What Your 23andMe Data Can Actually Tell You About Peptides

Your 23andMe raw data file has roughly 650,000 rows. Your genome has approximately 3 billion base pairs and an estimated 4 to 5 million variable positions. That means your raw data covers about 0.02% of your DNA. For most health questions, that slice is adequate because the variants that matter most are common enough to be on the chip. For peptide response prediction specifically, the question is which 0.02% your chip actually measured, and whether that set includes the variants relevant to your protocol. The answer, it turns out, is mostly yes for the high-impact axes and no for a targeted set of rare variants that matter in specific circumstances.

The good news is structural: the four genetic axes most predictive of peptide response are all common variants that appear across every major consumer genotyping array. COMT, BDNF, SOD2, and MC4R variants are directly genotyped on 23andMe (v4 and v5), AncestryDNA, and MyHeritage. If you have a raw data file from any of these providers, you almost certainly have reliable calls for the variants that explain the most inter-individual variance in peptide response. The gaps are real, but they are concentrated in a specific set of drug-interaction loci and rare pathway variants that matter mainly in edge cases.

Does your 23andMe version matter for peptide genetics?

Yes, and more than most people realize. 23andMe has deployed four meaningfully different genotyping arrays since 2007. The current chip (v5, in use since 2017) uses Illumina's Global Screening Array, which was designed with explicit pharmacogenomics coverage as a goal. The v4 chip (2013 to 2017) is the problem case: it dropped a significant number of SNPs relative to v3, including several pharmacogenomically relevant loci, reducing coverage of drug and variant pathway genes. If you tested between 2013 and 2017, your file has structurally different coverage from someone who tested before or after that window.

The most common consumer assumption is that two people both "have 23andMe data" and are starting from equivalent inputs. They are not. Depending on when you tested, your raw data file may or may not directly genotype certain metabolizer variants. That is not a failure of the service -- the earlier chips were designed for different primary use cases -- but it is a material consideration when using the data for peptide protocol design.

Version	Years active	SNPs on chip	PGx coverage	Peptide-relevant notes
v3	2010 to 2013	~960,000	Good	High coverage including many pharmacogenomics loci. Well-covered for core peptide axes.
v4	2013 to 2017	~580,000	Reduced	Dropped many SNPs from v3 including some pharmacogenomic variants. CYP3A4 *22 absent or unreliable. COMT and SOD2 still present.
v5	2017 to present	~640,000	Strong	Illumina GSA design includes explicit PGx targets. Best consumer array for peptide-relevant coverage. CYP3A4 *22 directly genotyped.
v1/v2	2007 to 2010	~550,000	Limited	Legacy arrays. Not suitable for reliable peptide genetics without clinical confirmation on key variants.

How to find out which version you have

Log in to 23andMe, go to Settings, then select "23andMe Data", then click "View" next to your profile. Your chip version appears in the header of your raw data file when you download it. Alternatively, open the downloaded file and look at the comment lines at the top starting with #. If you tested between 2013 and 2017, you almost certainly have v4. Testing before 2013 or after 2017 means v3 or v5 respectively. Your version tells you whether the CYP3A4 *22 variant -- the most clinically important peptide-adjacent metabolizer variant -- was directly measured or was absent from the chip.

Which peptide-relevant SNPs does 23andMe reliably genotype?

The variants that matter most for peptide protocol design are common enough that all major consumer arrays include them. A 2024 pharmacogenomics workflow study (Yin et al., Frontiers in Pharmacology, PMID 38529185) independently validated that the GSA chip -- the foundation of 23andMe v5 -- covers 21 of the 35 Tier 1 "Very Important Pharmacogenes" classified by PharmGKB, with 99.48% inter-run concordance. For the SNPs below, that concordance figure means your genotype call is highly reliable.

GLP1R rs6923761 -- the variant creating a 58% difference in semaglutide response speed between genotypes -- is where coverage becomes less certain. Its presence on the v5 chip manifest has not been confirmed in independent public sources, meaning it may require statistical imputation even on v5 data. A 2025 prospective study documented why this specific variant matters enormously:

The GLP1R rs6923761 A/A genotype was associated with a significantly higher monthly rate of weight loss (1.64% vs 1.04% in G-allele carriers), representing an approximately 58% difference in weight loss velocity on semaglutide 2.4 mg. These findings suggest GLP1R genotype should be considered a clinically meaningful predictor of GLP-1 receptor agonist outcomes.

Phan et al., Obesity (Wiley), 2025, PMID 40384505

When this variant is directly genotyped, the result is reliable. When it is imputed from nearby markers, it carries a confidence interval that the reported genotype does not make explicit. If you want a confirmed call on this specific variant and you tested before 2017 or with AncestryDNA, a dedicated peptide genetics panel or 23andMe v5 test gives you a direct measurement. For the full picture of what GLP1R predicts about semaglutide response and which related variants also matter, the semaglutide genetics guide covers the pharmacogenomics in depth.

Gene / SNP	23andMe v5	23andMe v4	AncestryDNA	MyHeritage
COMT rs4680	Direct	Direct	Direct	Direct
BDNF rs6265	Direct	Direct	Direct	Direct
SOD2 rs4880	Direct	Direct	Direct	Direct
NOS3 rs1799983	Direct	Direct	Direct	Direct
MC4R rs17782313	Direct	Direct	Direct	Direct
GLP1R rs6923761	Uncertain (possibly imputed)	Imputed	Imputed	Imputed
CYP3A4 *22 (rs35599367)	Direct	Absent / unreliable	Some versions	Imputed
CYP2D6 copy number variants	Not detectable	Not detectable	Not detectable	Not detectable

What does 23andMe miss for peptide genetics?

The false-positive nuance you need to understand

A landmark 2018 study in Genetics in Medicine (Tandy-Connor et al., PMID 29565420) found that 40% of medically classified variants reported by direct-to-consumer genetic tests were false positives when confirmed in a clinical laboratory. This number is real, but it applies almost entirely to rare pathogenic variants in cancer genes like BRCA1 and BRCA2 -- not to the common pharmacogenomic SNPs relevant to peptide response. For variants with a minor allele frequency above 1%, which includes all of COMT, BDNF, SOD2, and NOS3, the GSA chip's design accuracy is substantially higher. The 40% figure is not a reason to distrust your COMT genotype call. It is a reason to distrust any consumer array result for rare cancer variants, which is a completely different category of data.

The CYP metabolizer blind spots

CYP3A4 *22 (rs35599367) is the slow-metabolizer allele most relevant to drug interaction risk in peptide protocols. Someone combining research peptides with prescription medications needs to know if they are a CYP3A4 slow metabolizer, because reduced enzyme capacity raises serum concentrations of co-medications that clear through this pathway. On 23andMe v5, this variant is directly genotyped. On v4 chips, it is absent or unreliable. The coverage on AncestryDNA depends on which version of their array you tested on. If your protocol involves any prescription medications alongside peptides, knowing your v5 or clinical CYP3A4 status is a safety input, not just an optimization input. The CYP enzymes guide covers the full metabolizer phenotype spectrum and what each variant actually determines for a peptide protocol.

CYP2D6 is a more fundamental gap. The major star alleles that define ultrarapid and poor metabolizer phenotypes involve copy number variations -- gene duplications and deletions -- that no SNP genotyping chip can detect, because SNP arrays measure single-base changes, not structural gene rearrangements. A consumer array might detect a few CYP2D6 point mutations that suggest reduced activity, but it cannot confirm your full metabolizer phenotype without a gene dosage assay. For the peptides most commonly used in research protocols (BPC-157, TB-500, GHK-Cu, ipamorelin), this gap is less critical because these peptides clear primarily via proteolytic enzymes rather than CYP2D6. For anyone combining peptides with opioids, many antidepressants, or antipsychotics, a clinical CYP2D6 test provides information the consumer array cannot.

Rare tissue-repair pathway variants

The consumer genotyping arrays were designed to detect common variants present in at least 1% of the population. The tissue-repair pathway variants most relevant to healing peptide response -- including some MMP3, VEGF, IGF1R, and COL5A1 loci -- have some common SNPs on the arrays, but the rarer functional variants with the clearest experimental evidence are often below the chip's design frequency. What this means practically: for BPC-157 and TB-500 response, your 23andMe data gives a first-pass signal on the common variants in these pathways, but it cannot rule out functionally relevant rare variants that the chip never tested for. This is not a reason to dismiss the common-variant signal. It is a reason to understand what confidence level that signal carries.

How do you actually extract peptide insights from your raw data?

The raw data file is a plain text file with roughly 650,000 rows, each formatted as an rsID, chromosome position, and two-letter genotype call (for example, AA, AG, or GG). No interpretation is included. Extracting peptide insights requires matching your rsIDs against known variant associations. You can do this manually for a handful of specific variants or through an analysis service that processes all relevant rsIDs together.

AncestryDNA and MyHeritage raw files use the same rsID-based structure as 23andMe but have different header formats and slightly different SNP coverage. The analysis logic is identical. If you have data from any of these providers, the same lookup process applies. The differences that matter are the coverage gaps described above, particularly for GLP1R and CYP3A4. See the DNA-first decision framework for a step-by-step walkthrough of how to apply the four genetic axes once you have your variant calls in hand.

Is your 23andMe data still accessible after the 2025 bankruptcy?

23andMe filed for Chapter 11 bankruptcy on March 23, 2025. In June 2025, a US bankruptcy court approved a $305 million sale of 23andMe's genetic platform and data to TTAM Research Institute, a nonprofit organization co-founded by 23andMe co-founder Anne Wojcicki (source: Bloomberg Law, June 2025; NPR, June 2025). The sale closed in July 2025. Raw data downloads remained available through the 23andMe customer portal as of early 2026, but the ownership transition and the absence of federal genetic data protections in US bankruptcy proceedings created legitimate questions about long-term access.

The data security context matters separately. The 2023 23andMe credential-stuffing attack exposed profile data for approximately 7 million users. PIPEDA Findings #2025-001 (January 2025), a joint investigation by Canada's Privacy Commissioner and the UK's Information Commissioner's Office, concluded that 23andMe failed to adequately safeguard user data in that breach. The combination of the bankruptcy sale and the breach investigation underscores a practical recommendation: if you tested with 23andMe and have not downloaded your raw data file, do so now. The file is yours by right, and having a local copy preserves your access to your own genetic data regardless of what happens to the platform under new ownership.

The GHK-Cu peptide page includes the SOD2 variant context that is likely already in your 23andMe file, providing a concrete example of what a genetic match looks like once raw data is processed against a specific peptide's biological mechanism.

How reliable is a consumer-data peptide recommendation?

The honest answer depends on which peptide you are asking about. For semaglutide and GLP-1 class compounds, the key predictive variant (GLP1R rs6923761) may or may not be directly genotyped on your specific chip version -- but the related variants (MC4R rs17782313) that complete the picture are reliably present. For nootropic peptides, the COMT variant is directly genotyped everywhere. For antioxidant-modulating peptides like GHK-Cu and SS-31, the SOD2 variant is universally covered on all consumer arrays. These are not approximations. They are the same variant calls that pharmacogenomics researchers use in published studies, delivered through a consumer interface rather than a clinical lab.

Reliability drops at specific edges: CYP drug-interaction risk requires v5 or clinical testing for the most relevant loci, rare tissue-repair pathway variants are not on consumer chips, and imputation accuracy degrades for non-European ancestries because the reference panels used to fill in missing data are less complete for those populations. If you are in a population underrepresented in the 1000 Genomes or gnomAD reference panels, your imputed variants carry meaningfully more uncertainty than the same variants imputed for a European-ancestry individual. This is a structural limitation of consumer genomics broadly.

The appropriate framing is: your consumer genomic data gives a reliable, direct measurement for the common variants most predictive of peptide response. It gives a statistically inferred estimate for some less common variants. It gives nothing for structural variants and rare loci that were never on the chip. Knowing which category each relevant variant falls into is exactly what distinguishes a calibrated first pass from a false sense of precision.