Methodology
How DNA Layer builds your report
The deterministic evidence-matching methodology behind every DNA Layer report: the curated evidence panel, the clinical-grade safety floor, and where AI is and isn't involved.
Last reviewed 17 August 2026Deterministic matching first, always
Variant matching runs in deterministic software against a curated evidence panel, so the same raw DNA file always produces the same findings. This step never involves a language model — it is direct, repeatable software matching your file's genotypes against a curated panel of health-relevant variants.
Verified before explained
Only after a finding has been deterministically verified does a language model explain it in plain English. Language models are used for explanation, not discovery — they do not read your raw DNA file, and they cannot introduce a finding that the deterministic matching step didn't already verify.
The clinical-grade safety floor
Alongside the curated panel, every report is checked against clinical-grade sources as a safety floor: ClinVar (pathogenic/benign clinical significance), ClinGen Actionability (gene-condition risk), CPIC (prescribing/dosing guidelines), FDA pharmacogenomic labels, and PharmGKB (drug-gene interactions). This floor can only raise a finding's flagged severity, never lower it, and it never invents a finding that isn't already in your file — it exists so a clinically significant result is never understated.
Evidence and citations on every finding
Every finding in your layer carries its evidence strength, its sources, and its limitations alongside it. Citations are drawn outward from the finding's own variant — its own studies, population-level GWAS Catalog associations, and the clinical databases that catalogue it.
What this means for you
Most findings shift a probability rather than settle an outcome — lifestyle, environment, and chance all still apply. DNA Layer is for informational and educational use only, does not diagnose, treat, or prevent any disease, and important findings may require confirmation in a clinical laboratory and review by a qualified healthcare professional.
Frequently asked questions
How are my results generated?
Variant matching runs in deterministic software against a curated evidence layer, so the same file always produces the same findings. Language models only explain findings that have already been verified, never the raw file.
Does a predisposition mean I will develop a condition?
No. Most findings shift a probability rather than settle an outcome, and lifestyle, environment and chance all still apply. That is why evidence strength and limits sit next to every claim.