Ann Turner, co-author of the best book on DNA-based genealogy: Trace Your Roots With DNA, wrote me to say that she too has been comparing results from the two big genetic test companies, 23andMe and deCode. She wrote in response to my earlier posting comparing results between the two vendors.
The big news is that places where errors are showing up are probably not random. Here’s the argument, starting with her post on ancestry.com:
The two companies overlap on 562,532 SNPs. They agreed on 560,128 calls, or 99.6%. 23andMe didn’t make a call on 1,970 SNPs where deCODEme did, and deCODEme didn’t make a call on 399 records where 23andMe did. That leaves a mere 35 records where they actually made different calls [see the list below]. In all of those cases, one company would make a homozygous call while the other company made a heterozygous call — there were no cases where they made a completely discordant call.
Here’s the kicker from Ann’s letter to me:
Four of those (rs11149566, rs4458717, rs4660646, and rs 754499) were also found in Antonio’s list. That’s more than you would expect by chance.
Four out of 23 from Antonio’s list and four out of 35 on Turner’s list of discordant results indicates that these regions (at least) are unreliable.
This is why sharing results is so valuable and a key to great quantified self understanding.
This is a micrograph of the bead array on which these tests are conducted.
Turner’s 35 SNPs with different results, if case you also have done a comparison.
rs10435795 rs1045363 rs10743414 rs10945383 rs11149566 rs11179382 rs11707159 rs11915402 rs1209171 rs1221986 rs12907462 rs1303912 rs13422439 rs161381 rs17328647 rs1961196 rs1966357 rs2016461 rs2064034 rs2290516 rs2853981 rs3952469 rs4336661 rs4423481 rs4458717 rs4572718 rs4660646 rs6531490 rs6942478 rs7102702 rs754499 rs7812884 rs845217 rs9332128 rs9476380