Transparent Implementation and Benchmarking of ACMG/AMP Variant Classification in fastVEP

Background: The ACMG/AMP framework is the standard for germline sequence variant interpretation, but applying it in scalable automated software requires operational choices that the original framework and later ClinGen guidance do not always specify. We implemented an ACMG/AMP classification engine within fastVEP and evaluated which implementation choices improved agreement with expert-curated classifications without increasing concerning opposite-direction calls. Methods: The classifier implements all 28 ACMG/AMP criteria and can evaluate 15 from a sites-only VCF using nine public evidence sources. For selecting the optimal classifier, we compared symmetric and hybrid evidence-combining schemes, pathogenic class cutoffs, population-frequency thresholds, and benign-evidence thresholds against 673,660 ClinVar variants with review status of two stars or higher from April 2026. Clinical geneticists reviewed opposite-direction calls. We then applied the selected final classifier unchanged to 116,078 variants that newly reached the two-star threshold in the October 2026 ClinVar release. Results: The final fastVEP ACMG/AMP classifier uses the Tavtigian point scale on the pathogenic side and the 2015 ACMG/AMP combining table on the benign side. Using the 2015 table on both sides reduced pathogenic-side recall from 62.5 to 57.1 percent and lost 4,900 same-direction calls while avoiding 10 opposite-direction pathogenic calls. Using points on both sides increased benign-side recall from 66.4 to 77.0 percent, but increased opposite-direction benign calls from 27 to 399 and reduced recall of ClinVar uncertain variants from 97.1 to 69.5 percent. The 6-point Likely pathogenic cutoff was retained because lowering it to 5 points increased pathogenic-side recall to 73.9 percent but also classified 6,003 additional ClinVar uncertain variants and 17 additional ClinVar benign-side variants as pathogenic-side. The final classifier reproduced ClinVar's exact five-tier class for 431,727 variants or 64.1 percent and its pathogenic, benign, or uncertain direction for 534,335 variants or 79.3 percent. Only 51 calls or 0.0076 percent were opposite in direction to ClinVar. On the later variants, same-direction agreement was 72.6 percent and 20 calls were opposite in direction. Conclusions: The rules used to combine evidence affected classification more than several individual numeric thresholds. The hybrid scheme increased pathogenic-side recall while preventing a single Supporting benign criterion from producing a directional call. This conservative operating profile is suited to large-scale variant triage because it makes high-confidence calls while leaving variants uncertain when the available evidence does not support expert-level resolution. fastVEP is a research and decision-support tool and does not replace case-specific expert review.

Authors

Institutions

Publication Details

Journal
medRxiv
Published
2026-11-18
DOI
https://doi.org/10.64898/2026.11.16.26368228
Primary Topic
Genomics and Rare Diseases
Type
preprint

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Transparent Implementation and Benchmarking of ACMG/AMP Variant Classification in fastVEP

Kuan‐lin Huang, Mythily Ganapathi, Wendy Chung
3 links
medRxiv
Genomics and Rare Diseases
preprint

Transparent Implementation and Benchmarking of ACMG/AMP Variant Classification in fastVEP

Kuan‐lin Huang, Mythily Ganapathi, Wendy Chung
preprint en

Abstract

Background: The ACMG/AMP framework is the standard for germline sequence variant interpretation, but applying it in scalable automated software requires operational choices that the original framework and later ClinGen guidance do not always specify. We implemented an ACMG/AMP classification engine within fastVEP and evaluated which implementation choices improved agreement with expert-curated classifications without increasing concerning opposite-direction calls. Methods: The classifier implements all 28 ACMG/AMP criteria and can evaluate 15 from a sites-only VCF using nine public evidence sources. For selecting the optimal classifier, we compared symmetric and hybrid evidence-combining schemes, pathogenic class cutoffs, population-frequency thresholds, and benign-evidence thresholds against 673,660 ClinVar variants with review status of two stars or higher from April 2026. Clinical geneticists reviewed opposite-direction calls. We then applied the selected final classifier unchanged to 116,078 variants that newly reached the two-star threshold in the October 2026 ClinVar release. Results: The final fastVEP ACMG/AMP classifier uses the Tavtigian point scale on the pathogenic side and the 2015 ACMG/AMP combining table on the benign side. Using the 2015 table on both sides reduced pathogenic-side recall from 62.5 to 57.1 percent and lost 4,900 same-direction calls while avoiding 10 opposite-direction pathogenic calls. Using points on both sides increased benign-side recall from 66.4 to 77.0 percent, but increased opposite-direction benign calls from 27 to 399 and reduced recall of ClinVar uncertain variants from 97.1 to 69.5 percent. The 6-point Likely pathogenic cutoff was retained because lowering it to 5 points increased pathogenic-side recall to 73.9 percent but also classified 6,003 additional ClinVar uncertain variants and 17 additional ClinVar benign-side variants as pathogenic-side. The final classifier reproduced ClinVar's exact five-tier class for 431,727 variants or 64.1 percent and its pathogenic, benign, or uncertain direction for 534,335 variants or 79.3 percent. Only 51 calls or 0.0076 percent were opposite in direction to ClinVar. On the later variants, same-direction agreement was 72.6 percent and 20 calls were opposite in direction. Conclusions: The rules used to combine evidence affected classification more than several individual numeric thresholds. The hybrid scheme increased pathogenic-side recall while preventing a single Supporting benign criterion from producing a directional call. This conservative operating profile is suited to large-scale variant triage because it makes high-confidence calls while leaving variants uncertain when the available evidence does not support expert-level resolution. fastVEP is a research and decision-support tool and does not replace case-specific expert review.

medRxiv
Boston Children's Hospital (US), Columbia University (US), Icahn School of Medicine at Mount Sinai (US)
American Cancer Society
Good health and well-being
Genomics and Rare Diseases
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon