My apologies, mbreese, but I actually must go through your comment line by line. You seem like a good guy, but you are way off base about a lot of the facts her

My apologies, mbreese, but I actually must go through your comment line by line. You seem like a good guy, but you are way off base about a lot of the facts here. Your quote from the doctor was for an FDA regulated diagnostic test, with a known error rate. FDA validated diagnostics generally have much, much smaller sample sizes than 23andMe. For example, this FDA approved diagnostic only has a few hundred samples tested: http://www.accessdata.fda.gov/cdrh_docs/reviews/K083846.pdf http://www.accessdata.fda.gov/cdrh_docs/reviews/K083846.pdf 23andMe has done tens of thousands, on much more highly varied populations (including many South and East Asians in addition to Caucasians). Their confidence intervals are just much tighter. The 23and me test is not a diagnostic and doesn't have a known error rate for such purposes. No. The error rate of iSelect custom assays is well known and can also be empirically estimated from the above data. I know affymetrix (their vendor) has been working in getting things approved for diagnostic uses, but I don think that snp chip has been approved yet. Affy's genotyping platform was approved many years ago. 23andMe does not use Affy, but rather Illumina. So, even if you were cleared for the genetic condition by 23andme, could you trust the results to make a clinical judgement? So far the FDA says no. And if I remember correctly, I'm pretty sure 23andme would agree. Several different issues are being confused here: * whether the test is "diagnostic grade" * whether 23andMe says the test should be used for medicine * whether the FDA will permit 23andMe to sell the test at all * whether all applications of genetics are medical * whether there is even any point in distinguishing between medical and nonmedical interpretation when "interpretation" often just means "Googling the variant" or "looking up the Pubmed papers" These techniques work for research because you always run replicates, so any random error is taken care of. The raw base accuracy of these platforms is not really at issue. The primary source of error is not genotype assessment, but the genotype/phenotype maps derived by researchers in the literature. You can easily tell whether someone has an A/A at a position. You cannot easily tell whether that "A/A" means that they will come down with a complex condition like diabetes. The key question is whether you believe 23andMe should be free to report what was openly published in Pubmed about that "A/A" variant's effect, or whether they need a $50 million government license to do this.