Which DNA test can actually read your medication genes? We analysed the chips
Your DNA raw data can hint at how you handle some medications — but how much depends enormously on which test you took. We analysed the actual genotyping arrays behind the major consumer tests against all 1,210 variant positions that CPIC (the clinical pharmacogenomics standard) uses to define medication-gene alleles. The results: the 23andMe-base chip physically carries 47% of them — the AncestryDNA/MyHeritage base carries 3% — and some of the highest-stakes genes can't be read by any chip.
This is general educational information, not medical advice. Consumer DNA markers are statistical associations, not diagnoses, and are not a pharmacogenomic test — confirm anything that matters with clinical-grade testing and a professional.
Key findings
- The chip behind your test decides what's readable. Of CPIC's 1,210 pharmacogene-defining positions, the GSA v3 array (the family 23andMe v5 is built on) carries 570 (47%). The OmniExpress array (the family AncestryDNA v2 and MyHeritage are built on) carries 40 (3%) — it's an ancestry-era design, not a clinical one.
- Positions ≠ star alleles. Guideline-grade pharmacogenomics calls star alleles, each defined by a complete set of positions. Of CPIC's 1,330 named alleles, the GSA carries complete definitions for 552 (42%); the OmniExpress base for 33 (2%).
- Famous SNPs are there; full calls usually aren't. Example: CYP2C19 *2's headline variant rs4244285 is on the GSA — but *2's complete CPIC definition spans 4 positions and two of them aren't on the chip. The file shows the signal; it can't make the clinical call.
- 15 of 35 CPIC guideline genes have no SNP definitions at all — including HLA-A and HLA-B (abacavir, carbamazepine, allopurinol hypersensitivity). They require dedicated typing methods no consumer chip provides.
- CYP2D6 is structurally unreadable: 5 of its CPIC alleles are whole-gene deletions/duplications/hybrids that no SNP array can see, and even its point variants are only partially present (62 of 157 positions on the GSA).
The chip gap: 23andMe-base vs AncestryDNA/MyHeritage-base
Per gene, the split is stark. A selection (positions carried / CPIC defining positions):
| Gene (what it informs) | GSA v3 (23andMe base) | OmniExpress (Ancestry/MH base) |
|---|---|---|
| CYP2C19 (clopidogrel, antidepressants) | 26 / 43 | 5 / 43 |
| CYP2C9 (warfarin, NSAIDs) | 51 / 88 | 2 / 88 |
| VKORC1 (warfarin) | 1 / 1 | 1 / 1 |
| SLCO1B1 (statin muscle risk) | 23 / 35 | 5 / 35 |
| DPYD (5-FU/capecitabine chemo) | 76 / 83 | 7 / 83 |
| TPMT (thiopurines) | 35 / 45 | 2 / 45 |
| NUDT15 (thiopurines) | 3 / 20 | 0 / 20 |
| G6PD (oxidative-stress drugs) | 67 / 173 | 0 / 173 |
| CYP2D6 (20–25% of drugs) | 62 / 157 | 3 / 157 |
| RYR1 (anaesthesia risk) | 96 / 320 | 0 / 320 |
The pattern has a simple explanation: the GSA was designed in the ClinVar era with deliberate clinical content; the OmniExpress is a genome-wide ancestry and research backbone. If your raw file came from AncestryDNA or MyHeritage, it was never built to carry medication genetics — a point that matters to anyone running their file through an interpretation tool.
Why "the SNP is there" still isn't a clinical call
Clinical pharmacogenomics doesn't report single SNPs — it reports star alleles (like CYP2C19 *2) and diplotypes, each defined by a complete set of positions. Our strictest metric asks: for how many of CPIC's 1,330 named alleles does the chip carry every defining position?
- GSA v3: 552 alleles (42%) · OmniExpress base: 33 (2%)
CYP2C19 *2 makes the gap concrete: its famous loss-of-function variant rs4244285 is on the GSA, so a raw file genuinely shows the *2 signal — but the full CPIC definition includes rs58973490 and rs3758581, which aren't on the chip. That's why serious interpretation tools hedge their calls, and why clinical labs run dedicated pharmacogenomic panels instead of consumer arrays. The signal is real; the certainty isn't.
The genes no chip can read
Two whole categories of the pharmacogenome are out of reach for every consumer chip, whichever brand:
- The HLA genes and 13 other guideline genes have no SNP definitions. 15 of the 35 CPIC guideline genes — ADRA2C, ADRB1, ADRB2, COMT, CYP2C8, CYP3A4, GRK4, GRK5, HLA-A, HLA-B, HMGCR, HTR2A, IFNL4, OPRM1, SLC6A4 — have no CPIC SNP allele table at all. The HLA pair is the highest-stakes miss: HLA-B*57:01 (abacavir) and HLA-B*15:02 (carbamazepine, Stevens-Johnson syndrome risk) require dedicated HLA typing.
- CYP2D6 is defined by structure, not just SNPs. The biggest pharmacogene of all — involved in metabolising an estimated 20–25% of prescription drugs, from codeine and tramadol to antidepressants and tamoxifen — carries whole-gene deletions, duplications and CYP2D6/CYP2D7 hybrids (5 structural alleles in CPIC's definitions). SNP arrays read one base at one position; they cannot see copy number. Even when your raw file lists CYP2D6 SNPs, it cannot tell you your metaboliser status.
What a curated interpretation panel reaches
For context, we ran the same analysis lens over Quanome's own committed 66-marker interpretation panel: it reaches 10 of the CPIC guideline genes (CYP2C19, CYP2C9, VKORC1, CYP4F2, SLCO1B1, TPMT, NUDT15, DPYD, UGT1A1, ABCG2) — real, high-stakes pharmacogenomics (clopidogrel, warfarin, statins, thiopurines, chemotherapy) — while missing the structurally-unreadable ones above, by necessity. That's the honest ceiling of any raw-file interpretation, ours included.
What this means for reading your own file
- Check which chip your file came from. A 23andMe v5 file carries genuinely useful pharmacogenomic signal; an AncestryDNA or MyHeritage file mostly doesn't — its base array holds 3% of the defining positions. (Our guides: 23andMe, AncestryDNA, MyHeritage.)
- Read single markers as signals, not verdicts. The famous positions are often there (see CYP2C19 in your raw data); complete clinical-grade calls usually aren't.
- The biggest genes need real tests. CYP2D6 and HLA are invisible to every consumer chip. Anything that matters is a prompt to ask a clinician or pharmacist about clinical pharmacogenomic testing — never a reason to start, stop or change a medication.
You can explore your own file with all of this honesty built in — without uploading it anywhere: our free DNA raw data explorer reads a 23andMe, AncestryDNA or MyHeritage file entirely in your browser.
Methodology
- Chip data (primary sources): official Illumina manifest files — Infinium Global Screening Array v3.0 (GSA-24v3-0_A2, 654,027 markers, GRCh38) and Infinium OmniExpress-24 v1.3 (714,238 markers). 23andMe v5 is GSA-based; AncestryDNA v2 and MyHeritage are OmniExpress-based.
- Pharmacogene data (primary source): all allele definitions, defining sequence positions (with dbSNP rsIDs and GRCh38 coordinates) and structural-variation flags for the 35 genes with CPIC guideline gene-drug pairs, retrieved from the public CPIC API (2026-09-07): 1,210 defining positions, 1,330 named non-reference alleles.
- Matching: a CPIC position counts as "on the chip" if its rsID appears in the manifest, or (GSA, GRCh38) its chromosome+position matches a probe. An allele is "fully readable" only if it is non-structural and all its defining positions are on the chip. Analysis script:
scripts/chip_coverage_study.py— every published number is computed, none estimated. - Limitations, stated plainly: consumer products add undisclosed custom content on top of these base arrays (23andMe in particular adds pharmacogenomic content for its FDA-cleared reports), so real products may read somewhat more than the documented base manifest — this analysis measures the documented array backbone, the honest lower bound. OmniExpress matching is rsID-only (99.98% of its manifest rows are rsID-named). Array versions change over time. External context figures (CPIC's 34-gene/164-drug 2025 summary; the ~50-pharmacogene review; CYP2D6's drug share) are cited: Caudle et al., 2025, PMC9640910.
- Date: September 2026 (first published August 2026 as a panel-scope analysis; upgraded with the chip-coverage computation). Citation welcome — please link to this page.
Read the markers behind the numbers
Several markers here have their own plain-English guides: CYP2C19, CYP1A2 caffeine metabolism, APOE4, MTHFR, and Factor V Leiden. For the full picture of reading a raw file, see the complete guide to 23andMe raw data and how to read your AncestryDNA raw data, or browse the genetics section.
Understand your DNA markers, honestly and privately
Quanome maps your 23andMe or AncestryDNA file against a curated marker set — on your device, never uploaded — with an AI coach that's clear about what a raw file can and can't tell you. Free on the App Store and Google Play.
Quanome is free on the App Store and Google Play. Want product news too? Leave your email below.
Read your own raw DNA — privately, on your device
Quanome imports your 23andMe, AncestryDNA, or whole-genome file and parses it locally on your phone — never uploaded — with an AI coach that reads it alongside your labs and Apple Health, and is honest about what a raw file can and can't tell you. Learn more about Quanome →
Frequently asked questions
Does it matter which DNA test I took for medication genetics?
Enormously — that's this analysis's headline finding. The array family behind 23andMe v5 (Illumina GSA) physically carries 570 of the 1,210 variant positions CPIC uses to define pharmacogene alleles (47%). The array family behind AncestryDNA and MyHeritage (Illumina OmniExpress) carries just 40 of them (3%) — it was designed for ancestry, not clinical content. Same DNA, very different files.
How many genes affect how you respond to medications?
CPIC — the clinical pharmacogenomics standard — currently has prescribing-guideline gene-drug pairs for 35 genes (34 in its 2025 summary paper, covering 164 drugs). A broader evidence review counts roughly 50 pharmacogenes. So the clinically-relevant 'pharmacogenome' is on the order of 34–50 genes.
Can 23andMe or AncestryDNA raw data tell me about my medications?
Partly, and unevenly. The 23andMe-base array carries many famous single positions — CYP2C19 rs4244285 (clopidogrel), the warfarin variants, the statin-risk SLCO1B1 variant. But carrying famous positions isn't the same as making guideline-grade calls: only 42% of CPIC's named alleles have their complete definitions on that array, and on the AncestryDNA/MyHeritage base it's 2%. Treat anything you find as educational — never a basis to change a medication.
Why can't consumer DNA tests read CYP2D6?
CYP2D6 — the most important drug-metabolising gene, involved in an estimated 20–25% of prescription drugs — is defined partly by structural variation: whole-gene deletions, duplications, and CYP2D6/CYP2D7 hybrids. In CPIC's data, 5 of its named alleles are structural. SNP arrays read single positions and cannot see copy number or rearrangements, so no consumer chip can determine your true CYP2D6 metaboliser status.
What about HLA genes like the abacavir and carbamazepine risk alleles?
They're another blind spot: 15 of the 35 CPIC guideline genes — including HLA-A and HLA-B, which flag dangerous hypersensitivity reactions to abacavir, carbamazepine and allopurinol — have no SNP-based allele definitions at all. HLA typing needs dedicated laboratory methods, so no SNP chip can properly test them, whichever brand you bought.
Is consumer raw data a substitute for a pharmacogenomic test?
No. Even the best consumer array carries under half of CPIC's defining positions, can fully read a minority of alleles, and is structurally blind to CYP2D6 and the HLA genes. It can be a useful prompt to ask a clinician or pharmacist about clinical pharmacogenomic testing — but prescribing decisions need clinical-grade methods, not a consumer chip.
How was this analysed?
We computed every original number from primary data: the official Illumina manifest files for the GSA v3 array (654,027 markers, the 23andMe v5 base) and the OmniExpress-24 v1.3 array (714,238 markers, the AncestryDNA/MyHeritage base), matched against all 1,210 allele-defining variant positions for the 35 CPIC guideline genes, retrieved from CPIC's public API. The method, matching rules and limitations are in the Methodology section, and the analysis script is reproducible.
Quanome is live — free on iPhone and Android
Make sense of your DNA and health data privately. Download Quanome free on the App Store or Google Play.
Quanome is free on the App Store and Google Play. Want product news too? Leave your email below.