How to Read Your Raw DNA File With Free Tools: Finding a Variant, Checking the Strand, and Knowing What the Result Can and Cannot Tell You
2026-11-22Same post, two depths. Switch any time; the research version has the enzyme names, lab markers, and doses. Written from a research and education perspective: a model of what may be going on and what the published literature suggests would address it if the model is right, not medical advice or a treatment plan.
If you have tested with AncestryDNA or 23andMe, you own a text file with somewhere between 600,000 and 700,000 of your genetic positions in it, and you can look up any of them yourself in a few minutes. This page walks through exactly how, with free tools, and then spends as much time on what the result does not mean as on what it does, because the second part is where people get hurt.
How do you get the file?
In AncestryDNA, go to DNA, then Settings, then Download DNA data; you will confirm by email and receive a zipped text file. In 23andMe, go to Settings, then 23andMe Data, then Download Raw Data. Both files are plain text. Unzip the file and keep it somewhere private; it is you, permanently, and it identifies your relatives too. Do not upload it to a site you have not researched. The free-report form on this site asks for it only when a report needs it, and the file is not kept.
What is in the file?
Open it in a text editor (Notepad on Windows, TextEdit on a Mac). After a block of comment lines you will see one line per position, with four or five columns: the rsID (a name beginning with rs, which is the universal label for that position), the chromosome, the position on that chromosome, and your genotype, written as two letters such as AG or TT. Ancestry puts the two letters in separate columns; 23andMe puts them together. Each letter is one copy, one from each parent, in no particular order.
Some lines show a dash, two dashes, or 00. That is a no-call: the chip could not read that position in your sample. It is not a result, and it is not rare; a few percent of positions are no-calls in every file.
How do you find a specific variant?
You need the rsID. Every variant discussed on this site, in the caffeine, alcohol and cannabis post here, the genes table here, and the reports, is given with its rsID. Open the file, press Ctrl+F (or Command+F), type the rsID, and read the two letters on that line. That is the whole technique. If the rsID is not in the file, the chip did not test it; the two companies test different sets, and newer chip versions have dropped many of the positions older ones had.
Which letters mean what?
This is the step that trips people up. A variant is described in the literature as, for example, "C677T" or "the A allele," but the letters in your file are reported relative to a reference strand of the DNA, and the literature sometimes reports the opposite strand. DNA is two strands, and A pairs with T and C with G, so a variant written as "T" in one source can appear as "A" in your file for the same position, with the same meaning.
The reliable way to check is dbSNP, the public database at the National Center for Biotechnology Information. Search the rsID there. The entry shows the reference allele and the alternate allele on the forward (plus) strand, which is the strand both consumer companies report. If your file shows letters that do not appear in the dbSNP alleles at all (for example your file says AG and dbSNP lists C and T), the source you are comparing against is on the other strand, and you flip: A becomes T and G becomes C. For MTHFR C677T (rs1801133), the risk "T" allele appears in both Ancestry and 23andMe files as A, because they report the opposite strand from the literature; a file showing AG is one copy, AA is two, GG is none. Confirming the orientation in dbSNP for each variant before you draw any conclusion is the single most important habit on this page.
How do you find out what a variant does?
Three free sources, used together, in this order.
ClinVar, also at NCBI, is the clinical database. Search the rsID or the gene name. Each entry carries a classification: pathogenic, likely pathogenic, uncertain significance, likely benign, or benign, with the condition and the number of submitters. Read the review status (the stars). One submitter with no assertion criteria is an opinion; expert panel review is a consensus. For most of the variants discussed on this site, ClinVar will say benign or will have no entry, because they are common variants that change how well a pathway runs, not disease mutations. That is expected, and the Start Here post here explains why those two categories are different.
gnomAD, from the Broad Institute, tells you how common the variant is in populations. A variant carried by 30 percent of people is not a rare disease mutation however alarming the name sounds; a variant seen in a handful of people worldwide deserves a different level of attention. Search the rsID and look at the allele frequency by population.
SNPedia is a community-edited encyclopedia of variants with the literature summarized and the orientation stated on each page. It is useful and uneven; treat it as a starting point for finding the papers, not as the verdict, and check the orientation note at the top of each page against dbSNP.
For the pathway-level reading, which is what this site is about, the question is not "is this variant pathogenic" but "what does this variant do to the enzyme, and what does that change about the nutrient the enzyme needs." The genes table answers that for the variants that matter most for nutrition.
What can the file not tell you?
A consumer chip reads a fixed list of common positions. It does not read the gaps between them, so it misses rare variants, small insertions and deletions, repeat expansions, large deletions and duplications, and most of the mutations that cause rare genetic disease. A file with no concerning variants is not a clean genome; it is a clean list of the positions the chip looked at. If there is a serious family history or a real suspicion of a genetic disease, the right test is clinical sequencing through a genetic counselor, not the consumer file.
Consumer chips also have an error rate. For common variants it is low, around one in a few hundred to a few thousand calls, which is fine for a pathway read and not fine for a decision about surgery or medication. Any result that would change a medical decision is confirmed with a clinical-grade test. The companies say this in their terms, and they are right.
Third-party sites that produce a long report of alarming findings from your file are reading the same common variants and dressing them in disease language. A report that lists thirty "mutations" is listing thirty common polymorphisms. The pharmacogenomic variants (the CYP enzymes, the warfarin and clopidogrel genes) are the exception where a consumer file can carry information a prescriber would want, and even there a clinical pharmacogenomic panel is the standard.
What does a worked example look like?
Suppose you want to know your caffeine clearance genotype. The variant is CYP1A2 rs762551. You search your file for rs762551 and find AC. In dbSNP, rs762551 lists alleles A and C on the forward strand, so no flip is needed. The A allele is the fast allele and C is slow; AC means you carry one slow copy. In gnomAD the C allele is carried by around 30 percent of people of European descent. ClinVar has no clinical entry, because this is a metabolism variant, not a disease variant. SNPedia summarizes the Cornelis study on heart attack risk in slow metabolizers who drink several cups a day. The pathway read: caffeine lasts longer in you than in a fast metabolizer, and afternoon coffee is more likely to cost you sleep. That is the whole chain, and it took five minutes.
Now the same chain for a variant that sounds frightening. You search for HFE rs1800562, the C282Y hemochromatosis variant, and find AG. dbSNP lists G and A; A is the variant. One copy. gnomAD shows the A allele in about 6 percent of people of Northern European descent. ClinVar classifies it pathogenic for hereditary hemochromatosis, but the condition is recessive and penetrance even in two-copy carriers is incomplete. The reading for one copy: a modest tendency to absorb more iron, a reason to know your ferritin and transferrin saturation before anyone gives you iron, and not a diagnosis of anything. The ferritin post here has the numbers.
What should you do with what you find?
Write down the rsID, your genotype as the file shows it, the dbSNP orientation check, and the pathway meaning from the genes table. Bring that to the conversation with your clinician rather than the third-party report, and bring the lab guide here with it, because the useful next step from almost any variant on this site is a blood test that shows whether the pathway is actually running short, which the gene alone can never tell you. A variant is a tendency; a lab result is a fact.
Educational content written from a research perspective, not medical advice. I am a PhD scientist, not a physician, and I do not diagnose, treat, cure, or prevent disease. Consumer genetic results are not diagnostic; results that would change medical decisions are confirmed with clinical testing through a genetic counselor or physician.
File structure
AncestryDNA and 23andMe export tab-delimited text files of roughly 600,000 to 700,000 genotyped positions on the Illumina Global Screening Array (current versions) or earlier custom arrays. Columns are rsID (or an internal identifier where no rsID is assigned), chromosome, position, and genotype, reported in build GRCh37 (hg19) coordinates by both companies as of this writing; 23andMe reports the genotype as a two-character string, Ancestry as two columns. Alleles are reported on the forward (plus) strand of the reference assembly. No-calls appear as "--", "00", or "0 0" and typically account for 1 to 3 percent of positions. Mitochondrial, X, and Y positions are included; hemizygous male X and Y calls appear as single letters or duplicated letters depending on vendor. Chip versions differ in content (23andMe v3, v4, v5; Ancestry v1, v2), and later versions dropped many clinically annotated positions; rsIDs absent from a file were not assayed.
Strand orientation
The most common interpretive error. Consumer files report the plus strand; literature, older papers, and some databases report the minus strand for a subset of variants, and a few rsIDs have had their reference flipped between dbSNP builds. Procedure: for each rsID, confirm the forward-strand alleles in dbSNP (the current build, with the GRCh37 and GRCh38 entries both shown); if the alleles in the file are the complement of the alleles in the source being used for interpretation, apply the complement (A to T, C to G). Ambiguous SNPs (A/T and C/G variants) cannot be resolved by complement checking alone and require the vendor's orientation convention and the dbSNP reference. Worked case: MTHFR rs1801133 (C677T) is reported on the minus strand in the literature; consumer files show the risk allele as A (plus strand), so AA is homozygous 677TT, AG heterozygous, GG wild type. SNPedia states the orientation on each page ("Orientation: minus" or "plus"); its "rs1801133(A;A)" notation follows the plus strand of the current dbSNP build and the page explains the mapping.
Annotation workflow
- dbSNP (NCBI): alleles, strand, position in both builds, global and population allele frequencies (ALFA, 1000 Genomes, gnomAD), and links to ClinVar and PubMed.
- ClinVar: clinical significance (pathogenic, likely pathogenic, uncertain significance, likely benign, benign, with conflicting interpretations flagged), condition, review status (zero to four stars; practice guideline and expert panel review at the top), submitter count, and assertion criteria. A single-submitter entry without assertion criteria is weak; an expert panel classification is authoritative. Most nutritionally relevant common variants are classified benign or have no entry, which is correct: they are hypomorphic polymorphisms affecting enzyme kinetics, not Mendelian disease alleles. The distinction between pathogenicity classification and functional effect is set out in the Start Here post here.
- gnomAD (Broad): allele frequency by ancestry, homozygote counts, and constraint metrics for the gene (pLI, LOEUF), which indicate whether loss of function is tolerated; a variant common in the population is not a rare disease allele regardless of nomenclature.
- In silico predictors (REVEL, CADD, AlphaMissense, SIFT, PolyPhen-2) for missense variants without clinical annotation; these are calibrated for rare-variant pathogenicity assessment and are of limited use for common hypomorphic variants, where functional enzymology (residual activity, Km shifts, thermolability) is the relevant evidence. The Start Here clinician version covers calibration and limits.
- SNPedia: literature summaries with stated orientation; useful for locating primary papers; community-edited and uneven in quality.
- PharmGKB and CPIC: for pharmacogenomic variants, the evidence levels and dosing guidelines; star-allele nomenclature (CYP2C192, CYP2D64) maps to specific rsIDs, and consumer arrays capture only a subset of defining variants, so array-based star-allele calls are incomplete (CYP2D6 copy number variation in particular is not captured).
- Pathway reading: the functional consequence for the enzyme and the nutrient or substrate it handles, which is the frame used in the genes table here and the reports on this site.
What array data cannot do
Arrays genotype predefined common positions; they do not sequence. Rare single-nucleotide variants, small insertions and deletions, repeat expansions (FMR1, HTT, C9orf72), copy-number variants, structural rearrangements, and most Mendelian disease alleles are not represented, and a file with no flagged positions carries no information about them. Imputation from array data (as performed by some third-party services) infers untested positions statistically with ancestry-dependent accuracy and should not be used for clinical inference. Genotyping error for common variants is low (concordance with sequencing over 99.5 percent for well-performing positions) but non-zero, and call quality is lower for rare and poorly designed probes; any result bearing on a medical decision is confirmed by clinical testing (Sanger or NGS in a CLIA or equivalent laboratory). Clinically relevant exceptions where array data have reasonable concordance and can usefully prompt confirmation: HFE C282Y and H63D, Factor V Leiden (rs6025), prothrombin G20210A (rs1799963), SERPINA1 S and Z alleles, APOE epsilon status (rs429358 and rs7412, which 23andMe historically reported and Ancestry does not), and the major pharmacogenomic SNPs.
Third-party interpretation services generate reports by matching file positions to curated lists; the outputs range from careful (pharmacogenomic and carrier-status-focused tools) to alarmist (long lists of common polymorphisms labeled as mutations with disease associations drawn from underpowered studies). The appropriate reading of any such report is: identify the rsIDs, re-annotate against dbSNP, ClinVar, and gnomAD, and discard associations resting on single small studies.
Worked examples
CYP1A2 rs762551. File: AC. dbSNP forward alleles: A, C; no flip. C is the lower-inducibility allele; AC is a slow-metabolizer heterozygote. gnomAD: C allele frequency about 0.3 in European ancestry. ClinVar: no entry. PharmGKB: level 2B for caffeine and cardiovascular outcomes. Pathway read: prolonged caffeine exposure; the caffeine, alcohol and cannabis post here.
HFE rs1800562 (C282Y). File: AG. dbSNP: G (reference), A (variant); no flip. Heterozygous. gnomAD: A frequency about 0.06 in Northern European ancestry. ClinVar: pathogenic, hereditary hemochromatosis, expert panel reviewed; autosomal recessive with incomplete penetrance even in homozygotes (clinical iron overload in a minority of C282Y homozygotes, more often men). Pathway read for a heterozygote: modestly increased iron absorption; measure ferritin and transferrin saturation before any iron supplementation; confirm clinically if saturation exceeds 45 percent. The ferritin post here.
MTHFR rs1801133 (C677T). File: AA. dbSNP: G (reference), A (variant) on the plus strand; the literature's "T" is the plus-strand A. Homozygous 677TT. gnomAD: A frequency about 0.3 in European and East Asian ancestry, lower in African ancestry. ClinVar: benign for the Mendelian phenotypes historically attributed to it, with the ACMG practice guideline recommending against MTHFR testing for thrombophilia. Pathway read: thermolabile enzyme with about 30 percent residual activity, FAD-dependent, rescued by riboflavin; homocysteine and RBC folate as the functional markers; the genes table.
Data handling
The file identifies the individual and relatives; store it encrypted, share it only with services whose data policies have been read, and understand that uploading to genealogical matching services makes relatives findable. The free-report form on this site uses the file for a single analysis and does not retain it.
Position
Consumer array data are adequate for reading common pathway variants and for prompting clinical confirmation of a handful of well-characterized disease and pharmacogenomic alleles; they are not adequate for excluding genetic disease or for making medical decisions without confirmation. The useful next step from nearly any array finding is a functional laboratory test, because a genotype describes a tendency and a biomarker describes the current state. The lab guide here carries the markers that correspond to each pathway.
Educational and research content, not medical advice. The author holds a PhD in cellular and molecular nutrition, is not a licensed physician, and does not diagnose, treat, cure, or prevent disease. Consumer genotyping is not diagnostic; clinically actionable findings are confirmed through a genetic counselor or physician in an accredited laboratory.
