strong above 10, Moderate from 5. Truncating and in-frame: Strong above 50, Moderate from 20, Supporting from 10. When O7 comes from GDC residue counts, samples in studies that also feed the GDC are left out. When cBioPortal does not answer and O7 does not apply, GDC cases with the allele are counted instead.
O5 · Same residue as an established oncogenic change
A missense change at a residue where another missense change is on the canonical list. AlphaMissense of this change is compared with the best-scoring canonical change: Moderate when it is equal or higher, within 0.02, or both are at least 0.906; Supporting when it is at least 0.792.
O6 · Computational evidence supports a damaging effect
Missense changes with AlphaMissense at least 0.792 (ClinGen-calibrated Supporting), or BayesDel without allele frequency at least 0.13 when AlphaMissense has no score. Not used together with O2. Splice predictions are not used.
O7 · Mutation hotspot or critical functional domain
Hotspot-like recurrence from NCI GDC open-access mutations (a stand-in for Cancer Hotspots): the residue needs at least 10 cases with a missense change and at least 5 times the cases expected from the gene's missense cases spread evenly along the protein. Strong above 50 cases with at least 10 carrying this change, otherwise Moderate. Without a hotspot, a missense or in-frame change inside a listed critical domain gets Supporting.
O8 · Missense change in a missense-constrained gene
Missense changes in genes with gnomAD v4.1 missense Z of at least 3.09; not used when O7 applies.
O9 · In-frame indel or stop-loss changing protein length
In-frame insertions and deletions outside UCSC simple repeats and RepeatMasker elements, and stop-loss changes: Moderate.
O10 · Functional studies show an abnormal result
Needs a reading of published functional studies: chosen by hand.
O11 · Tumour phenotype consistent with the gene
Needs the tumour's own findings (for POLE exonuclease-domain changes, residues 268–471, a high tumour mutational burden): chosen by hand.
B1 · Common in the general population
gnomAD v4.1: stand-alone Benign above 1% and Strong above 0.1%, overall or in a continental group (afr, eas, nfe, amr, sas) with more than 2,000 alleles.
B2 · Does not fit the gene's mode of action (stand-alone VUS)
Needs knowledge of the gene in this cancer: chosen by hand.
B3 · Computational evidence does not support a damaging effect
Missense changes with AlphaMissense at most 0.169 (or BayesDel without allele frequency at most -0.18), outside a splice region.
B4 · Synonymous change
Synonymous changes: Supporting. Strong needs a splice prediction showing no effect, which is not used here, so Strong is chosen by hand.
B5 · In-frame indel in a repetitive region
In-frame insertions and deletions overlapping a UCSC simple repeat or RepeatMasker element: Supporting.
B6 · Functional studies show no damaging effect
Needs a reading of published functional studies: chosen by hand.
B7 · Tumour phenotype not consistent
Needs the tumour's own findings: chosen by hand.
Data sources
Allele counts and numbers in exomes and genomes, by genetic-ancestry group. Used for: O3, B1.
Released without restrictionGermline submission counts, somatic clinical impact and oncogenicity classifications for the exact allele. Used for: Tier IV benign consensus, tier evidence.
Public domain (US government)AlphaMissense and BayesDel (no allele frequency) scores. Used for: O5, O6, B3.
Scores in dbNSFP’s commercial build onlyEvidence items and assertions for the variant, its codon, exon and gene. Used for: Tier evidence, evidence table.
CC0Drug-response biomarkers for the allele and the same codon (via MyVariant.info, GRCh37). Used for: Tier evidence, evidence table.
CC0Devices whose biomarker covers the variant exactly, at its codon, a codon range, an exon class or a gene-wide class. Used for: Tier evidence, evidence table.
Public domain (US government)Samples with the allele and with any missense change at its residue. Used for: O4, evidence section.
Public study data; counts onlyCases with the allele; cases with a missense change at the residue and in the gene. Used for: O7, O4 fallback, evidence section.
Open-access dataConsequence, protein change, exon and protein length on the MANE Select transcript; GRCh37 position. Used for: All codes.
Open (Apache 2.0 software, open data)Simple repeats (TRF) and RepeatMasker elements over in-frame indels. Used for: O9, B5.
Free data tracksGene role (loss of function or activating) across cohorts. Used for: O2.
CC0"Tumor suppressor" and "Proto-oncogene" keywords (genes IntOGen does not list); critical domain coordinates. Used for: O2, O7 domain arm.
CC BY 4.0Tumour-type hierarchy (codes, names, main types, tissues). Used for: Tumour picker and matching.
CC BY 4.0Source substitutions
- AACR Project GENIE counts (O4)
- Public cBioPortal studies; NCI GDC when cBioPortal does not answer.
- Cancer Hotspots (O7)
- Hotspot-like residue recurrence counted in the NCI GDC, with an enrichment test against the gene’s missense load.
- REVEL (O5, O6, B3)
- AlphaMissense with its ClinGen calibration; BayesDel without allele frequency when AlphaMissense has no score.
- SpliceAI (O6, B3, B4)
- Not used. B3 is not applied in splice regions, and B4 Strong is left to the reader.
- OncoKB cancer gene list (O2)
- IntOGen roles, with UniProt keywords for genes IntOGen does not list.
- SpliceVault (O2 at splice sites)
- Not used: canonical splice-site changes get O2 Strong.
Sources not used: OncoKB, COSMIC, OMIM, REVEL, SpliceAI, CADD, PolyPhen-2, VEST, ClinPred, VarSome.
Limitations
- There is no sample: call quality (variant allele fraction, run recurrence) and possible germline origin are not checked, unlike in a laboratory report.
- Codes that need functional studies, the tumour’s own findings or knowledge of the gene in a given cancer (O10, O11, B2, B6, B7) are never applied automatically; readers can add them on the page.
- Evidence is matched by protein change on the MANE Select transcript. Changes named on other transcripts, and biomarkers written in free text, may be missed or matched too broadly.
- Tumour types written as free text are mapped to OncoTree by name; unusual names fall back to phrase matching, and evidence for a related type is counted as another tumour type.
- Public cohorts overlap and a sample may be counted in more than one study.
References
- ACGS SVIG-UK somatic variant interpretation guidelines v1.1 (26 May 2026)(opens in a new tab)
- Horak et al. 2022, Genet Med 24:986 (ClinGen/CGC/VICC oncogenicity SOP)(opens in a new tab)
- Li et al. 2017, J Mol Diagn 19:4 (AMP/ASCO/CAP somatic variant tiers)(opens in a new tab)
- Bergquist et al. 2025, Genet Med 27:101402 (AlphaMissense calibration)(opens in a new tab)
- Pejaver et al. 2022, Am J Hum Genet 109:2163 (BayesDel calibration)(opens in a new tab)