AIstanbul Research Group
Clinical geneticists interpreting a patient's DNA constantly hit "Variants of Uncertain Significance" (VUS): changes where the evidence isn't enough to call them safe or harmful.
Core trick
Maps public variant evidence to ACMG criteria in deterministic code, then audits per-ancestry reference coverage so the language model only interprets results.
Tags
Description
Clinical geneticists interpreting a patient's DNA constantly hit "Variants of Uncertain Significance" (VUS): changes where the evidence isn't enough to call them safe or harmful. Existing tools aggregate that evidence and score it, but none ask the question that matters most: can this evidence be trusted for this specific patient?
vus-lens does. For any variant, it gathers public evidence, maps it to the ACMG rules deterministically, then audits its own confidence, flagging exactly where the evidence is unreliable, especially for Turkish and Middle Eastern patients who are barely represented in the reference databases medicine relies on. A language model reads all the evidence and reasons about it live, but a deterministic engine sets the classification: the AI interprets, it never decides.
We also measured its safety: across 1,277 known-pathogenic variants, it never once returned a benign call, while its benign logic provably fires on known-benign variants. It never gives false reassurance.
Built from scratch this week with Claude Code. Open-source. Live at vuslens.aistanbulresearch.com/live.
README excerpt
# VUS Confidence Auditor (`vus-lens`) Clinical-genetics decision support that aggregates the evidence for a genetic variant from **open public sources**, maps it to ACMG criteria **deterministically** (no LLM in the call), and then does what most aggregators don't: it **audits the reliability of its own evidence** and declares where that evidence is too weak to trust for *this* patient's ancestry. > **Decision support — not a classifier, not a diagnosis.** > The final interpretation, and the clinical decision, belong to the clinician. **Live demo:** <https://vuslens.aistanbulresearch.com/live> Prepared for the **2026 _Built with Claude: Life Sciences_ Hackathon**. **Demo Video:** <https://youtu.be/GDhIpS2ensk> --- ## What it does 1. **Deterministic ACMG engine (no LLM).** Frequency criteria (PM2 / BS1 / BA1) from allele frequency; in-silico (PP3 / BP4) from REVEL read on the canonical / MANE transcript, with AlphaMissense carried only as a Layer-2 cross-check (never scored into the class); ClinVar significance read directly. Same input always yields the same output, and every step is inspectable. 2. **Ancestry-aware confidence auditing.** When a frequency criterion is assigned, the tool reads the *actual* allele number for the patient's ancestry (gnomAD v4 Middle-Eastern `mid` group + the Turkish Variome) and flags — by the rule of 3 — when the sample is too small to support the call. This is the axis that tools built on Western reference data