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

Pathogenicity Scoring

ACMG Bayesian classification, CancerVar, EVE, gnomAD, ClinGen validity

Variants Classified
4 P + 1 LP
ACMG/AMP research classification (EZH2 V662A: VUS in ClinVar)
Highest ACMG Points
21 pts
IDH2 R140Q and SETBP1 G870S, tied (PS1+PS3+PM1+PM2+PP3+PP5)
CancerVar Tier I
IDH2 R140Q
OPAI=0.99, FDA-approved enasidenib
gnomAD Absent
3/5
EZH2, SETBP1, PTPN11 AC=0

Unified Scoring Table

Variant CADD 27
[27] P 2019
CADD: predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res (2019)
REVEL 28
[28] NM 2016
REVEL: An Ensemble Method for Predicting the Pathogenicity of Rare Missense Variants. Am J Hum Genet (2016)
AlphaMissense 19
[19] J 2023
Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science (2023)
ESM-2 LLR 18
[18] Z 2023
Evolutionary-scale prediction of atomic-level protein structure with a language model. Science (2023)
PrimateAI-3D CancerVar Tier ACMG Points Classification
DNMT3A R882H 28.5 0.742 0.9953 -8.383 0.869 Tier II (0.54) 20 Pathogenic
IDH2 R140Q 32.0 0.891 0.9872 -1.478 0.855 Tier I (0.99) 21 Pathogenic
SETBP1 G870S 28.7 0.716 0.9962 -9.804 N/A Tier II (0.99) 21 Pathogenic
PTPN11 E76Q 28.6 0.733 0.9972 -1.865 N/A Tier II (0.98) 15 Pathogenic
EZH2 V662A 28.1 0.781 0.9984 -2.966 N/A Tier II (0.91) 9 Likely Pathogenic
CADD, REVEL, AlphaMissense and ESM-2 read from acmg_aggregation.json as regenerated on 2026-11-03; CADD and REVEL are the Ensembl VEP values, except SETBP1 G870S REVEL, which VEP did not return and comes from myvariant.info. CADD: Combined Annotation Dependent Depletion (phred-scaled, ≥25.3 = ClinGen SVI PP3 threshold). REVEL: Rare Exome Variant Ensemble Learner (≥0.644 = ClinGen SVI PP3 threshold). AlphaMissense: DeepMind proteome-wide missense prediction (≥0.564 = likely pathogenic). ESM-2: masked marginal LLR (more negative = more disruptive). PrimateAI-3D: primate conservation + 3D structure (≥0.803 = damaging); 3/5 variants not in API. CancerVar: AMP/ASCO/CAP somatic classification with OPAI deep learning score.

ACMG Bayesian Points

Bayesian point system (Tavtigian et al. 2020) 26
[26] SV 2020
Fitting a naturally scaled point system to the ACMG/AMP variant classification guidelines. Hum Mutat (2020)
for ACMG/AMP variant classification. Pathogenic threshold: ≥10 points. Likely Pathogenic threshold: ≥6 points. Evidence aggregated from 11 sources.
Variant PS1PS3PM1PM2PM5PP3PP5 Total Class
DNMT3A R882H StrongStrongStrong--StrongStrong 20 Pathogenic
IDH2 R140Q StrongStrongStrongSupporting-StrongStrong 21 Pathogenic
SETBP1 G870S StrongStrongStrongSupporting-StrongStrong 21 Pathogenic
PTPN11 E76Q StrongStrongModerateSupporting-Strong- 15 Pathogenic
EZH2 V662A -SupportingModerateSupportingSupportingStrong- 9 Likely Pathogenic
Point values: Very Strong = 8, Strong = 4, Moderate = 2, Supporting = 1; in-silico evidence (PP3) is capped at Strong (ClinGen SVI, Pejaver et al. 2022). 4 variants reach the Pathogenic threshold (≥10 points); 1 is Likely Pathogenic (6 to 9 points). IDH2 R140Q and SETBP1 G870S score highest (21 pts), with PS1, PS3, PM1, PM2, PP3, PP5 evidence. EZH2 V662A scores lowest (9 pts) as a novel unreported variant with no PS1/PP5 evidence.

CancerVar AMP/ASCO/CAP Classification

CancerVar applies the AMP/ASCO/CAP 2017 consensus guidelines for somatic variant classification. The OPAI score comes from CancerVar's deep learning model, which predicts oncogenicity from both functional and clinical features (0.0 to 1.0). Tier I = strong clinical significance, Tier II = potential clinical significance.
Variant CancerVar Score Tier OPAI Score Therapeutic (CBP1) Prognostic (CBP3) Predictive Tools
DNMT3A R882H 10 Tier II 0.54 Supporting Supporting Strong
IDH2 R140Q 12 Tier I 0.99 Supporting Supporting Strong
SETBP1 G870S 10 Tier II 0.99 Supporting Supporting Supporting
PTPN11 E76Q 9 Tier II 0.98 Supporting - Supporting
EZH2 V662A 8 Tier II 0.91 Supporting Supporting Strong
CancerVar REST API (Li Q et al. Sci Adv 2022). The Therapeutic and Prognostic columns are CancerVar's automatic CBP1 and CBP3 criterion scores, not a statement that a usable therapy exists for the variant. IDH2 R140Q is the only Tier I variant (FDA-approved enasidenib 21
[21] EM 2017
Enasidenib in mutant IDH2 relapsed or refractory acute myeloid leukemia. Blood (2017)
). All 5 variants are clinically significant (Tier I or II).

gnomAD v4 Population Frequency

Variant gnomAD Status PM2 Strength Interpretation
DNMT3A R882H AC=427 alleles, AF=2.65e-4 Not_Met Present in gnomAD at AF=3.69e-04 -- does not meet PM2
IDH2 R140Q AC=71 alleles, AF=4.40e-5 Supporting Very rare in gnomAD (AF=6.58e-05, AC=71)
SETBP1 G870S Absent (AC=0) Strong Completely absent from gnomAD (0 alleles across exomes + genomes)
PTPN11 E76Q Absent (AC=0) Strong Completely absent from gnomAD (0 alleles across exomes + genomes)
EZH2 V662A Absent (AC=0) Strong Completely absent from gnomAD (0 alleles across exomes + genomes)
gnomAD v4.1.0, GRCh38. EZH2 V662A, SETBP1 G870S, and PTPN11 E76Q are completely absent from gnomAD (PM2_Strong). IDH2 R140Q is very rare (AC=71 alleles, AF=4.40e-5 exome + genome, PM2_Supporting). DNMT3A R882H is present at AF=2.65e-4 (AC=427 alleles, exome + genome), attributable to CHIP contamination in gnomAD blood-derived samples (PM2 not met).

ClinGen Gene-Disease Validity

Gene Disease Association Classification
DNMT3A Heyn-Sproul-Jackson syndrome Limited
Tatton-Brown-Rahman overgrowth syndrome Definitive
IDH2 mitochondrial disease Definitive
SETBP1 complex neurodevelopmental disorder Definitive
Schinzel-Giedion syndrome Definitive
PTPN11 Noonan syndrome Definitive
Costello syndrome Disputed
cardiofaciocutaneous syndrome Disputed
Noonan syndrome with multiple lentigines Definitive
EZH2 Weaver syndrome Definitive
ClinGen gene-disease validity curations from the ClinGen download of 2 Nov 2026 (exact rows, clingen_validity.json). Every curation is a germline, autosomal dominant disorder; ClinGen has no gene-disease validity curation for AML, MDS, JMML or CHIP for any of these 5 genes. Each gene has at least one Definitive germline curation. DNMT3A: 1 Definitive (Tatton-Brown-Rahman), 1 Limited. PTPN11: 2 Definitive (Noonan syndrome, Noonan syndrome with multiple lentigines), 2 Disputed (Costello, cardiofaciocutaneous).

EVE Evolutionary Variant Effect

EVE (Evolutionary model of Variant Effect) uses deep generative models trained on evolutionary sequences to predict pathogenicity 41
[41] J 2021
Disease variant prediction with deep generative models of evolutionary data. Nature (2021)
. EVE has a known limitation with gain-of-function variants: activating mutations at conserved positions may score as Uncertain because the model penalizes changes from the evolutionary consensus, but GoF mutations create novel functions not captured by conservation alone.
Variant EVE Score EVE Class (25th/75th) ESM-1b Score GoF Blind Spot
DNMT3A R882H 0.6197 Uncertain -12.728 -
IDH2 R140Q 0.8863 Pathogenic -13.606 -
SETBP1 G870S 0.7460 Uncertain -12.747 Yes (GoF)
PTPN11 E76Q 0.3068 Uncertain -7.606 Yes (GoF)
EZH2 V662A 0.7825 Uncertain -8.243 -
EVE scores from dbNSFP v4.x via myvariant.info, read from eve_scores.json. IDH2 R140Q is the only variant EVE classifies Pathogenic at its recommended 25th/75th-percentile threshold. EZH2 V662A scores 0.7825, which EVE calls Uncertain at that threshold and Pathogenic only at 60% and above; a much higher earlier value for this variant came from a wrong hg19 coordinate and was corrected in March 2026. PTPN11 E76Q scores lowest (0.307), illustrating the gain-of-function blind spot where experimentally confirmed GoF (DMS 99th percentile) is missed by conservation-based models.

popEVE Population-Calibrated Severity

popEVE (Orenbuch et al. 2025) calibrates EVE scores against population-level constraint data to produce a severity ranking. Scores below -5.056 indicate 99.99% likelihood of deleteriousness.

popEVE scores are not available via the myvariant.info API and require bulk download from the EVE model website. An earlier proxy severity ranking (EZH2 V662A highest, 0.957; PTPN11 E76Q lowest, 0.696) is superseded: it was built from older CADD, REVEL and AlphaMissense values and the withdrawn wrong-coordinate EVE value for EZH2, and needs a rerun before it can be used. These were never direct popEVE scores.

The discrepancy between EVE and experimental DMS data for PTPN11 E76Q (EVE Uncertain vs DMS 99th percentile GoF) underscores the importance of multi-tool concordance and experimental validation over any single predictor.
References
  1. Tavtigian SV et al. Fitting a naturally scaled point system to the ACMG/AMP variant classification guidelines. Hum Mutat (2020). DOI
  2. Richards S et al. Standards and guidelines for the interpretation of sequence variants. Genet Med (2015). PubMed
  3. Rentzsch P et al. CADD: predicting the deleteriousness of variants throughout the human genome. Nucleic Acids Res (2019). PubMed
  4. Ioannidis NM et al. REVEL: an ensemble method for predicting the pathogenicity of rare missense variants. Am J Hum Genet (2016). PubMed
  5. Cheng J et al. Accurate proteome-wide missense variant effect prediction with AlphaMissense. Science (2023). DOI
  6. Lin Z et al. Evolutionary-scale prediction of atomic-level protein structure with a language model. Science (2023). DOI
  7. Frazer J et al. Disease variant prediction with deep generative models of evolutionary data. Nature (2021). PubMed
  8. Stein EM et al. Enasidenib in mutant IDH2 relapsed or refractory acute myeloid leukemia. Blood (2017). PubMed
  9. Li Q, Ren Z, Cao K, Li MM, Wang K, Zhou Y. CancerVar: An artificial intelligence-empowered platform for clinical interpretation of somatic mutations in cancer. Sci Adv (2022). PubMed