How Accurate Are Personal Genome Tests?
April 30, 2008
I’ve had my DNA sequenced by 2 of the 3 companies now offering this service to the paying public. I purchased the tests for 23andMe and Iceland-based deCode. I am still plodding my way through the results — it’s sort of an education. One question I had was how well do the two results matched? I give the same DNA to both companies; the results ideally should be identical. DeCode claimed to test for 1,000,000 SNPs and 23andMe for 500,000, so the problem of lining all these results up to see what differs is not trivial. Luckily another user has just done this.
Antonio Oliveira also used both 23andme and deCode. He writes in his new blog:
In order to determine the accuracy of the genome profile provided by 23andMe and deCODEme I arranged to be genotyped by both companies and wrote a computer program to compare the results. The downloaded files contains 576,105 snips in the case of 23andMe and 1,013,349 snips for deCODE. After removing the no-calls and matching the two files by SNP identification, 560,299 snips were present in both files. The comparisson revealed 23 cases in which the results do not agree.
Oliveira made a chart of his results, categorized by chromosome.
The 23 errors makes the agreement between the two sets of data about 99.995% accurate, or an error rate of .005%, which is pretty good for medicine. A better test might be to repeat the test on the same DNA, but I assume the manufactures of the chip have done that. The 23 “unequal” SNPs caught here in disagreement are not SNPs currently associated with any diseases, so these particular errors are inconsequential. I don’t know if there are location biases in the errors, but presumably errors can appear in significant locations — at that very low rate. However if your computer had the same error rate, you’d notice.