An end-to-end account of the pipeline, the statistics and the reference databases that turn a raw DNA file into a set of personal reports.
102 Traits Scored
717 Drugs Analyzed
3,321 Conditions in Catalog
5,000+ Variants Detailed
A consumer DNA test produces a raw genotype file: a list of positions in the genome and the letters found at each one. The arrays in current use read between roughly 550,000 and 960,000 positions, and files from sequencing services carry millions. Our guides take a 23andMe file and an AncestryDNA file apart line by line. ReadYourDNA turns that file into five kinds of result.
Predisposition read from many variants at once, expressed as an ancestry-corrected score and a population percentile, for 102 complex traits from height and BMI to Alzheimer's and Type 2 Diabetes. Every published model kept for a trait is scored, and the models are averaged.
85 traits read directly from one or a handful of well-studied positions, among them lactose tolerance, eye colour, earwax type and coffee intake. The result is a genotype lookup rather than a score, and each card carries a confidence rating out of ten.
Two lanes. Medicines with a published CPIC or DPWG prescribing guideline are answered from star-allele diplotypes called by PharmCAT and the guideline is cited. Everything else is background research from PharmGKB, marked as such.
The uploaded data annotated against ClinVar pathogenic variants for inherited conditions, graded by a three-tier confidence system for how well supported each match is. Educational annotation, not a diagnosis or a carrier test.
Individually researched positions, each assembled from clinical significance, published trait associations, population frequencies, drug relationships and the papers behind them, and written up in plain language.
Two further views draw on the same processed genome. The karyotype places typed positions, carrier findings and the variants behind each trait score onto the 23 chromosome pairs at their real coordinates. The dashboard carries the genome-rarity score. A genome assistant answers questions by calling the same result files these pages read, so it can quote a result but cannot invent one.
Every upload runs the same pipeline, unattended, in five stages. The stages are very unequal in length, so the progress bar weights them by measured duration rather than counting them off evenly. The share beside each stage below is its measured portion of a complete run.
Format detection dbSNP Liftover 8% of the run
A raw DNA file is identified by its structure rather than by the brand on it, which is what makes every layout on the consumer market readable: one genotype column or two allele columns, tab-separated or comma-separated, quoted fields or plain, any of the common compression formats, plus VCF files from sequencing services. Positions are checked against the reference genome, converted into a normalized VCF, and short insertions and deletions are recovered. A file that is not genetic data, that covers fewer than 100,000 positions, or that touches fewer than fifteen of the numbered chromosomes is refused with the reason given, rather than analyzed into a report that would say "nothing found" because almost nothing was tested.
Three things are settled here because everything later depends on them. Genetic sex is read from Y-chromosome call rate and X-chromosome heterozygosity, since it changes how variants on X, Y and mitochondrial DNA must be interpreted. The reference genome has been revised several times and different tests are built on different editions, so the edition in use is detected and every position is translated onto one shared edition, with a second copy lifted to the newer edition for the pharmacogenomic caller. Ancestry is assigned at the end of the stage.
dbSNP gnomAD v4 10% of the run
On its own, a position in a genome is a pair of letters. Each one is given a standard identifier and confirmed reference and alternate alleles from dbSNP, and how often each allele occurs across world populations from gnomAD, a resource aggregated from more than 800,000 people. Those frequencies are the foundation of everything that follows: the expected polygenic score, the spread that score is judged against, how rare a genotype is, and the plausibility checks in carrier screening.
The same genetic difference can be written down in more than one valid way, so every variant is rewritten in a single canonical form against the reference. Without that step a variant in the uploaded file and the identical variant in a research database read as near-misses instead of a match. Insertions and deletions are stored alongside the reference sequence they were written against, because a genotype of "A/A" says nothing without it.
ClinVar gnomAD 4% of the run
The finished library is matched against a curated panel of ClinVar variants classified as pathogenic or likely pathogenic, covering a catalog of 3,321 conditions. Matching is strand-aware, and a single copy on the X chromosome in a male genome is treated differently from a single copy on an autosome. Every match is then graded for how well supported it is, rather than reported flat.
The grading is the part that matters, and it has its own section below. A well-reviewed rare variant should not be discarded for being rare, and a shaky match should not be presented as a finding.
PharmCAT 3.2.0 PharmGKB CPIC DPWG 1% of the run
Two engines run over the same genome on deliberately separate genes, so they can never contradict each other on the same result. PharmCAT reads star-allele diplotypes for 23 clinical pharmacogenes from the newer-edition copy of the data and applies CPIC and DPWG prescribing guidelines. A separate lookup covers single-variant associations from PharmGKB for every other gene, with those 23 genes excluded from it by name.
Three genes are supplied to the caller as outside calls, because it cannot derive them from array data on its own: mitochondrial MT-RNR1, an HLA-B*57:01 screen built from a tag variant, and a conservative CYP2D6 diplotype assembled from its common single-letter alleles. Each carries the limit of how it was made into the result.
PGS Catalog 77% of the run
Every published score kept for each of the 102 traits is computed against the genome and the results are averaged. That is 1,699 published score files in total, and it is roughly three-quarters of the wall-clock time, which is why the progress bar weights the stages by measured duration instead of counting them off evenly.
A compact orientation summary is written at the end of the run for the genome assistant, and the large intermediate files are deleted. What is kept is the processed genome and the result files each page reads.
How unusual a genetic result is depends on how common it is in the population a person comes from. A variant that is rare in one population is ordinary in another. A reference population is therefore chosen before anything is scored.
An allele frequency is the share of copies in a population that carry a particular version of a position. It is an input to nearly every number on the site: the score an average person is expected to carry, the natural spread that score is measured against, how rare a genotype is, and the plausibility checks that separate a real carrier finding from a mis-calling probe. Using the wrong population's frequencies distorts all four at once.
Find the markers shared between the uploaded file and the 1000 Genomes reference panel
Run principal component analysis over the combined data: the genome plus 2,504 reference individuals
Take the position of the genome in the top five principal components
Take the twenty nearest reference individuals in that space and let them vote
Assign the super-population with the most votes, and read every downstream frequency from it
An earlier version reported an ancestry mixture, by weighting the distance from the genome to the centre of each population. Those fractions were a geometric artifact rather than a measurement: no real genome sits exactly on a centre, so every genome came out admixed, and the fractions then re-weighted the allele frequencies used for scoring and carried the artifact downstream. A calibrated admixture estimate needs a supervised method that is not run here, so a single nearest population is the strongest claim the data supports. The vote is taken over the individual panel members rather than over population centres, because the admixed American cluster is not a sphere and a nearest-centre call misplaces it.
EUR
AFR
EAS
SAS
AMR
Latino / Admixed American
Reference panel: 2,504 individuals across 26 sub-populations
# Example ancestry output EUR:1.0000
A polygenic trait is one that many positions each nudge slightly, rather than one gene deciding it. A polygenic score adds those small effects into a single number. For each of 102 traits, every published score kept for that trait is computed separately and the results are averaged. Across the catalog that is 1,699 published score files; the median trait draws on nine of them and the largest on 101. Here is the arithmetic behind one of those scores.
STEP 1 - THE RAW SCORE
PRSyou = Σi βi · Gi
Each variant's published effect size (βi) multiplied by how many copies of the effect allele the genome carries (Gi = 0, 1 or 2), summed over the score. The effect sizes are deposited by the research teams that published them, in the PGS Catalog.
STEP 2 - THE EXPECTED SCORE
PRSexp = Σi βi · 2pi
The score an average person of the assigned ancestry would carry, where pi is that population's frequency for each variant, read from gnomAD. This anchors the result to the right baseline instead of to a global average.
STEP 3 - THE NATURAL SPREAD
σ = √( Σi 2pi(1−pi) · βi² )
How widely scores vary across a population: the genetic spread at each variant, weighted by its effect size. Numerator and denominator are measured over the same set, the variants the file was actually typed at, so a sparse array and a dense sequencing file are each judged against the spread expected for their own coverage rather than pushed toward the middle.
STEP 4 - THE ANCESTRY-CORRECTED SCORE
Z = (PRSyou − PRSexp) / σ
How many standard deviations one published score places a genome above or below the population average. Z greater than 0 is above-average genetic predisposition, Z below 0 is below average. This Z is computed separately for every published score behind the trait.
Published scores report their own directions, which do not all agree in sign, so the signs are aligned before the Z values are averaged into one ensemble result. Five traits carry a corrected sign, where the deposited direction runs against the phenotype the trait page names. Combining independent scores is more stable than trusting any single one of them.
A percentile expresses rank within a group. At the 76th percentile, 76 people in 100 carry a lower score and 24 carry a higher one. It is a position among people, not the probability of anything happening. The ensemble score is ranked against a European-ancestry reference distribution to read a percentile and a band off it. Because that distribution is European, a numeric percentile is served only where the assigned ancestry is European; every other genome is served the qualitative band instead, since the number would carry a precision the reference cannot support.
Very Low
≤ 10th percentile
Low
Average
High
Very High
> 90th percentile
The middle band is deliberately wide. Under equal fifths, a perfectly ordinary genome came out labelled Very Low or Very High on roughly 40 of its traits and Average on 20, which described the banding rather than the person, and Very High meant one person in five. The outer bands are now the top and bottom tenth.
85 traits on the site are not polygenic at all. Most are read from a single well-studied position - lactose tolerance (MCM6), earwax and body odour type (ABCC11), coffee intake (AHR) - and the result is a lookup rather than a score: the genotype is read and the write-up for that exact genotype is shown. Nothing is modeled, ranked or averaged.
A handful read several positions together, because one is not enough. Lactase persistence arose separately in Europe, the Middle East and Africa, so that card reads all three. Eye colour reads a switch and two modifiers through ordered rules, because the switch is recessive and summing would push brown-eyed readers into an intermediate band. Where several positions are combined, the linkage between them is measured on our own genomes first: two variants that travel together are one signal counted twice.
Genotype keys are reconciled against real array data and gnomAD reference alleles, because a strand difference between a chip and a database would otherwise invert a result silently. Evidence varies a great deal across these traits, so each carries a confidence rating out of ten computed from the evidence itself - how well the direction is established, how much of the trait the gene actually decides, replication, how closely the frequencies agree, chip coverage, effect size and whether the result varies between readers at all. The card shows every component and its reason, so a low number can be checked rather than merely trusted.
Pharmacogenomics is the study of how genetic variants change the way a person processes and responds to medicines. The evidence behind that field is of two very different strengths, so the results are served in two lanes that are never mixed: one carries answers backed by a published prescribing guideline, the other carries individual research findings and says so.
A star allele is a named version of a gene, defined by a fixed list of positions that tell one version from another. PharmCAT 3.2.0 reads those positions and calls a diplotype, the pair of versions inherited from each parent, for 23 clinical pharmacogenes.
A medicine appears in this lane only when an expert body has published a guideline for it against one of those genes: CPIC in the United States or DPWG in the Netherlands. That covers 105 medicines across 131 drug-gene pairs, and a medicine with no published guideline is kept out of this lane even when the gene behind it was read perfectly.
Guidelines are cited, not reproduced. They are written for prescribers and carry dosing detail that is only meaningful alongside the rest of a patient's circumstances, so each result links its source guideline instead of paraphrasing it.
A separate lookup covers single-variant associations curated by PharmGKB, each carrying that database's own evidence level from 1A down to 4. These are research findings about a variant and a drug, not a clinical recommendation, and the lane is labelled that way wherever it appears.
A direction is only presented as confident when at least two independent studies or two genes agree. A single unreplicated finding stays informational, and genuinely conflicting research is shown as Mixed rather than averaged into a false winner.
The 23 genes the guideline lane covers are excluded from this lookup by name, so the two lanes can never produce competing answers about the same gene.
Direction belongs to a drug-gene pair, not to a drug. Amitriptyline and tramadol are both handled by CYP2D6 and point opposite ways: a slow enzyme lets amitriptyline accumulate, and a slow enzyme makes less of tramadol's working form. Each of the 131 pairs is therefore curated onto one of four axes rather than derived from a property of the drug.
69 of the pairs. Enzymes such as CYP2D6, CYP2C19 and CYP2C9 decide whether a drug is broken down slowly, normally or rapidly. Slow clearance lets a standard amount sit higher and for longer; rapid clearance can empty it before it works. Four further pairs set a dose outright, as warfarin does through VKORC1 and CYP4F2.
53 of the pairs. Here the gene decides whether a serious reaction is likely, not how much drug is present: DPYD and fluoropyrimidines, TPMT or NUDT15 and thiopurines, G6PD and oxidative drugs, RYR1 and anaesthetic agents, HLA-B*57:01 and abacavir. Rendering any of these as more or less drug would be wrong.
Five of the pairs. The gene changes how well a drug works without changing how much of it is present. ABCG2 moves urate rather than allopurinol, so its guideline raises the dose for lost efficacy. CFTR decides whether ivacaftor has anything to act on. IFNL3 predicts response to interferon therapy.
Medicines in the guideline lane are sorted into six groups, and the groups double as the page's navigation. Two of them, "May not work as well" and "May affect you more strongly", always render even when empty, because on a quiet genome an absent section is indistinguishable from a check that never happened.
Check before taking
The gene is not adjusting how much of a drug reaches the body. It is flagging one the body may not be able to handle safely.
May not work as well
The enzyme clears the drug faster than average, or the drug needs that enzyme to switch it on and the enzyme runs slow, or the gene predicts response rather than exposure.
May affect you more strongly
The enzyme clears the drug more slowly than average, so a standard amount sits higher and for longer, or the drug needs that enzyme to switch it on and the enzyme runs fast.
Your genes disagree
Two genes govern the drug and they point in opposite directions, so there is no single answer to give. Worth raising precisely because it is not clear-cut.
Not covered by your chip
The gene the drug depends on is not among the positions the chip reads, or the gene was read but not the specific version this drug reacts to. No result either way, which is not the same as a clear one.
Nothing to change
For most, the governing genes were read and came back typical. For a few, a gene is not typical but its effects cancel out. This is the group a reader would take reassurance from, so it is the group that carries the coverage note.
A star-allele call is made from a fixed list of positions that tell one version of a gene from another: 18 of them for NUDT15, which gates the thiopurines, 171 for G6PD, 313 for RYR1. No consumer array carries all of them, and how many were read varies from chip to chip, so the count is stated on each result rather than claimed in general. Unread positions cannot un-find a variant, so a positive finding is worth exactly as much at partial coverage as at full coverage. Only a reassuring answer is weakened by it. Coverage is therefore attached to results that report nothing to act on, and it names the least-covered gene behind the answer rather than an average, because an average advertises the coverage of the gene that did not produce it.
Nineteen of the 23 genes have such a list. The remaining four are supplied to the caller from outside and have none, so they carry no count at all: their limits are different in kind and a number cannot express them. CYP2D6 is reconstructed from its common single-letter alleles and never assesses copy number, whole-gene deletion or duplication. HLA-B is a screen for the *57:01 allele through one tag variant, which resolves abacavir and nothing else: the other HLA-B alleles that carbamazepine, allopurinol and dapsone react to are not typed, and those medicines are reported as unread rather than clear.
THE 23 PHARMACOGENES CALLED
CYP2D6 * CYP2C19 CYP2C9 CYP2B6 CYP3A4 CYP3A5 CYP4F2 DPYD TPMT NUDT15 UGT1A1 SLCO1B1 VKORC1 ABCG2 G6PD NAT2 HLA-B * HLA-A * MT-RNR1 * RYR1 CACNA1S CFTR IFNL3
* Supplied to the caller from outside the VCF, or not readable from array data at all. A gene whose defining positions are absent from the chip returns “not determined”, which records a missing measurement and is never rendered as a normal result.
Evidence basis: PharmCAT · CPIC · DPWG · PharmGKB - plus single-variant associations from PharmGKB for genes outside the 23, for example F5 and COMT.
No dosing text, anywhere. These pages carry no dose, no schedule and no instruction to start, stop or change a medicine, in either lane. A separate summary can be generated for a clinician instead: the diplotype calls, the variants behind each one, an explicit statement of what this kind of data could not assess, and the guideline citation for every medicine. It carries no advice either, because what a prescriber needs is the measurement and its source, not a paraphrase of their own guideline.
A carrier has one copy of a variant for a recessive condition. That person is usually unaffected, but can pass the variant on, and a child inherits the condition only if both biological parents pass on a variant in the same gene. The uploaded data is annotated against pathogenic variants curated in ClinVar, covering a catalog of 3,321 conditions across 3,825 genes, and every match is graded for how well supported it is. This is educational annotation, not a diagnosis and not a carrier test.
The panel starts from ClinVar variants classified as pathogenic or likely pathogenic, with concrete alleles and adequate expert review, enriched with population frequencies from gnomAD. Conditions that follow a dominant pattern are labelled as dominant rather than dressed as harmless carrier status, because a single copy in a gene such as BRCA1 is not a carrier result at all.
Variants too common to be genuinely pathogenic, above roughly 1% frequency in the reference population, are removed, with one reviewed exception list: well-established pathogenic founder alleles that are common in Europeans, such as the main cystic fibrosis deletion (CFTR F508del), the main recessive hearing-loss deletion (GJB2 35delG), the two haemochromatosis alleles (HFE C282Y and H63D), the prothrombin G20210A variant and the alpha-1 antitrypsin S and Z alleles, are kept and carry their own caveat on the card. Variants that turn up far more often than their published frequency implies are flagged as probable genotyping-chip artifacts rather than findings: that pattern is the signature of a mis-calling probe, not of a rare allele. Palindromic positions, where a chip cannot tell which DNA strand it read, are flagged for the same reason.
Genotypes are matched to the surviving variants with strand awareness. One copy of a pathogenic-classified variant is annotated as a variant found, since for most recessive conditions a single copy is typically harmless. Two copies, or a single copy of an X-linked variant in a male genome, is flagged for clinical confirmation. The annotation records what was matched; it does not determine that someone is or is not a carrier.
Each finding is sorted into confident, plausible or flagged. Confident requires all of: a ClinVar review level of at least three stars, a population frequency high enough for a chip probe to be reliable, and a clean pass through the artifact checks above. The tiering exists so that a well-reviewed rare variant is not discarded for being rare, and a shaky match is not presented as a result.
Array limitation: screening is restricted to pathogenic variants that are physically present on the genotyping chip. Whole-exome or whole-genome sequencing is required for comprehensive carrier screening, and any positive finding should be confirmed with a clinical-grade test. See the Limitations section.
Beyond the summary reports, the site holds a library of 5,000+ individually researched positions, a genome-wide measure of how unusual a combination of variants is, and a view that places all of it on the chromosomes it came from.
Each entry is built from an evidence pack assembled for that one position: identity from dbSNP, clinical classification from ClinVar, trait associations from the GWAS Catalog, drug relationships from PharmGKB, frequencies from gnomAD, and the papers behind them. Every numbered citation is checked back against that pack before publishing. A reliability tier records how well a chip can call the position at all.
Across a fixed panel of independent markers, each genotype is scored for how surprising it is given how common it is in the assigned ancestry, and the scores are combined into one figure ranked to a percentile. The panel is pruned so that nearby positions travelling together are not counted twice. The result sits on the dashboard.
The 23 chromosome pairs are drawn with their real banding pattern, and typed positions, carrier findings, gene positions and the variants behind each trait score are placed on them. Everything is drawn in one genome edition: because trait scoring runs in an older one, positions are translated by identity rather than by coordinate first.
ReadYourDNA builds on eleven public genomics resources. Each card names the release in use where that release is pinned. Reference data is refreshed by rebuilding the analysis pipeline rather than continuously, so a database update reaches a set of results the next time that file is processed.
PGS Catalog Trait scores
Published polygenic score weights. 1,699 score files across the 102 traits scored here.
gnomAD v4 Allele freq
Population allele frequencies aggregated from over 800,000 individuals across diverse ancestries, used for frequency, ancestry weighting and rarity.
ClinVar Carrier
NCBI's curated database of human variants and their clinical significance, with the review level attached to each classification.
dbSNP Annotation
NCBI's reference catalog of known variants. Resolves identifiers, confirms reference and alternate alleles, and recovers indels.
1000 Genomes Ph3 Ancestry
2,504 individuals across 26 populations. The reference panel the ancestry assignment votes over.
PharmCAT 3.2.0 Medications
Open-source engine that calls star-allele diplotypes and applies CPIC and DPWG prescribing guidelines. Bundles current guideline and nomenclature data.
PharmVar Medications
The star-allele nomenclature the diplotype calls are written in, distributed inside PharmCAT.
CPIC Guidelines
Clinical Pharmacogenetics Implementation Consortium. Evidence-based prescribing guidelines per gene-drug pair, released under CC0.
DPWG Guidelines
Dutch Pharmacogenetics Working Group. The second guideline body behind the medication results, distributed through PharmGKB.
PharmGKB Medications
Curated variant-drug knowledge base. Supplies the background-research lane and the 1A to 4 evidence level on each association.
GWAS Catalog Variants
EMBL-EBI's curated collection of genome-wide association results across thousands of traits.
Data licensing & attribution. Pharmacogenomic associations are derived in part from PharmGKB, used under the Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0). The source data was modified, by re-mapping phenotype names, normalizing DNA strand and re-scoring evidence for display, and those changes are our own, not endorsed by PharmGKB. Star-allele diplotypes are called with PharmCAT (Mozilla Public License 2.0), implementing CPIC and DPWG guidelines (CPIC content is released under CC0) and PharmVar nomenclature. ClinVar, dbSNP, gnomAD, the GWAS Catalog, the PGS Catalog and 1000 Genomes are used under their respective open-data terms.
Scientific credibility requires being clear about what these analyses cannot tell you. Please read these carefully before interpreting any result.
A polygenic score measures relative genetic predisposition against a population. A high score means elevated genetic predisposition, not certainty; a low score is not protection. Genetics is one part of the picture and the size of that part varies enormously: across the traits scored here, heritability runs from roughly 15% to 90%, and published scores capture only a fraction of that genetic component.
Every trait page carries an Evidence rating that describes the published science behind the trait, not the reader's DNA. Strong means the trait has been measured in large, well-replicated studies and the result draws on several independent published models. Moderate means reasonably well studied, with fewer or smaller models behind it. Limited means few published scores exist, often only one, and those results are a rough indication rather than a settled estimate. The rating says nothing about how extreme a particular score is: a very high percentile on a Limited trait is still a weakly evidenced number.
The assignment names the single reference population a genome sits closest to, and every allele frequency downstream comes from that population. For a genome with substantial ancestry from more than one group, that is an approximation, and the further a genome sits from the group it was assigned to, the rougher the frequencies behind its scores become. Calibrated admixture proportions need a supervised method that is not run here, and reporting fractions this method cannot support would be worse than naming one population.
Most large genetic studies to date were conducted in people of European ancestry, so both the published score weights and the reference distribution behind the percentiles are most accurate there. Every percentile is therefore a rank against a European-ancestry reference population, and each trait states this in its Study Details. The further an ancestry is from that reference, the rougher the estimate - a disclosure, not a correction: the underlying research imbalance cannot be removed.
Unread positions cannot un-find a variant, so a positive finding is worth the same whether a gene was read completely or partly. A reassuring result is the opposite: it is only as good as the coverage behind it. That asymmetry is why coverage is stated on results that report nothing to act on, and why an absent measurement is labelled as not determined rather than as normal, on every surface where the distinction exists.
Consumer DNA chips read roughly 550,000 to 960,000 positions, which is around 0.02% of the 3.2-billion-letter genome. Most rare variants, large structural changes and repeat expansions are not captured at all. Some insertions and deletions are recovered through dbSNP, but coverage remains limited to what the chip physically measures. Sequencing files carry far more positions and are analyzed the same way.
CYP2D6 depends on gene copy number, whole-gene deletions and duplications that array data cannot see, so only its common single-letter alleles are called and the rest is left unassessed. HLA-B is screened for the *57:01 allele through a single tag variant, which is not HLA typing and says nothing about the other HLA-B alleles that other drugs react to. Both should be confirmed by clinical testing before any decision.
Comprehensive carrier screening requires full exome or genome sequencing. An array-based approach misses pathogenic variants that are not on the chip and cannot reliably detect certain variant types, for example repeat expansions in FMR1 for Fragile X or deletions in SMN1 for spinal muscular atrophy. A result with no findings does not rule out genetic risk.
ReadYourDNA reports are intended for informational and educational purposes. They do not constitute medical advice, diagnosis or treatment. Do not make health decisions based solely on this data. Consult a qualified healthcare provider or certified genetic counselor before acting on any genetic result.
Educational use only. ReadYourDNA reports are for informational and educational purposes and do not constitute medical advice, diagnosis, or treatment. Do not make clinical decisions based solely on these results. Always consult a qualified healthcare provider or certified genetic counselor.
Every analysis on this page runs on the file you upload and comes back as a plain-language report.
Raw DNA Data Reports for 23andMe & AncestryDNA · About ReadYourDNA · FAQ · Methodology & Science · Pricing · Support and complaints · Legal & Contact · Health Traits · Pharmacogenomics · Carrier Status · Variant Explorer · The Human Karyotype · Privacy Policy · Terms of Service · Accessibility · Consumer Health Data Privacy · Your Privacy Choices · Editorial Standards · Your Data, Checkable · Sample Report · ReadYourDNA and Promethease compared · What is inside a 23andMe raw data file · What is inside an AncestryDNA raw data file · What a chatbot can and cannot do with your raw DNA file · Where a raw DNA file can still be uploaded in 2026
Written, maintained and operated by Martin Pechhacker, sole proprietor, Vienna, Austria. Trade register GISA 40064420. support@readyourdna.com