Methodology

How Varia works, how findings earn their labels, and how the catalog stays current.

Reading Preference

How Varia works

Varia performs client-side interpretation against a curated variant catalog, surfaces the controlling authority classification (ClinVar, CPIC, ACMG, ClinGen, PharmGKB) verbatim, applies its own evidence grade only where no authority covers the variant, and publishes a full editorial event log. Curation discipline and measurement integrity, not classifier novelty, are the product.

Everything runs in your browser, so your genome stays on your device. Out of the whole genome, Varia keeps only the findings that have earned their place in the science, and for each one it shows you the established authority behind the call. Every decision is logged.

How Varia works and the catalog

How Varia works

Varia is post-test interpretation software that reads genotype calls from a DNA file you already have from another testing system (a consumer chip export, a clinical lab, or whole-genome sequencing). Varia does not perform genotyping, sequencing, or any in vitro molecular detection. Varia reports population-level associations from published science, not per-user risk scores, percentiles, disease diagnoses, or screening directives.

You load a DNA file from your device and Varia returns curated findings drawn from peer-reviewed scientific literature and named institutional authorities. The product runs entirely in your browser. Your DNA file is never uploaded.

Varia is not a Genetic Health Risk Assessment System under 21 CFR 866.5950 because Varia performs no variant detection and emits no per-user risk output. Where relevant, FDA General Wellness guidance (January 6, 2026 reissue), the Cures Act software provision for displaying medical information (21 USC 360j(o)), and general-wellness device policy (21 CFR 880.6315) may offer additional context, but those frameworks have limits: General Wellness does not cleanly cover disease-risk surfaces, the CDS non-device line applies to health care professional-facing tools with independent review, and Varia is consumer-facing post-test interpretation software.

Three editorial commitments shape every finding Varia displays:

Curated. Varia's current archive represents 64 carefully chosen variants across 6 health areas. By contrast, Promethease ships 111,702+ entries unfiltered and 23andMe reports against an opaque internal catalog. Varia picks the smaller set deliberately, because each that we report on has been through a seven-pass editorial review and survives a documented evidence threshold.

Transparently sourced. Every finding cites its primary literature. Every authority Varia consults appears on the Institutional Authority References page with the date Varia last reviewed it. Every detector that watches for upstream changes (ClinVar, ClinPGx, ClinGen, PharmGKB, PharmVar, Retraction Watch, ACMG SF) runs on a documented schedule with results committed to the editorial event log in our public repository.

Editorially disciplined. Varia distinguishes findings that come with clinical authority (ACMG/AMP classifications, CPIC pharmacogenomic guidelines) from findings that come from Varia's own editorial assessment (common variants and lifestyle traits where no clinical authority exists). The framework is explicit. The reasoning is visible. The cadence at which findings get reviewed is public.

Varia is not a substitute for clinical genetic testing or genetic counseling. The recommendation boundary is bright: Varia surfaces variant names and biomarker or drug class names. It does not recommend doses, supplements, products, or protocols.

Varia is post-test interpretation software. It reads a DNA file you already have from another testing system (a consumer chip export, a clinical lab, or whole-genome sequencing). Varia does not run a new genetic test, does not perform sequencing in a lab, and does not give you a personal risk score or screening recommendation.

You open your DNA file from your device, and Varia analyzes it directly in your web browser. Your file never leaves your device.

Varia is not a Genetic Health Risk Assessment System under FDA rules because it performs no variant detection and no per-user risk output. FDA General Wellness guidance (reissued January 2026), the Cures Act exclusion for software that displays medical information (21 USC 360j(o)), and general-wellness device policy (21 CFR 880.6315) may add context in some cases, but Varia's primary posture is post-test interpretation of genotype calls you already hold.

Three editorial commitments shape every finding Varia displays:

Curated. Varia reports on 64 variants chosen with care, organized across 6 health areas. Other consumer-genomics products either show you every they find without filtering (Promethease shows over 111,000), or report on a much larger set without telling you why they picked those (23andMe). Varia keeps the list intentionally short because every variant has been reviewed through a seven-step quality process before earning a spot in the catalog.

Transparently sourced. Every finding Varia shows you links back to the scientific papers it draws from. The major medical authorities Varia consults (the same authorities your clinician uses) are listed openly with the date each was last reviewed. Computer programs check those authorities for updates on a published schedule, and every change gets logged in a public record so anyone can see what changed and when.

Editorially disciplined. When a clinical authority has already graded a variant (like for rare-disease risks or medication response), Varia shows you their grade directly. For variants where no clinical authority weighs in (common health-trait variants, lifestyle traits), Varia applies its own grading and shows the reasoning behind it. Either way, you see how the grade was reached and when it was last reviewed.

Varia is not a substitute for clinical genetic testing or working with a genetic counselor. The line Varia holds is clear: Varia tells you what variants you have and what published science says about them. Varia does not recommend specific doses of medication, supplements you should take, or lifestyle protocols you should follow.

The catalog

Varia tracks 64 variants across 6 health areas: Heart & Lipids, Brain & Cognition, Metabolism & Longevity, Medication Response, Cancer Risk, and Autoimmune. Each area groups related variants with sub-topics such as blood sugar metabolism, Alzheimer's clearance pathways, and pharmacogenomic drug response.

The catalog scope is intentional. Over the last two decades, genomic interpretation has been a bottleneck. Sequencing throughput scaled roughly six orders of magnitude in the same period, but the functional and clinical interpretation of variants grows linearly through peer-reviewed publication. Most of the human genome remains uninterpreted at the clinical-actionability level. A consumer-genomics product that surfaces every variant present in your file either oversells the actionability of those findings or hides the most defensible findings inside an unmanageable list.

Varia chooses the smaller, defensible list. Every variant in the catalog passes one of two qualification paths:

  • Authority-backed: the variant has an ACMG/AMP classification in ClinVar (Pathogenic, Likely Pathogenic, or VUS) with documented review status, OR a CPIC pharmacogenomic guideline at evidence level A or B, OR a PharmGKB clinical annotation at level 1A or 1B. For these variants, Varia surfaces the institutional grade verbatim and adds a confidence chip indicating ClinVar review status.
  • Varia-graded: the variant is a common-trait risk modifier (lipid metabolism, methylation pathway efficiency, inflammation response) or lifestyle association. No clinical authority grades these. Varia applies its own evidence assessment and labels the grade transparently with citations behind it.

HLA loci (HLA-B*15:02, HLA-B*57:01, HLA-DQ2 / HLA-DQ8 and similar) are deferred from the V1 catalog. Consumer chip imputation is unreliable for HLA, and whole-genome calling requires specialized pipelines beyond Varia's current scanner. When HLA returns, it returns with whole-genome-only restriction.

Varia tracks 64 variants across 6 health areas: Heart & Lipids; Brain & Cognition, which includes Alzheimer's risk and the 19 region surrounding APOE; Metabolism & Longevity, which folds in blood sugar, and detoxification, longevity and aging biology, and inflammation; Medication Response, how your body processes medications; Cancer Risk, common low-penetrance cancer associations from published genome-wide studies; and Autoimmune, common autoimmune associations with explicit limits on what consumer files can assess (including HLA).

The catalog size is intentional. Over the last two decades, our ability to sequence DNA has grown roughly a million-fold, but our scientific understanding of what each variant actually does has only grown step by step through peer-reviewed publication. Most of the human genome remains scientifically uninterpreted at a level that supports actionable medical decisions. A consumer-genomics product that shows you every variant in your file is either overselling what those findings can tell you, or burying the strongly-supported findings inside an unmanageably long list.

Varia chooses the smaller, defensible list. Every variant in the catalog earns its spot through one of two paths:

  • Authority-backed: a recognized medical authority has already published a grade for this variant. ACMG and ClinVar grade rare-disease risk variants. CPIC and PharmGKB grade medication response variants. When that authoritative grade exists, Varia shows it to you directly with a chip indicating how strong the consensus is.
  • Varia-graded: the variant is a common health-trait or lifestyle variant. No clinical authority grades these kinds of variants. Varia applies its own evidence assessment and shows the reasoning publicly.

HLA loci (a family of immune-system variants linked to drug allergies and some autoimmune conditions) are not in the V1 catalog. Consumer DNA chips can't reliably read these regions, and reading them from whole-genome sequencing requires specialized analysis software Varia doesn't yet run. When HLA returns, it will be available only for whole-genome sequencing files.

Varia's V1 catalog is intentionally small and editorially defended entry by entry. View the accepted variant catalog for the full rationale.

How findings get their labels

Each variant routes to its controlling authority (ClinVar/ACMG for pathogenicity, CPIC/PharmGKB for pharmacogenomics), surfaced verbatim; Varia's own strength grade applies only to common-variant complex-trait associations where no authority adjudicates. Grades are reproducible from published criteria and traceable to sources.

Varia does not make up its own verdicts. Where an expert body has already graded a variant, Varia shows that grade as published. Where none has, Varia adds its own evidence rating and says so. The hard part, and the point, is what gets left out.

The scanner

Varia ships a purpose-built, client-side genomic interpretation engine in the browser, not an off-the-shelf VCF parser wired to a lookup table. It matches variants by rsID with strand normalization and dbSNP merge-table resolution, auto-detects GRCh37 vs GRCh38 and rejects files when the build cannot be determined, phases compound diplotypes such as APOE ε2/ε3/ε4, and reports three-state coverage (tested-variant, tested-reference, not-tested) so absence is never silently miscalled. Input-confidence and provenance detection route low-confidence files to conservative summaries. Your file never leaves the device.

Varia's scanner runs entirely in your browser, and it's built to interpret your file, not just read it. It works out which genome build you have, handles multi-part findings like your APOE type, and tells you when a position wasn't tested rather than guessing. If a file looks like chip or imputed data, Varia weighs each finding more cautiously. Nothing uploads.

Staying current

Six detector channels monitor ClinVar, ClinPGx, Retraction Watch, PharmVar, and ACMG Secondary Findings on documented cadence. Alerts log to the public editorial event log; Eric consolidates genuine review items monthly.

Varia checks major medical authorities for updates on a published schedule and logs every change publicly. Once a month, items that truly need a human call roll into one review surface.

What Varia's results are and are not

Varia is a wellness-informational product. The Varia findings page describes variants from peer-reviewed scientific literature and surfaces classifications from named institutional authorities. The product is not a clinical diagnostic test, does not replace consultation with a qualified clinician, and is not authorized as a clinical decision support tool by the FDA.

Varia does not diagnose disease. Varia does not prescribe treatment. Varia does not replace clinical genetic testing.

Findings that display with clinical-grade authority chrome (ACMG classifications, CPIC guidelines) reflect the input data type your file represents. Clinical-grade interpretation of an ACMG Pathogenic variant requires confirmatory testing in a CLIA-certified laboratory and consultation with a clinician or genetic counselor. The Varia Genomic Brief that Varia generates is a non-directive literature and provenance summary for you to share with your clinician, not as a substitute for clinical evaluation.

The recommendation boundary is bright: Varia surfaces variant names and biomarker or drug class names. Varia does not recommend doses, supplements, products, lifestyle protocols, or treatment plans. When a finding suggests a clinical context (drug-gene pair, cardiovascular risk, Alzheimer's risk), the appropriate next step is a conversation with a clinician, not action on Varia's display alone.

Varia is a wellness-informational product. The Varia findings page describes what variants you have and what published science says about them, drawing on classifications from named medical authorities. Varia is not a clinical diagnostic test, does not replace your clinician, and is not approved by the FDA as a clinical decision tool.

Varia does not diagnose disease. Varia does not prescribe treatment. Varia does not replace clinical genetic testing.

Findings that display with clinical-grade visual confidence (ACMG classifications, CPIC guidelines) reflect what your file represents. If you have a rare-disease finding shown at clinical-grade confidence, confirmatory testing at a clinical laboratory plus consultation with your clinician or a genetic counselor remain the standard path. The Varia Genomic Brief that Varia generates is designed for you to share with your clinician, not as a substitute for the consultation itself.

The line Varia holds is bright: Varia tells you what variants you have and what published science says about them. Varia does not recommend specific doses of medication, supplements, products, lifestyle protocols, or treatment plans. When a finding suggests a clinical context (drug-gene interaction, cardiovascular risk, Alzheimer's risk), the appropriate next step is a conversation with your clinician, not action on Varia's display alone.

Citation standards

Varia cites primary literature from a and journal allowlist. Tier 1 includes high-impact peer-reviewed venues with sustained editorial reputation (NEJM, JAMA, Nature, Cell, Science, JCEM, Endocrine Reviews, Nature Immunology, Immunity, JEM, Circulation: Genomics and Precision Medicine, JACC, European Heart Journal, Neuron, Diabetes Care, and others). Tier 2 includes subspecialty venues with rigorous review and clinical or mechanistic depth (ATVB, Molecular Neurodegeneration, Translational Psychiatry, Neurobiology of Aging, and others). The full allowlist lives in the Varia conventions document and is enforced by the editorial process.

Citation freshness. When a primary citation is older than ten years, Varia notes that in the currency note. Older citations remain valuable for foundational mechanisms but get flagged so readers can weigh them against more recent work.

flagging. During editorial development, citations occasionally appear with provisional verification status before final PubMed verification completes. Provisional entries are explicitly flagged in the source-list display. The audit pipeline closes provisional flags through programmatic PubMed verification.

Conflict-of-evidence disclosure. When the literature is split on a variant association, Varia surfaces both sides through a conflict callout. Eleven named conflict types are in scope: failed replication, effect direction reversal, heterogeneity by ancestry, heterogeneity by phenotype, heterogeneity by sex, suspected publication bias, retracted supporting paper, withdrawn by authors, expression of concern, post-publication correction, and major methodology revision. Each labeled conflict appears in the finding card alongside the citations that contradict the main interpretation.

Varia draws from a list of scientific journals organized into two tiers. includes the highest-impact peer-reviewed venues with long-standing editorial reputations: the New England Journal of Medicine, JAMA, Nature, Cell, Science, the Journal of Clinical Endocrinology and Metabolism, Endocrine Reviews, Nature Immunology, Circulation, the Journal of the American College of Cardiology, the European Heart Journal, Neuron, Diabetes Care, and others. includes specialty venues with rigorous review and depth in their clinical or mechanistic areas: ATVB (cardiovascular biology), Molecular Neurodegeneration, Translational Psychiatry, Neurobiology of Aging, and others. The complete list is published in Varia's conventions document.

Citation freshness. When a scientific paper Varia cites is more than ten years old, Varia notes that on the finding. Older papers can still be foundational, but readers can weigh them against more recent work.

flagging. During editorial development, citations occasionally appear with provisional verification status before final verification through PubMed completes. Provisional entries are explicitly flagged so you can see which citations are still being checked. The audit pipeline resolves provisional flags through automated verification.

Disclosure when the literature is split. Sometimes scientific studies disagree about a variant's effect. When that happens, Varia surfaces both sides through a conflict callout on the finding card. Eleven named conflict types include failed replication, effect direction reversal, heterogeneity, publication bias suspected, retractions, withdrawals, expressions of concern, post-publication corrections, and major methodology revisions. Each conflict appears alongside the citations that contradict the main interpretation.

See also: The Genome, Medication Response, Sources.