I can tell you, I’m a real molecular/microbiologist!

Tl;Dr: It’s mostly bunk.

Most of these services don’t analyze your entire genome*, but instead just regions of genes, looking for something called SNPs: single nucleotide polymorphisms. DNA is composed of four nucleobases, commonly represented by their initials: T and A, C and G. A SNP is a spot on a gene where there’s some variability in these, e.g., a C or A or even a T instead of a common G.

Through whole genome sequencing and statistical analysis, these companies were able to identify frequency trends in SNPs according to where the person lives and their self-reported ancestry. Now they use a cheaper, less comprehensive (but still fairly accurate) process to look for the SNPs that data suggests are most strongly correlated with different regions/ancestries and dole out your supposed ancestry.

There are problems.

Conclusions are only as good as your data, and the data are often based strongly on self-reporting, which is usually pretty inaccurate.

SNPs aren’t static - every child has some, about 20 to 60, that their parents don’t have. Many detrimental SNPs can lead to death, so most that persist have no effect, though some are weakly detrimental or, even more rarely, beneficial. That means there’s a limited pool of viable options, so your kid might have spontaneously developed a few strongly associated with a region they’re not at all connected to. You have a few too, as does your coparent and all of your parents. Through a couple of generations of new SNPs, a person’s ancestry results can shift. Through random chance and no new SNPs, one might inherit a combination of SNPs commonly seen in other regions, simply through the right combination of ancestors not at all from that area.

Some SNPs are better than others. Those on what are called “highly conserved” genes, i.e., fuck this gene up at all and you die, tend to be less common and more stable. If a defined group has an unusual SNP or SNPs on these regions, it’s a far better indicator of relatedness than a SNP on a gene for something like vitamin C synthesis, which we have but the process is broken so it doesn’t matter if we break it more.

In summary, these services are built on data of varying quality (shitty data) and moving targets of variable utility (shitty targets).

Thank you for coming to my TED Talk.

*If you can swing it, genome sequencing and analysis can be really interesting and useful for healthcare decisions. You can learn a lot about how you, specifically, work, and we’re learning more all the time.

Just be sure to get sufficient sequencing coverage, at least 30x if you want “good enough”, 100x or more if it’s medically vital and/or you’re looking for rare genes. 1x is fairly worthless, paying for it is a waste of money.