23andMe’s New Ancestry Composition (Ethnicity) Chromosome Segments

I was excited to see 23andMe’s latest feature that provides customers with Ancestry Composition (ethnicity) chromosome segment information by location.  This means I can compare my triangulation groups to these segments and potentially identify which ancestor’s DNA that I inherited carry which ethnicity – right?? Another potential way to help discern whether I should ask Santa for lederhosen or a kilt?

Not so fast…

Theoretically yes, but as it turns out, after working with the results, this tool doesn’t fulfill it’s potential and has some very significant issues, or maybe this new tool just unveiled underlying issues.

Rats, I guess Santa is off the hook.

Let’s take a look and step through the process.

Ancestry Composition Chromosome Painting

To see your Ancestry Composition ethnicity chromosome painting, sign into 23andMe, then go to the Reports tab at the top of your page and click on Ancestry. Please note that you can click on any of the graphics in this article to enlarge.

23andme-eth-seg-1

Then click on Ancestry Composition, which shows you the following:

23andme-eth-seg-2

Scrolling downs shows you your chromosomes, painted with your ethnicity. This isn’t new and it’s a great visual.

You may note that 23andMe paints both “sides” of each chromosome separately, the side you received from your mother and the side you received from your father. However, there is no way to determine which is which, and they are not necessarily the same side on each chromosome.

If one or both of your parents tested at 23andMe, you can connect your parents to your results and you can then see which ethnicity you received from which parent.

Let’s work through an example.

23andme-eth-seg-3

This person, we’ll call her Jasmine, received two segments of Native ancestry, one on chromsome 1 and one on chromosome 2, both on the first (top) strands or copies. She also received one segment of African on DNA strand (copy) 1 of chromsome 7.

Caveat

Words of warning.

JUST BECAUSE THESE ETNICITIES APPEAR ON THE SAME STRANDS OF DIFFERENT CHROMOSOMES, STRAND ONE IN THIS CASE, DOES NOT MEAN THEY ARE INHERITED FROM THE SAME PARENT.  

Each chromosome recombines separately and without a parent to compare to, there is no way to know which strand is mother’s or father’s on any chromsome. And figuring out which strand is which for one chromsome does NOT mean it’s the same for other chromsomes.

In fact, Jasmine’s mother has tested, and she has NO African on chromosome 7. However, Jasmine and her mother both have Native American on chromosomes 1 and 2 in the same location, so we know absolutely that Jasmine’s strand 1 on chromosome 7 is not from the same parent as strand 1 on chromosome 1 and 2, because Jasmine’s mother doesn’t have any African DNA in that location.

If you’re a seasoned 23andMe user, and you’re saying to yourself, “That’s not right, the chromosome sides should be aligned if a parent tests.”  You’re right, at least that’s what we’ve all thought.  Keep reading.

Let’s dig a bit further.

Connecting Up

23and Me encourages everyone to connect their parents, if your parents have tested.

Jasmine’s mother has tested and is connected to Jasmine at 23andMe.

23andme-eth-seg-4

Even though the button says “Connect Mother,” which makes it appear that Jasmine’s mother isn’t connected, she is. Clicking on Jasmine’s “Connect Mother” button shows the following:

23andme-eth-seg-5

Furthermore, if the parent isn’t connected, you don’t see any parental side ethnicity breakdown – and we clearly see those results for Jasmine.  Below is an example of the same page of someone whose parents aren’t connected – and you can see the verbiage at the bottom saying that a parent must be connected to see how much ancestry composition was inherited from each parent.

23andme-eth-seg-not-connect

If a child is connected to at least one parent, 23andMe, based on that parent’s test, tells the child which sides they inherited which pieces of their ethnicity from, shown for Jasmine, below.

23andme-eth-seg-6

In this case, the mother is connected to Jasmine and the father’s ethnicity results are imputed by subtracting the results where Jasmine matches her mother. The balance of Jasmine’s DNA ethnicity results that don’t match her mother in that location are clearly from her father.

23andMe may sort the results into the correct buckets, but they do not correctly rearrange the chromosome “copies” or “sides” on the chromosome browser display based on the parents’ DNA, as seen from the African example on chromosome 7. Either that, or the ethnicity phasing is inaccurate, or both.

You can see that 23andMe tells Jasmine that all of her Native is from her mother’s side, which is correct.

23andMe tells Jasmine that part of her North African and Sub-Saharan African are from her mother, but some North African is also from her father. You can see Jasmine’s African on her chromosome 7, below.

23andme-eth-seg-7

There is no African on Jasmine’s mother’s chromosome 7, below.

23andme-eth-seg-8

So if African exists on chromosome 7, it MUST come from Jasmine’s father’s side. Therefore, side one of chromosome 7 cannot be Jasmine’s mother’s side, because that’s where Jasmine’s African resides.

This indictes that either the results are incorrect, or the “sides” showing have not been corrected or realigned by 23andMe after parental ethnicity phasing, or both.

Here’s another example. Jasmine shows Middle East and North Africa on chromosomes 12 and 13 on sides one and two, respectively.

23andme-eth-seg-9

Jasmine’s mother shows Middle East and North Africa on chromosome 14, only, with none showing on chromosome 12 or 13.

23andme-eth-seg-10

Yet, 23andMe shows Jasmine receiving Middle East and North African DNA from her mother.

23andme-eth-seg-11

Jasmine is also shown as receiving Sub-Saharan African and West African from her mother, but Jasmine’s mother has no Sub-Saharan or West African, at all.

Interestingly, when you highlight both West African and Sub-Saharan African, shown below, it highlights the same segment of Jasmine’s DNA, so apparently these are not different categories, but subsets of each other, at least in this case, and reflect the same segment.

23andme-eth-seg-12

23andme-eth-seg-13

Jasmine’s mother shows this region of chromosome 7 to be “European” with no further breakdown.

Clearly Jasmine’s sides 1 and 2 have not been consistently assigned to her mother, because Jasmine’ African shows on both sides 1 and 2 of chromosomes 12 and 13 and Jasmine’s mother has no African on either on those chromosomes – so those segments should be assigned consistently to Jasmine’s father’s side, which, based on Jasmine’s match to her mother on chromosome 1, side 1 – Jasmine’s father’s “copy” should be Jasmine’s side 2.  This tool is not functioning correctly.

Jasmine’s father is deceased, so there is no way to test him.

The information provided by 23and Me contradicts itself.

Either the ethnicity assignment itself or the parental ethnicity phasing is inaccurate, or both. Additionally, we now know that the chromosome “sides,” meaning “copies” are inaccurately displayed, even when one parent’s DNA is available and connected, and the sides could and should be portrayed accurately.

This discrepancy has to be evident to 23andMe, if they are checking for consistency in assigning child to parent segments.  You can’t assign a child’s segment to a parent who doesn’t carry any of that ethnicity in a common location.  That situation should result in a big red neon sign flashing “STOP” in quality assurance.  Inaccurate results should never be delivered to testers, especially when there are easy ways to determine that something isn’t right.

The New Feature – Ethnicity Segments

Like I said, I was initially quite excited about this new feature, at least until I did the analysis. Now, I’m not excited at all, because if the results are flawed, so is the underlying segment data.

My original intention was to download the ethnicity segment information into my master spreadsheet so that I could potentially match the ethnicity segments against ancestors when I’ve identified an ancestral segment as belonging to a particular ancestral line.

This would have been an absolutely wonderful benefit.

Let’s walk though these steps so you can find your results and do your own analysis.

When you are on the Ancestry Composition page, you will be, by default, on the Summary page.

23andme-eth-seg-14

Click on the Scientific Details tab, at the top, and scroll down to the bottom of the page where you will see the following:

23andme-eth-seg-15

You will be able to select a confidence level, ranging from 50% to 90%, where 50% is speculative and 90% is the highest confidence. Hint – at the highest confidence level, many of the areas broken out in the speculative level are rolled up into general regions, like “European.”  Default is 50%.

23andme-eth-seg-16

Click on download raw data and you can then open or save a .csv file. I suggest then saving that file as an Excel file so you can do some comparisons without losing features like color.

In my case, I saved a 50% confidence file and a 90% confidence file to compare to each other.

I began my analysis with both strands of chromosome 1:

Strand 1 was easy.  (Click on graphic to enlarge.)

23andme-eth-seg-17

At the 50% confidence level, on the left, three segments are identified, but when you really look at the start and end positions, rows one and two overlap entirely. Looking back at the chromosome browser painting, this looks to be because that segment will show up in both of those categories, so this isn’t an either-or situation. Row 3 shows Scandinavian beginning at 79,380,466 and continuing through 230,560,900, which is a partial embedded segment of row 2.

At the 90% confidence level, on the right, above, this entire segment, meaning all of chromosome 1 on side 1, is simply called European.

You can see how this might get complex very quickly when trying to utilize this information in a Master DNA Spreadsheet with your matches, especially since individual segments can have 2 or 3 different labels.  However, I’d love to know where my mystery Scandinavian is coming from – assuming it’s real.

Now, let’s look at strand 2 of chromosome one. It’s a little more complex.

23andme-eth-seg-18

I’ve tried to color code identical, or partially-overlapping segments.

The red, green and apricot segments overlap or partially overlap at the 50% level, on the left, indicating that they show up in different categories.

The red segments are partially the same, with some overlapping, but are grouped differently within Europe.

The green Native/East Asian segments at the 90% level are interrupted by the blue unassigned segments in the middle of the green segments, while at the 50% confidence level, they remain contiguous.

All of the start and end segments change, even if the categories stay the same or generally the same. The grey example at the bottom is the easiest to see – the category changes to the more general “European” at the 90% level and the start segment is slightly different.

Jasmine and Her Mother

As one last example, let’s look at the segments at the 50% confidence level, which should be the least restrictive, that we were comparing when discussing Jasmine and her mother.

You can see, below, that Jasmine’s Native portion of chromosome 1 and 2 are either equal to or a subset of her mother’s Native portion, so these match accurately and are shown in green.

This tells us that Jasmine’s mother’s side of chromosomes 1 and 2 is Jasmine’s “copy 1” and given that we can identify Jasmine’s mother’s DNA, all of Jasmine’s “copy 1” should now be displayed as her mother’s DNA, but it isn’t.

23andme-eth-seg-19

On chromosomes 7 and 12, where Jasmine’s copy 1 shows African DNA, her mother has none. All African DNA segments are shown in red, above.

Furthermore, 23andMe attributes at least some portion of Jasmine’s African to Jasmine’s mother, but Jasmine’s mother’s only African DNA appears on chromosome 14, a location where Jasmine has none. There is no common African segment or segments between Jasmine and her mother, in spite of the fact that 23andMe indicates that Jasmine inherited part of her African DNA from her mother.  It’s true that Jasmine and her mother both carry African DNA, but not on any of the same segments, so Jasmine did not inherit her mother’s African DNA.  Jasmine’s African DNA had to have come from her father – and that’s evident if you compare Jasmine and her mother’s segment data.

Where Jasmine has African DNA segments, above, I’ve shown her mother’s corresponding DNA segments on both strands for comparison. I have not colored these segments. Conversely, where Jasmine’s mother has African, on chromosome 14, I have shown Jasmine’s corresponding DNA segments covering that segment.  There are no matches.

Clearly Jasmine did not inherit her African segments from her mother, or the segments have been incorrectly assigned as African or European, or multiple problems exist.

Summary

I initially thought the Ancestry Composition segments were a great addition to the genealogists toolset, but unfortunately, it has proven to be otherwise, highlighting deficiencies in more than one of the following area:

  • Potentially, the ancestry composition ethnicity breakdown itself.  Is the underlying ethnicity assignment incorrect?  In either case, that would not explain the balance of the issues we encountered.
  • The chromosome “sides” or “copy” shown after the parental phasing – in other words, the child’s chromosome copies can be assigned to a particular parent with either or both parents’ DNA. Therefore, after parental phasing, all of the same parent’s DNA should consistently be assigned to either copy 1 or copy 2 for the child on all of their chromosomes.  It isn’t.
  • The child’s ethnicity source (parent) assignment based on the parent’s or parents’ ethnicity assignment(s).  Hence, the African segment assignment issues above.
  • The ethnicity phasing itself.  The assigning of the source of Jasmine’s African DNA to her mother when they share no common African segments.  Clearly this is incorrect, calling into question the validity of the rest of the parental ethnicity phasing.

Unfortunately, we really don’t have adequate tools to determine exactly where the problem or problems lie, but problems clearly do exist. This is very disappointing.

As a result, I won’t be adding this information to my Master DNA spreadsheet, and I’m surely glad I took the time to do the analysis BEFORE I copied the segment data into my spreadsheet.  In my excitement, I almost skipped the analysis step, trusting that 23andMe had this right.

All ethnicity results need to be taken with a large grain of salt, especially at the intra-continent level, because the reference populations and technology just haven’t been perfected.  It’s very difficult to discern between countries and regions of Europe, for example.  I discussed this in the article, “Ethnicity Testing – A Conundrum.”

However, it appears that adding parental phasing on top means that instead of a grain of salt, we’re looking at the entire shaker, at least at 23andMe – even at the continent level – in this case, Africa, which should be easily discernable from European. Parental phasing by its very nature should be able to help refine our results, not make them less reliable.

Is this new segment information just showing us the problems with the original ethnicity information?  I hate to even think about this or ask these difficult questions, but we must, because testers often rely on minority (to them) ethnicity admixture information to help confirm the ethnicity of distant ancestors. Are the display tools or 23andMe’s programs not working correctly, or is there a deeper problem, or both?

I think I just received a big lump of coal, or maybe a chunk of salt, in my stocking for Christmas.

Bah, humbug.

22 thoughts on “23andMe’s New Ancestry Composition (Ethnicity) Chromosome Segments

  1. I kept telling people that with 23andme just because you connect at least 1 parent, the chromosomes may not line up correctly. So you shouldn’t assume that all of the top or all of the bottom belongs to one parent, as in my case that did not happen but for some of the chromosomes.

    And I’m ecstatic about this new feature as I wanted to know the exact start/end points of my paternal aunt’s small 0.3% Iberian/European segment that matches up with her niece (my paternal 1st cousin) and also with me, although my portion doesn’t end the same as theirs but is rather cut short. Soon I plan on getting my brother tested to see if he got that same segment. And that other cousin, her brother did not get that segment.

  2. Although this 23andme feature is a “lump of coal” (smiling), this newbie to chromosome analysis is excited about the future aspects of comparing chromosome segments for ethnicity. As I do have ancestry on my mother’s side that has remained hidden and is the “source” for her mother referring to their ancestry as “black dutch”. I don’t have a 23andme account so I mentioned this new feature to a fellow researcher who informed me that something similar is available on Gedmatch. I took a look and ran my mother’s and mine which seemingly are very helpful – but of course their is no parental assigning of chromosome ethnicity. By comparing my gedmatch chromosome ethnicity results to my mother’s gedmatch chromosome ethnicity results, I believe I can infer what ethnicity I inherited from my father. Right? Roberta, have you tried the gedmatch tool for chromosome ethnicity by chromosome? If so, do you have thoughts on its functionality and applicability? If you’ve already blogged about this gedmatch tool, would you please send the link?

    Love following your blog! You keep me on my toes!

    Deborah

    • I repeated your exercise with my son, who has a small East Asian segment on his copy 2 (paternal side) that is not found in his father. But this segment disappears at the 70% level. His most surprising result at the 90% level is an Italian segment on his copy 1, which does show up in me.

  3. Both before and after “The New Experience”, 23andMe identified several “Native American” segments for me, my five siblings, and my daughter. My father had no such segments, and for the most part, 23andMe didn’t show him as the source for any of those I inherited.

    However, they did — and still do — treat a segment on chromosome 15 as paternal. The really odd thing is that four of us inherited this same segment, and it’s attributed to our father for three of us. For the fourth, it’s identified as maternal — or more precisely, as “not paternal” since our mother hasn’t been tested.

    But here’s the really weird part. I passed on this segment to my daughter, and it is correctly identified as paternal for her. But a chromosome comparison with her grandfather shows that she inherited *nothing* from my father on chromosome 15. That’s right, my daughter’s paternal copy appears to precisely match my own *maternal* copy.

    In addition, I have a maternal cousin at 23andMe who matches me at this same location — and has a Native American segment in the same location.

    Clearly, something is amiss.

    • One positive thing, though, is that it IS possible to make very precise comparisons — something Ancestry seems unwilling to allow.

      Based on our downloaded Ancestry Composition data, for all five of us who inherited the Native American segment on chromosome 15, the segment runs from position 78774676 to 94085546. For my brother Curt, my sister CJ, and me this segment is shown on strand 2 — which is supposedly from our father. For my sister Kim, it’s on strand one — from our mother.

      My father’s non-European percentage is shown as 0 in Ancestry Composition, yet he supposedly passed on 0.3% of Curt’s Native American DNA (and the only segment my brother has on strand 2 is the one on chromosome 15); 0.3% to me (again, my segment on chromosome 15 is shown on strand 2); and 0.3% to CJ (guess where).

      Meanwhile, as mentioned, I apparently managed to pass the same segment on to my daughter, but on my maternal copy of chromosome 15. She has no shared segments with her paternal grandfather on this chromosome. Zero. (Their overall sharing consists of 1673 cM in 24 segments, so please don’t raise the question of whether he really IS her grandfather. He is.)

      But my sister Kim’s results aren’t perfect, either. There is a segment on chromosome 18 which is reportedly of paternal origin for her. The thing is, this segment was also inherited by my sister CJ and me, and I passed it on to my daughter.

      Not only is the segment maternal for both CJ and me, but chromosome 18 is another one where my father and daughter share nothing. However, Kim and my daughter DO share in region in which the NA segment is located.

      I think one issue MIGHT be incorrect “child phasing” for my father. It’s based on just one of us, even though all six offspring have been tested. I’m not sure *which* of us is the lucky one, but my guess is that this is the one cause of our results being out of sync.

  4. Nice article Roberta. 23andMe still haven’t assigned me 30% of my DNA. They just put it under Broadly Northwestern European. So I am still waiting to see where 23andMe will give me. I am sort of surprised that they haven’t figure out my 30%.

  5. Pingback: 2016 Genetic Genealogy Retrospective | DNAeXplained – Genetic Genealogy

  6. Bummer! Your results now make me question mine…and I had none of the problems you mention. Both my brother and I (and my dad) have tested at 23andMe. Both my dad and my mom have one parent who is a first-generation American; Dad is Irish on both sides and mom is Italian and German-Danish, and that’s basically what shows up in our ethnicity. No wrong assignments to parents graphically or in the data download, whether at speculative or conservative levels. Ditto for my brother’s results. Interestingly, after I bumped up the segments against my maternal (Italian) half-first cousins, I found that for all but 2 chromosomes, the segments that 23andMe identified as Southern European or better yet Italy were in alignment with where my brother and I matched our cousins.

    All that said, my view is that if you and others are getting garbage data, my (seemingly) properly aligned data is suspect. This would’ve been so cool if they had done it correctly! 😦

  7. My mother is white and my dad is Asian. Both sides in my chromosome pair 2 are completely Asian, which means that they are both from my dad. Considering that my results are still unphased, is it normal?

  8. Pingback: Which DNA Test is Best? | DNAeXplained – Genetic Genealogy

  9. I’d be curious if anything has changed in their methods, and Jasmine’s results, since you wrote this article. I’m reading through their methods page, and it looks like they’re doing the analysis absolutely correctly (and I work in a population genetics lab):

    https://customercare.23andme.com/hc/en-us/articles/115004339467-How-Ancestry-Composition-Works

    It’d be strange to see the problems you describe, given the methods they say they’re using. My parents are getting their DNA done. I’ll have to compare my profile before and after phasing, to see how it changes.

      • I’ve checked my post-phasing reports, and those of my parents, and you’re right—they don’t completely agree, particularly around the edges of small haplotype blocks; and where the ethnicities are very similar (British, say, vs Northern European).

        I’m not convinced this is a sign that their assignments are wrong, per se. I think what they’re doing is phasing the DNA, sure, but then calculating the ancestry independently between the parents and offspring. This could make the ancestry assignment a little different for a few reasons. With offspring, you don’t need to worry about recombination breakpoints, so your Markov model will have different probabilities in parents and children (in the children you only need to decide, is the ancestry changing as we move along the chromosome; in the parents you also need to decide, have the chromosomes been swapped—which will happen around twice per chromosome).

        Another thing that may or may not be at work (if I were doing the analysis, it’s certainly something I’d consider doing): if they condition their ancestry probabilities on mean background ancestry for a given profile, you could very definitely change some of the haplotype calls. If I know for sure that someone has a lot of African admixture, say, I’d be more comfortable making a borderline call as African, than if they were otherwise completely northern European. The same if they had a few obvious Native American blocks. I have a (very) tiny Native American haplotype on a maternal chromosome that my mother doesn’t have. Maybe because I have a bunch of other Native American haplotypes from my father (or maybe because in my mother, the algorithm thought it was likely the site of a recombination event.)

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s