PDA

View Full Version : Why 23andMe only tells you about the last 500-1000 years



Tomenable
04-30-2017, 10:53 AM
I noticed already some time ago, that:


23andMe experts are very careful (which is why they have all of these "Broadly something" things). Nothing is noise there. I would say that they are actually too careful with detecting admixtures.

By too careful I mean reluctant to tell you about minor or "uncertain" admixtures.

And I found similar opinions by others:


In my observation 23andme underestimates Eastern Euro admixture when someone still plots in a very West Euro countries. It might look unsure so it ends up mostly in "Broadly..." or "Unassigned", even in Speculative mode.
Which is probably why Poles usually get the same % of "North Slavic" in DNA.Land as their % of "East Euro" in 23andMe, but Westerners usually tend to get more "North Slavic" than "East Euro"

As it seems, 23andMe tend to label such admixtures as "Broadly something".
23andme shows your ethnic heritage from the last 500 years in form of bounded geographical areas you are native to . on gedmatch your genome gets broken down further in more detailed components in a larger than 500 year past timeframe.

But I wasn't sure why was this the case.

Until I found this post by user Kurd:

http://www.anthrogenica.com/showthread.php?4576-Do-you-trust-23andme-s-Ancestry-Composition-or-GEDmatch-calculators-more&p=145261&viewfull=1#post145261


23andMe's speculative mode greatly overestimates major components, and underestimates minor components. This is due to their methodology of snipping the genome into 100 SNP segments to compare against the limited references they have. So for example, if 60% of the the segment indicates Middle Eastern, and 40% indicates S Asian, that segment is assigned 100% Middle Eastern. In effect 40% of the segment, which is S Asian is ignored, and the whole segment is assigned Middle-Eastern.

Also, their methodology includes segment smoothing, which means if there are chunks of minor components in a segment, they are ignored.

That is how Iranians and W Asians turn out 98-100% Middle Eastern, and folks in neighboring Pakistan turn out 98-100% S Asian in speculative mode.

This naturally is unrealistic and uninformative, because you don't need a test to tell you that. Conservative mode is better with regards to inflation of major components and underestimation of minor components, but the trouble here is that people get 5-70% unassigned. This is where your minor components are hidden.

The above translates to 23andMe being useless for figuring out your minor components to any degree of accuracy.

Some other comments:


I know I'm at least 1/8 Polish and I get 11.3% Eastern European on speculative. If I go on standard mode it drops to 4.2% and if I go onto conservative I get none at all.

When using Eurogenes Mixed mode population sharing on Gedmatch I'll usually get something like,

56.4% Orcadian + 43.6% South_Polish @ 3.91

or

77.5% North_German + 22.5% Ukrainian @ 4

It shows me as way more Eastern European than 23andme does.

FTDNA seems a bit weird, but still generally gets my ancestry right,

I get 56% for the British Isles, 20% Eastern European,

The rest is like 5% per of the other European populations + Turkey.

I have found that these genetic calculators tend to get the whole picture right. But there seems to be quite a bit of variation between them. I imagine this is because they're using different populations to compare you with, different samples, and different methods for calculating ancestry. And also because there's a great deal of overlap between Europeans.

Another one:




I don't really agree because I've noticed very peculiar things happen with the 23andme calculator. I've seen families where one parent is Sicilian and the other is Irish. The Sicilian parent scores 15% Middle Eastern, while the child scores 0%. I've yet to see a half North European, half SE European (Cretan, Sicilian, Greek islander) score significant Middle Eastern even though full members of said groups score a significant amount.

Likewise, I suspect 23andme underestimates admixture when it is small. I have seen that in people with under 10% of African, the African is greatly underestimated on 23andme.

What do others think? I think due to its algorithms, 23andme ends up giving misleading results, but it is good for telling if a given ancestral component exists... just not necessarily the amounts.

And it underestimates admixture even more if it is both small and old (= fragmented, small segments).

Basically, 23andMe was designed for people in the Americas to tell them where they ancestors lived in 1492 AD. It seems to be accurate at predicting continental breakdown of ancestry for Latin Americans. In general it seems quite accurate for people with recently mixed ancestry, but not necessarily for older mixtures.

So it tends to be less accurate for Europeans than for people in the New World. Their methodology is designed in such a way, that they can ignore minor and older admixtures. 23andMe ancestry report tells you more about recent geographical affinities than about deep genetic affinities.

A Norfolk L-M20
04-30-2017, 02:01 PM
That's pretty much my conclusion. They wanted a product that fits the North American markets. 23 fanboys will wave the white paper around and loudly proclaim that 23 focuses onto 300 - 500 years (I've seen them claim 500 and 300 years) by filtering out short segments. Trust me, it doesn't work. I don't want to bore people with it, but I could use the example of English testers, although I suspect that it could apply to most other European populations. The English get in the following order: British & Irish, French & German, Scandinavian, and more often than not, a small percentage of Southern European. It is uniform to English people of known local ancestry. However, if you apply the white paper, then all of us English testers then have had the same French & German, Scandinavian, and often Southern European direct ancestors invisibly admix into our trees over the past 300 to 500 years. Yes, I know, the Huguenots, sorry no, that cannot account for that uniform population background. In actual fact, the most common known admixture for many English families in recent centuries is Irish and Scottish, which should reinforce our British & Irish ancestry.

What I am saying is that the 23 algorithm cannot deal with populations that have been admixed further back in history, where that admixture has become a part of the population background. This is probably made worse by poorly laid out data sets and population references. So, as I'm sure that my Normand, Dutch, French, and perhaps German friends on here can also confirm - 23 ancestry composition does not work. It cannot deal with old admixture. My English testers also do have a lot of "Broadly NW European. I have a whopping 36% after phasing (it increased the uncertainty with phasing - previously it was 29%). Old admixture that doesn't fit in elsewhere?

jpb
04-30-2017, 02:12 PM
I agree with Norfolk. I think my 23andme results are kind of similar to his (I am like 50-50 British Isles/German and I often cluster near North Dutch and SE English), but with more EE and AJ and a whopping 37.5% Broadly NW Euro. I also agree with the Eastern European point. My whole family, including my self, has at least 6% East Euro on Ancestry from our East German/Polish ancestry. I feel 23andme really underestimated it. I am waiting for my 1/8 Pomeranian (central Poland, but supposedly ethnic German) grandma's 23 results. She is even more mixed German-British Isles than I am so I feel they're going to have a lot of trouble with her, as she also has very minority Italian and Native American.

jpb
04-30-2017, 02:17 PM
Also, my SSA segment, proven by Dr. McDonald, is listed as Unassigned on 23. :\

wombatofthenorth
05-02-2017, 03:34 AM
I agree with Norfolk. I think my 23andme results are kind of similar to his (I am like 50-50 British Isles/German and I often cluster near North Dutch and SE English), but with more EE and AJ and a whopping 37.5% Broadly NW Euro. I also agree with the Eastern European point. My whole family, including my self, has at least 6% East Euro on Ancestry from our East German/Polish ancestry. I feel 23andme really underestimated it. I am waiting for my 1/8 Pomeranian (central Poland, but supposedly ethnic German) grandma's 23 results. She is even more mixed German-British Isles than I am so I feel they're going to have a lot of trouble with her, as she also has very minority Italian and Native American.

Even 100% Polish people won't get 100% Eastern European. Look at the size of the region, of course not everyone will be able to score 100% of anything. So you need to look a bit at ratios. MyOrigins 2.0 tried to be able to get say 100% Polish or Hungarians to get 100% Eastern European, but look what happens. A huge area just scores 100% EE and you don't get any hints from ratios as to what part and if you had any say German or Swedish or even something exotic it just gets turned all into Eastern European, even French sometimes get it marked a lot of Eastern European, that's what happens when you try to force that huge region to be able to have many people able to score 100% of something, it's worse.

Mixed
05-02-2017, 11:38 PM
23andme is not good for Slavs. I've even read that here. I know someone who is half Russian and 23 gave them 0% Eastern European. Either way there is no perfect company but 23andme is not the best and all that it's hyped up to be.