Variant classification that doesn't go stale.
43% of ClinVar submissions are Variants of Uncertain Significance — leaving patients in diagnostic limbo while evidence accumulates. Manual classification is slow, inconsistent, and doesn't monitor the literature. Skippy Variants applies all 28 ACMG/AMP criteria programmatically using the Tavtigian Bayesian framework, and alerts you automatically when new evidence warrants reclassification.
What a response looks like
Classification with Bayesian posterior, VUS reclassification alerts, and ancestry-stratified population frequency — all three in one API.
Predicted null variant (frameshift) in BRCA2, a gene where loss of function is an established mechanism of BRCA2-associated cancer.
Absent from gnomAD v4 controls (0 alleles in 807,162 chromosomes). Extremely low population frequency.
Multiple computational tools predict deleterious effect: CADD=37.2, REVEL=0.91, spliceAI 5′ score=0.04.
Simulated output representative of real API responses. Tavtigian 2018 Bayesian ACMG/AMP framework. CAP/CLIA documentation per variant.
Six capabilities. One API.
ACMG/AMP classification
All 28 criteria (PVS1, PS1–PS4, PM1–PM6, PP1–PP5, BA1, BS1–BS4, BP1–BP7) applied programmatically using the Tavtigian 2018 Bayesian framework. Every criterion documented with evidence source. Pathogenic / Likely Pathogenic / VUS / Likely Benign / Benign.
Population frequency lookup
gnomAD v4 allele frequency by ancestry cohort — European, African, East Asian, South Asian, Latino, Ashkenazi. Ancestry-stratified output for BA1/BS1/PM2 evidence, with ACMG signal annotation returned alongside frequency data.
gnomAD → ACMG signals
Convert raw gnomAD population frequency data directly into structured ACMG criterion signals. BA1 (>5% in any cohort), BS1 (elevated for disorder), PM2 (absent from controls) — returned as structured evidence codes ready for the classifier.
Reclassification monitoring
Continuous monitoring: when new ClinVar submissions, functional studies, or population data warrant reclassification, an alert is generated automatically. VUS → Likely Pathogenic notifications without waiting for annual manual review.
Bayesian posterior scoring
Compute the posterior probability of pathogenicity directly from ACMG criterion signals using the Tavtigian odds-of-pathogenicity product. Prior = 0.1. Thresholds: Pathogenic ≥ 0.99, Likely Pathogenic ≥ 0.90, Likely Benign ≤ 0.10.
Ancestry equity audit
Flag variants where frequency evidence is derived predominantly from European cohorts — a known source of classification bias. Returns per-ancestry allele counts and coverage gaps to support equitable variant interpretation.
All 28 ACMG/AMP criteria. Bayesian posterior.
Implements the Tavtigian 2018 Bayesian reformulation (PMID 29300386) of the Richards 2015 ACMG/AMP guidelines (PMID 25741868). Posterior = Prior × ∏(odds for met criteria). Prior = 0.1. Thresholds: Pathogenic ≥ 0.99 · LP ≥ 0.90 · LB ≤ 0.10 · Benign ≤ 0.01.
Predicted null variant (frameshift, nonsense, canonical splice) in a gene where LOF is the disease mechanism.
Same amino acid change as established pathogenic variant; de novo confirmed; well-established functional evidence; significantly elevated prevalence in cases vs controls.
Mutational hotspot; absent from gnomAD controls (PM2); in-frame deletion in non-repeat region; novel missense at known pathogenic residue.
Cosegregation in multiple family members; missense in low-benign-variation gene; multiple in silico tools predict deleterious.
Allele frequency >5% in gnomAD — stand-alone benign classification, no further criteria needed.
Greater than expected frequency for disorder; observed in healthy homozygous adult; well-established functional studies show no effect; in-frame deletion in repetitive region.
43% of ClinVar submissions are Variants of Uncertain Significance — leaving patients and families in diagnostic limbo while evidence accumulates in journals no lab has time to monitor.
A VUS classified today may be reclassified to Likely Pathogenic next quarter when a functional study is published. Continuous monitoring means labs are automatically notified when reclassification is warranted — without waiting for annual manual literature reviews.
“Fabric Genomics — the leading independent AI variant interpretation tool — was acquired by GeneDx in April 2025 and removed from the open market.”
Skippy Variants is the neutral, API-accessible alternative for genomics labs and clinical teams not operating on the GeneDx platform.
Labs, boards, and R&D teams
Built for clinical laboratory standards
ACMG/AMP criteria documented with explicit evidence sources per criterion code. Classification is traceable — not inferred — meeting CAP/CLIA documentation requirements for variant reporting in clinical genomics labs.
Somatic variant interpretation includes AMP/ASCO/CAP and OncoKB tiering with oncology-context evidence — relevant for FDA companion diagnostic and CDx-adjacent evidence generation.
HIPAA-ready with BAA available. Every classification produces an immutable audit record with all 28 criteria codes, sources, odds, and posterior probability.
Often deployed together
Novel variants need clinical interpretation before PGx diplotype assignment. Variants provides the upstream ACMG evidence; PGx translates it into prescribing action.
Genetic evidence from GWAS and ClinVar drives repurposing scores. Variants provides upstream interpretation for novel candidate variants feeding the discovery pipeline.
AMP/ASCO/CAP and OncoKB somatic tiering alongside ACMG germline classification — both available through the same API for comprehensive genomic reporting.
Skippy Variants is a decision-support tool for qualified genomics professionals. It does not replace laboratory director review, clinical validation, or physician interpretation. All variant classifications should be confirmed against current institutional guidelines and patient-specific clinical context before reporting. HIPAA-ready; BAA available.
See Variants in your genomics workflow
We work with genomics labs, tumor boards, and life sciences R&D teams. Let's talk about your variant interpretation problem.