AI-powered prioritization and interpretation of rare disease variants

VarSage

VarSage is built to fill the space between raw variant files and human clinical reasoning: filtering, ranking, explaining, and documenting candidate variants without pretending that software replaces expert review.

by Milad EIDI

A clinical genetic researcher, Montpellier, France

VarSage began from a practical need: facilitate variant reviews on research based patients, more auditable for the people doing the work.

The project reflects a clinical-genomics mindset: keep evidence domains separate, preserve provenance, show uncertainty plainly, and make every conclusion available for human review.

Why it exists

Variant interpretation can become a maze of annotation fields, inheritance assumptions, population frequency thresholds, phenotype terms, and report wording. VarSage turns that maze into a structured review path.

AI can be enabled for selected steps, but its role remains bounded: support summarization and review prioritization, never issue a diagnosis or laboratory classification.

Design principles

Evidence firstPhenotype fit, inheritance, quality, population data, assertions, and ACMG-oriented evidence are kept visible and distinct.

Local by defaultRuns are designed around local files and local reports, with optional external AI only when configured and enabled.

Review, not replacementOutputs are decision support for trained experts, with caveats and confirmatory actions kept in view.

What VarSage helps with

Rare-disease prioritization, HPO-driven phenotype matching, optional AI HPO extraction, Jannovar annotation, BED-region filtering, coordinate annotation database merging, provisional ACMG/AMP evidence summaries, and exportable HTML, TSV, Word, and carrier-screening review reports.

Open VarSage Knowledge Base Changelog About Contact Privacy Policies