43:["$","title",null,{"children":"23andMe Ancestry Inference | lorenasandoval88 | Observable"}] 4b:T1107,

View the notebook: 2021 ancestry inference papers ↗

23andMe v5: overlap with the ancestry papers

Your v5 manifest contains 34 SNP IDs listed in the VISAGE paper’s Table 1.

Marker list checked	Total markers	Confirmed rsID matches	Additional location-only match
VISAGE 2021, Table 1 Biallelic ancestry markers	100	34	1
Current official VISAGE online grid	115	33	1
PCA paper’s specific PC1 marker set	2,000	Not calculated; exact list not obtained	—

Why the VISAGE counts differ: the published table includes rs4959270, which appears in your manifest. The current online grid includes rs6437783 instead. These lists were checked separately; the complete historical 115-marker panel remains unverified.

Finding	How to interpret it
Confirmed rsID match	The same SNP identifier appears in the marker list and your manifest.
Location-only candidate: rs2789823	Your manifest contains its build-37 position under i710492, but no rsID alias confirming identity. It is excluded from confirmed counts.
Missing from the core manifest	This does not establish absence from every v5 export. Your manifest was assembled from 64 files using a 95% presence threshold.

These results measure marker coverage, not ancestry-inference accuracy.

Still unresolved: overlap with the PCA paper’s specific 2,000-marker PC1 set requires its exact marker list.

Sources: VISAGE paper · Current official marker grid · PCA paper

4c:Te30,

How I would rerun ancestry inference using v5 SNPs

Train on approximately 2,000 people from 1000 Genomes and test on approximately 500 held-out people. These numbers refer to people; the SNPs would come from the verified overlap between the v5 manifest and 1000 Genomes.

I would include the 34 SNPs from the VISAGE paper’s Table 1 that were identified in the v5 manifest, alongside the broader set of shared v5 SNPs, subject to genotype-quality and allele-matching checks.

Step	What I would do
1. Prepare the data	Obtain 1000 Genomes genotypes and population labels. Keep confidently matched autosomal v5 SNPs, including the 34 matched VISAGE markers. Check genome build, positions, and allele orientation.
2. Split the people	Reserve roughly 20% for testing, with each population represented in both sets. Keep close relatives together and save the sample lists for reproducibility.
3. Prepare the markers	Use training data to filter poorly measured or very rare SNPs and reduce redundant, correlated markers through LD pruning. Prioritize retaining the 34 VISAGE markers when choosing between redundant SNPs, and document any exclusions. Any additional informative-marker selection uses training data only.
4. Train the model	Fit PCA, then train a classifier on the PCA coordinates to predict broad reference-group labels. Choose settings through cross-validation within the training set.
5. Test once	Apply the saved preprocessing, PCA weights, and classifier to the held-out samples. Report overall accuracy, performance within each group, and a confusion table showing which groups were mistaken for others.
6. Apply to a v5 profile	Process an actual 23andMe genotype file using the same marker definitions and saved model. The manifest alone lists markers; it does not contain a person's genotypes.

Interpretation: this evaluates reference-group assignment. Classifier probabilities are not ancestry mixture percentages. The 1000 Genomes AMR label includes admixed American populations.

Methods: VISAGE paper · 1000 Genomes reference data · PLINK PCA · Keeping training and testing separate

4d:[{"id":"d749e568167b6de811ee97c221adf9651260142bf0de2da98205fa163e56e75e75915816d63b5c7caf86f53f1284ab7dd8711da0a161873872bacfac01237b9d","href":"https://static.observableusercontent.com/files/d749e568167b6de811ee97c221adf9651260142bf0de2da98205fa163e56e75e75915816d63b5c7caf86f53f1284ab7dd8711da0a161873872bacfac01237b9d","name":"v5_ancestry_paper_overlap.zip","type":"application/x-zip-compressed","lastModified":1792598399452,"size":31587},{"id":"96dcb308259cb2ccc24d0a0b5833684ac4696e190b9431e223844fac2abf6138101c9c39630e8aa7a0a357dd38407f7a9ae2b226aa5a8997532d0b19d1e6aed8","href":"https://static.observableusercontent.com/files/96dcb308259cb2ccc24d0a0b5833684ac4696e190b9431e223844fac2abf6138101c9c39630e8aa7a0a357dd38407f7a9ae2b226aa5a8997532d0b19d1e6aed8","name":"train_2003.ids","type":"application/octet-stream","lastModified":1792600487470,"size":16024},{"id":"2f371953d39235de783b7399d74e9f29651ac58822789d1aa0819acad24a5cbe190b5777213f731aca8881cf5ea8e5d24df52f1dd59c87dedb080034ed40a0b5","href":"https://static.observableusercontent.com/files/2f371953d39235de783b7399d74e9f29651ac58822789d1aa0819acad24a5cbe190b5777213f731aca8881cf5ea8e5d24df52f1dd59c87dedb080034ed40a0b5","name":"test_501.ids","type":"application/octet-stream","lastModified":1792600487471,"size":4008}] 47:{"id":"972ac8ed342db5dd","slug":"23and-me-ancestry-inference","title":"23andMe Ancestry Inference","thumbnail":"9c63c15de16707e0ecc5c25ad826e7d4e8fec8ea16a560cf16ea1bad48dc47fe","access_level":"public","create_time":"$D2026-10-21T15:58:16.695Z","update_time":"$D2026-10-21T19:48:26.106Z","view_time":null,"first_public_time":"$D2026-10-21T16:16:59.381Z","has_unpublished_edits":null,"fork":false,"stars":0,"starred":false,"owner":{"type":"team","id":"5a5147324ed15c62","login":"lorenasandoval88","name":"","avatar_url":"https://avatars-next.observableusercontent.com/c8926e2434569cc1f72aeac62929e824fdec3f0ec037d4a41a38a7728d14e950","tier":"starter_2024","bio":"","following":false},"version":30,"publish_version":30,"role":null,"authored":false,"authors":[{"type":"individual","id":"5c479a2c5ae5f52f","login":"lorenasandoval","name":"","avatar_url":"https://avatars-next.observableusercontent.com/c8926e2434569cc1f72aeac62929e824fdec3f0ec037d4a41a38a7728d14e950","tier":null,"bio":"","following":false}],"members":[{"type":"team","id":"5a5147324ed15c62","login":"lorenasandoval88","name":"","avatar_url":"https://avatars-next.observableusercontent.com/c8926e2434569cc1f72aeac62929e824fdec3f0ec037d4a41a38a7728d14e950","tier":"starter_2024","role":"editor"}],"comments":[],"custom_thumbnail":true,"trashed":false,"lapsed":null,"subscribed":null,"forks":0,"collection":{"id":"ac0528c4971e1d13","slug":"genomic-ancestry","title":"Genomic Ancestry","owner":{"id":"5a5147324ed15c62","login":"lorenasandoval88","name":""}},"type":"notebook","body":{"titling":"auto","title":"23andMe Ancestry Inference","cells":[{"id":1,"value":"# 23andMe Ancestry Inference","pinned":false,"hidden":false,"mode":"md","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":2,"value":"$4b","pinned":false,"hidden":false,"mode":"html","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"},{"id":3,"value":"$4c","pinned":false,"hidden":false,"mode":"html","author":"$undefined","summary":"$undefined","tool":"$undefined","providerMetadata":"$undefined","attachments":"$undefined","output":"$undefined","database":"$undefined"}],"files":"$W4d","stdlib":"2"}} 42:null