How we turn raw data into plain-English interpretations.

Vogelview is deterministic first, AI second. Every finding starts from a curated knowledge base built on public research catalogs. This page describes what those catalogs are, how we grade the evidence, and where our interpretations stop.

Interpretation pipeline

Parse

A deterministic parser extracts variants (VCF, 23andMe TXT, AncestryDNA, MyHeritage, Nebula) or biomarkers (PDF, CSV, structured lab formats). Values are normalized into a canonical unit.

Match

Each variant/biomarker is matched against our curated knowledge base (built from the sources above) with exact, positional, and semantic matching.

Score

Matches carry an evidence tier (Strong / Moderate / Preliminary / Emerging) derived from the underlying study count, sample size, and replication.

Render

Quote → Meaning → Impact templates are filled with your actual data. A language guard strips diagnostic claims. Confidence and evidence indicators travel with every finding.

Evidence tiers

Every finding in a Vogelview report carries an evidence tier. The tier controls both our visual treatment (badge colour, confidence chip) and the language used in the interpretation. More evidence, more assertive copy. Less evidence, more caveats.

Strong

Backed by at least one large meta-analysis or multiple replicated cohort studies. Clinical guidelines often reference the association. We use assertive — but still hedged — language.

Moderate

Supported by multiple independent studies with consistent direction of effect, but the size of effect varies or replication is incomplete. We explain the trend and the uncertainty.

Preliminary

Based on one or two primary studies, or on association signals that have not yet been replicated. We surface the finding with clear caveats and never frame it as actionable in isolation.

Emerging

Interesting early signals we track but do not yet surface as a finding. Reserved for research curiosity, not clinical interpretation.

Source catalogs

We draw on publicly accessible, peer-reviewed databases. No single source is treated as authoritative; where sources disagree, we surface the disagreement and default to the most conservative interpretation.

GWAS Catalog

Curated repository of published genome-wide association studies maintained by EMBL-EBI and NHGRI. Our primary source for trait-variant associations with published effect sizes.

ClinVar

NCBI database of clinically reported variants with expert-panel and lab classifications. Used for clinical significance, especially for pharmacogenomic and rare-disease variants.

PharmGKB

Pharmacogenomics knowledge resource from Stanford. Source for drug-gene interactions, metabolizer phenotypes (CYP2D6, CYP2C19, CYP1A2, etc.), and CPIC dosing guidelines.

gnomAD

Genome Aggregation Database at the Broad Institute. We use it for population-frequency context — how common your variant is across the world and in specific ancestries.

dbSNP

NCBI Short Genetic Variations database. Reference for canonical rsIDs and nomenclature.

LOINC

Logical Observation Identifiers Names and Codes. Standard vocabulary for mapping lab test names across providers so your Quest, LabCorp, or hospital-lab PDF parses into the same canonical biomarker.

NHANES reference ranges

CDC population-level reference distributions used to contextualize biomarker values beyond single-lab reference intervals.

CPIC guidelines

Clinical Pharmacogenetics Implementation Consortium drug-gene guidelines. What we reference for pharmacogenomic actionability.

When sources disagree

Scientific literature is not monolithic. Two studies on the same variant can report different effect sizes; different labs set different reference ranges for the same biomarker. We handle disagreement in three ways:

Default to the conservative reading. If a finding could be Moderate or Preliminary, we tag it Preliminary. Users are better served by under-confidence than over-confidence.
Surface the disagreement. When sources meaningfully differ, the finding explains both positions rather than picking one silently.
Let you challenge us. Every finding has a feedback control. Flagged findings enter a manual review queue maintained by our team.

What Vogelview is not

Not a diagnostic tool. We do not diagnose any condition. Our language is educational and hedged by design. Findings are described as "may suggest", "tends to be associated with", or "can be linked to" — never as "you have" or "you will develop".

Not a replacement for a clinician. No finding should be acted on without discussing it with a licensed healthcare provider who can place the result in the context of your full history, other tests, and physical examination.

Ancestry-biased data. Most published genetic associations were studied in populations of European ancestry. Effect sizes derived from those cohorts may not transfer cleanly to other ancestries. We flag this on findings where the underlying studies are clearly ancestry-skewed.

Single-timepoint lab values. A blood test is a snapshot. Many biomarkers (cortisol, inflammation, iron) are highly dynamic. We contextualize rather than conclude from a single draw.

Researchers & clinicians

Auditing our interpretations, checking citations, or interested in partnership? Drop a line — we respond within a business day.

research@vogelview.health See also: Security & data handling →