These SNP arrays, however, are exceedingly accurate and reproducible in their calls. A common medium-density SNP array used for human genetics is the Illumina CytoSNP-850K [1], which routinely gets >99.5% accuracy on calls. This is at least as good, or better than, whole genome sequencing.
The challenge in ancestry assignment is not determining what is in the sample, but rather in the interpretation and assignment of ancestral haplotypes.
[1] https://www.illumina.com/content/dam/illumina-marketing/docu...
>99.5% accuracy is great for a scientific measurement device that's only been developed very recently; it's absolutely rubbish if you want to sell the results to consumers.
Or potentially put people in jail based on these results...
You can get considerably better accuracy, you just have to be willing to fork over $1500 rather than $60. How much are you willing to pay to know if your 31% German vs 30%?
There was a good 5-10 years where microarrays were the go-to method for genomic investigation. Since RNA-seq has supplanted it, there has been no revolution in overturning those old results. At best, RNA-seq reveals what microarrays cannot: novel sequences, specific sequences that have difficulty hybridizing, etc.
The variation between these services is due primarily to how the data is interpreted and presented to customers, not the reads themselves.
In this case "SNP microarray" is a "digital" (or binary) readout of the genomic DNA sequence; specifically the array has attached to it sequences that match both the "reference allele" (e.g., a "G") and the "alternate allele" (e.g., "A" or some other base). Note that across the world's population, most sites in the human genome are "bi-allelic" meaning that there's really only two variants at each site, though of course it is possible to have tri-allelic or even fully polymorphic (A, T, C, or G) sites. For the purposes of ancestry, bi-allelic identification of commonly polymorphic sites is adequate, though the more sites and the rarer those sites are in world populations, the more specific you can be re: ancestry. This is why the gold standard for ancestry and haplotyping is whole genome sequencing-- you simply get far more informative sites to map to other populations.
>An entire DNA sample is made up of about three billion parts, but companies that provide ancestry tests look at about 700,000 of those to spot genetic differences.
>According to the raw data from 23andMe, 99.6 per cent of those parts were the same, which is why Gerstein and his team were so confused by the results. They concluded the raw data used by the other four companies was also statistically identical.
>Still, none of the five companies provided the same ancestry breakdown for the twins.
When my doctor asks how tall I am, if I'm 6'1", then 6'1" is a good answer. 6' (or even better: about 6') can also be an acceptable answer, as it's an approximation. I wouldn't say I'm 6'8", and then try to rationalize it by saying my tape measure is cheap, and anyway it's still right to about 10%.
Reporting what you don't know can be just as important as reporting what you do know. It doesn't matter how good your laboratory method is if you can't report results in a way that doesn't mislead customers.
So these people getting results and finding out they aren't related to their parents, are the results suspect?
Ditto the results about genes associated with dementia or heart disease?
I'm aware of the companies having disclaimers, I thought that was on the basis of them not really know what the genes did, not that the test itself was rubbish.
Is it a case of them adding too many decimal places to the percentages? Making statistics out of noise?