Preprint describes SeverYX, a transparent ensemble framework for classifying variant severity from multiple predictors
SeverYX consolidates conservation, splicing, and functional-impact scores into a three-tier severity classification using a rule-based, effect-aware decision architecture benchmarked on over 270,000 ClinVar variants.
A preprint posted to bioRxiv introduces SeverYX, an ensemble framework designed to reconcile disagreements among computational variant effect predictors and produce a stable, three-tier severity classification — 'High', 'Medium', and 'Low' — for variants assessed in clinical genomics workflows.
A persistent challenge in variant interpretation is that individual predictors — covering sequence conservation, splicing impact, and functional consequences — frequently disagree with one another or return incomplete evidence. SeverYX addresses this through a transparent, rule-based decision architecture that the authors describe as 'effect-aware', meaning it weights predictor outputs differently depending on the likely functional mechanism of a given variant class.
The framework was benchmarked against a curated ClinVar dataset of 270,208 variants on MANE transcripts, comprising approximately 86,000 pathogenic or likely pathogenic variants and 184,000 benign or likely benign variants. The authors report performance metrics stratified by variant type.
SeverYX joins a growing family of ensemble and meta-predictor approaches — including those recently calibrated for ACMG/AMP evidence codes — aimed at making computational evidence in variant classification more reproducible and auditable. This preprint has not been peer-reviewed and the tool should be treated as a research prototype.
Sources
Read the original reporting — these are the public sources this summary draws from.
-
Primary sourcePreprint bioRxiv · 2026-11-04SeverYX: An ensemble framework for robust variant severity classification