Saturday, January 17, 2015

Which DNA Company Has The Best Matching System?

I am now finally able to evaluate the new matching system, at AncestryDNA, which uses the Timber filter to eliminate supposedly IBS segments. I was quite shocked to find a 3rd cousin 1x removed, who was a 95% confidence level match, now missing from my Mom's match list. This match is in a Circle with us and definitely isn't considered a match anymore.  A cousin lost a total of 8 matches who were previously 95% confidence level matches, which no longer match at all.

Now that I have this comparison information I can evaluate the 3 companies I have experience with i.e: Family Tree DNA, AncestryDNA, and 23andMe.
  1. 23andMe's system appears to be the best. Their matching procedure isn't as rigid as the others, which may result in false matches,  but I believe they have a good confidence rating system which should provide enough guidance when it comes to establishing the likelihood that you are actually related.
  2. Family Tree DNA uses IBS small segments to determine matches. This sort of rating system can cause good matches to be lost.
  3. I rate AncestryDNA well below the others when it comes to matching. They phase results and now use a Timber filter to determine IBS segments. AncestryDNA processes as many as 30,000 kits a month. I believe the extra processing they do with the kits affects the accuracy of their match results. They claim their processes have a low error rate. I personally think the error rate is higher than they claim. Their computer system is full of glitches which may also affect the match results? Both my Mom and I lost 3 previously 95% confidence matches ( one was a 3rd cousin 1x removed). A third cousin of mine lost 8 matches in the previously 95% confidence level. After phasing these matches looked strong, the Timber filter eliminated them completely. Timber also pushed a few 95% confidence matches into the Extremely High confidence level. A couple of these matches have extensive trees going back 8 generations. I can't find a common ancestor or even common places with them?
I have to say I was sold on phasing after listening to AncestryDNA representatives talk about it. I believe it works well on a small scale, with more quality control than is  possible when working with a high volume of kits.

I think it would be best if all of the companies would use the same match criteria as 23andMe, which is at least one segment which is 7 cMs and 700 SNPs. 23andMe sets a cap on the number of matches to avoid the problem AncestryDNA had with massive numbers of matches. I think the cap is a better idea than trying to filter out IBS segments and in the process losing good matches. is more of an entertainment site rather than a serious genealogy site. A great blog post explaining the pitfalls of this approach was published this week "Ignorance is Bliss at AncestryDNA". I would still test with them; it's just important to realize their match procedure and confidence levels are flawed. I probably lost half a dozen matches I felt were likely true matches at AncestryDNA when the Timber filter was used. That's not a good thing, but I can live with that because the trees are so helpful.

No comments: