Sometimes our customers ask us about the accuracy of their 23andMe data. How certain are we that they really do have the genotypes we report?
The answer is, very certain. We typically claim that the genotyping technology we use reports the correct call for more than 99.9 percent of the approximately 580,000 single-letter DNA variations we report. Those variations, known as single-nucleotide polymorphisms, or SNPs, are the raw data upon which the 23andMe Personal Genome Service is built.
But apparently one of our customers decided to do more than ask. He designed a simple experiment to test the reproducibility of the SNP chips that 23andMe and other companies use to give people access to their genetic data.
Antonio C.B. Oliveira signed himself up for 23andMe and a similar service, then wrote a computer program to compare the genotype calls at the 560,299 SNPs where both companies reported data. He found a grand total of 23 discrepancies, or about 0.004% of the common reported SNPs.
“This error rate seems to me to be quite acceptable,” Oliveira concluded on Longa Vista, a blog he established to share his results.
Oliveira is actually the second customer to conduct such an experiment. When Ann Turner compared two datasets for the same person (not herself) back in January, she discovered 35 discrepancies.
And when Craig Venter and his colleagues published his full genome last October, they also found the discrepancies between their own sequencing results and two different SNP genotyping platforms to be negligible.
We aren’t surprised — these results confirm experiments that 23andMe performed before launching our Personal Genomics Service last year. But now 23andMe can say with confidence that independent testing has shown that the Illumina genotyping technology we use is extremely robust.