23andMe Introduces the Automated Family Tree

23andMe DNA kits are on sale through October 15, 2019.

Clustering methods are all the rage in genetic genealogy these days, for good reason.  They offer a tidy visualization of which of our DNA matches also share DNA with one another.  That’s information we can use to corroborate our biological family trees and to identify brick-wall ancestors.  What clusters don’t do, however, is generate a family tree for you.

23andMe just changed all that!

A few days ago, 23andMe surprised us all with a new “auto-built” tree feature based on shared DNA clusters.  The tool, simply called Family Tree, is currently available to customers who have signed up for 23andMe’s beta testing program.  It’s accessed via the Family & Friends menu.  (You can sign up, too, using the instructions at the bottom of this post.)

The tree generated for my father looks like this. I’ve replaced the match names with letters to protect their privacy.  These matches all share 50 cM or more with my father.

What’s going on here?

23andMe’s FAQ says:

When you participate in DNA Relatives or make one-to-one connections with other 23andMe customers, 23andMe compares your DNA and predicts your genetic relationships.

To build your predicted family tree, we then go a step farther: we also look at the relationships between your relatives, and calculate many different ways you could all be related to each other. The tree you see represents the most likely tree out of all the possibilities we calculated.

Without explicitly stating so, 23andMe has clustered the 15 matches who share the most DNA with my father (including myself) and fit them into a hypothetical tree based on whom else they match, how much DNA they share with us and with one another, and their ages.

 

Genetic Clustering

To see what I mean, compare the family tree above with this cluster report from the DNAgedcom Client for the same set of matches.  Matches are listed in order across the top and down the left side.  Where two matches also match one another, the intersecting box is colored in.  The cells on the diagonal are always filled in, because that’s where each match is compared to themselves.

 

Match G is a first cousin once removed to my father.  Match G also shares DNA with Matches A through F and with Match H, but Match H doesn’t share DNA with any of the others.  23andMe’s system has figured out that Match H is related through one branch of the tree while Matches A through F are related through another.

What’s more, the algorithm is comparing how much each pair shares to estimate how they are related to one another.  For example, Match C is the father of Match D, so 23andMe put them as parent–child in the tree.  That one’s easy, of course.

A more sophisticated analysis suggests how Matches A, B, and C might be related to one another.  Match C shares approximately 800 cM with both Match A and Match B, which is in the first cousin range.  However, Matches A and B only share about 200 cM with one another, so the three of them can’t all be first cousins to one another.  An alternate scenario is that Matches A and B are 2nd cousins to one another and Match C is their great uncle.  This possibility fits better to both the DNA data and to their ages (which we users can’t always see but 23andMe’s system can).

And that’s precisely what turned out to be the case based on a quick dive into their family trees.  I already knew how Match C was related to me, and it only took a few minutes of online research to corroborate that Matches A and B are the grandchildren of his brothers.

 

You Can Annotate the Tree

An exciting feature is that you can annotate the tree with the names, birth/death years, and locations of the people in the tree.  Eventually, the deceased people in your tree will be visible to your matches, much like at 23andMe’s competitors, enhancing your genealogical research and that of your matches.

Tree annotations are still in development, and there are limitations.  For example, Maria Eva Weigel in the screenshot above was married twice.  Her daughter Elizabeth Mühl was the only child of her first marriage, so the descendants of her second husband (Jacob Schmitt) are half cousins to us.  There is currently no way to indicate half relationships in the system.

Nor is there a way to correct placements that turn out to be wrong.  All of the matches in my father’s auto-tree were on the correct branches, but several were in the wrong generation (see below).  Other users are reporting that maternal relatives are linked to the paternal side and vice versa.  23andMe plans to introduce a tree editor to allow us to correct these issues, but it’s not available yet. (That’s the down side of participating in a beta program.)

 

Cautions

Key to remember is that the tree isn’t “sided”, meaning that you can’t tell whether your matches are maternally or paternally related to you by which side of the diagram they’re on.  For example, in my auto-tree, my paternal matches are on the left, while in my father’s tree, his paternal matches are on the right.

Also, the system doesn’t appear to be using haplogroups or X segments to “side” people.  My father shares a substantial amount of DNA with Match D on the X chromosome, which men only inherit from their mothers, but 23andMe gave no indication that Match D is a maternal match.  Perhaps they will use X DNA and haplogroups to “side” our trees in the future.

Finally, bear in mind that the proposed relationships are educated guesses, not fact.  Of 13 matches in my father’s tree (excluding myself and the child of Match C), I’ve been able to place 10 of them definitively.  All of them are positioned on the correct branches in the auto-tree, but five of them are in the wrong generation.  Four of those five are estimated to be one generation more distantly related than they really are, e.g., a 2nd cousin once removed said to be a third, while one is a true third cousin said to be a 2nd cousin twice removed.  I expect the estimates to improve over time as they gather more information from real-world relationships and as the database grows.

Oh, and the font size … oh my!  The tool was clearly programmed by someone who has 20-something year old eyes, whereas mine are a few decades older.  The names are so small they’re almost impossible to read.  That’s what customer feedback is for, right?!

 

Sign Up for the Beta Program

Enrolling in the beta program gives you access to new features early.  Companies beta test newly developed products to see how customers respond.  For that reason, you should understand that beta tools may have unexpected glitches and are likely to change over time as the company continues to perfect them.  If you’re not comfortable with that, wait for the finished product rather than joining the beta program.

If you’d like to go ahead and join, here’s how:

  1. Log into your 23andMe account and click on your name at the top-right to get a pull-down menu.  Click “Settings”.
  2. On the next page, scroll down to the “Preferences” section and click the hot pink “Become a tester” button.
  3. That’s it!  You should be enrolled in the Family Tree beta program within a day or so.
  4. The tree is accessed via the Family and Friends menu in your 23andMe account.
  5. The system may walk you through a couple of introductory screens before you can see the tree.  These are expected to change as 23andMe develops the system.

 

35 thoughts on “23andMe Introduces the Automated Family Tree”

  1. As expected, it doesn’t handle half and double relationships well. My father had double cousins though his mother’s sister who was married to his father’s uncle. My father also had half-sisters through his mother. Additionally, several descendants from my grandfather’s siblings have tested. All of these groups appear as being descended from the wrong couple. There are four groups of us that are related in varying ways. Two groups are related to both my grandparents, one group is related to my grandfather only, and the other group related to my grandmother only. It has confused the heck out of their predictions for this mess.

    1. It’s not incorporating half relationships at all yet, and it’ll be a long while before *any* automated method can deal with double relationships! My hope is that once they introduce the tree editor, the AI will be able to learn from the corrections we make to improve future predictions.

      1. I do have one half 1C1R that is listed as 1C2R so it is assigning the distance incorrectly although it is an equivalent relationship.

    2. The double relationships definitely complicate things! My father was born in the Smokey Mountains and it’s not uncommon for me to be distantly related to all four grandparents of my paternal matches. 🙁 On the plus side, I work in software testing so this stuff is just job security for me. 🙂

  2. Hi Leah,
    I was amazed to see this Family Tree suddenly appear on 23andme, but I have possibly a slightly silly question…

    How do I export it? (Basically to show it to my mum, who’s the amateur genealogist).

    Maybe I just need to open 23andme on something bigger than my phone and screen-shot it…?

    Or is there a better way…?

    Unrelated thought: I’ve just been talking with the first person I didn’t know on 23andme, about our shared relatives. It would make sense to be able to fit my tree in with hers. Might that ever happen? Should it happen…?!?

    Matthew

    1. There’s no export yet. For the blog post, I used a screenshot from my computer.

      Here’s what their FAQ says about seeing other user trees:

      Who can see the information in my Family Tree?
      Your full Family Tree is visible only to you — no one else, including your DNA Relatives, can see the unique configuration of people that make up your Family Tree.

      Self-reported information about living relatives: Only you can view information you add about living relatives. All relatives you write in are assumed to be living until you mark them as deceased.

      Self-reported information about deceased relatives: If you opt in to DNA Relatives or connect with other users, those users can see information you’ve added about deceased relatives.

      1. I’ll log in on something bigger than my phone! 😉

        I need to learn more about opting in to ‘DNA Relatives’….

  3. Love the new 23andme tool!
    Ancestry keeps making tweeks and restoring old functions to their original match system and the Thrulines (why they won’t just call it a Pedigree is beyond me.)

    1. Technically, ThruLines isn’t a pedigree. It’s a group of DNA matches who all likely descend from the same common ancestor.

  4. I just don’t get it..if like me you where adopted …you have nothing and no one to start with ..so all that is very sad for us …

  5. Leah,

    Once I’ve filled in names, will the tree start expanding? Or is what I have right now “it” until a new match comes along?

    Lisa

  6. How dependent is this on close relationships? For example I have one half 1C1R that I know is related to me through my maternal grandmother. Everyone else is third cousin or higher and it’s a crapshoot unless we share the half 1C1R. There is another that could be related on my father’s side based on the last name. Then there is the grandchildren of my brother. That’s no help since they descend from my mother and father…lol Interesting though. I got both sets of grandparents reversed in my first attempt..geeze. I thought each was male on the left and female on the right but closer inspection proved that wrong. I think. Maybe once more people fill out trees it will be more easily filled in?

    1. They’re not assigning sides at this point, and a lot of people are reporting that their maternal relatives are getting mixed up with paternal and vice versa. It’s early days. It’ll be really exciting if they start using X-DNA, haplogroups, and segment mapping to assign sides.

  7. What I find bizarre is the Family Tree has connected one of my maternal 2nd cousin and two of my maternal 2nd cousins once removed on my father’s side. The reason I call this bizarre is that my father also tested at 23andMe, is linked to me, and has no connection whatsoever to my 2nd cousin or the 2nd cousins once removed.

    These cousins

    (1) Do not appear in my father’s DNA Relatives list
    (2) Do not appear as “Relatives in Common” with either me or any of my five full siblings
    (3) Share **zero** DNA with my father when directly compared to him in the chromosome browser
    (4) Do not appear anywhere in my father’s beta “Family Tree”
    (5) Appear in the Family Trees of all five of my full siblings, but as **maternal** relatives
    (6) Appear in **my** Family Tree as great grandchild (and 2nd great grandchildren) of my father’s paternal grandparents

    So basically, my 2nd cousin should be my father’s 1st cousin once removed even though they share absolutely NO DNA!

    I contacted 23andMe customer service, including an explanation and a screen shot. The response?

    “It’s important to note that our relationship prediction algorithm has been updated, so the predictions you are seeing are the most accurate predictions we can provide at this time.”

    Huh? What I want to know is how the algorithm can possibly connect my 2nd cousin to me through my father, when it’s very clear my father and my 2nd cousin ARE NOT RELATED.

    What in the world is the basis for the “relationship prediction algorithm” — updated or not — linking my 2nd cousin through my father? A game of darts?

    1. Similar problems here. In order for the 23andMe tree to be correct in some places, my maternal grandfather would need to have children with paternal grandfather (yes, I wrote that correctly!). The problem with the 23andMe tree seems to be related to two issues: 1) the children of my maternal grandmother’s second marriage and 2) no DNA matches on my paternal grandmother’s side. The children of my maternal grandmother’s second marriage only share DNA with me, not the descendants of my grandmother’s first marriage which there are also DNA matches to. The “algorithm” does not know what to do with these “hanging” close relatives so it assigns them to my paternal grandmother.

      1. Agreed. They’re not fully considering how our matches match one another. I’m still excited at the possibilities of this tool, though. I hope they continue to invest in it.

  8. I just can’t imagine the programming that would cause this result: that three relatives would be assigned to the side of a parent who shares *no* DNA with them.

    It doesn’t appear that my father is connected incorrectly — he does at least appear as my father. But on the other hand, way back in the days when 23andMe still had the experimental ABO Blood Type lab, they could not determine what my father’s was. The reason was two miscalled SNPs, which I brought to the attention of the lab as probable miscalls. They assured me that this was extremely unlikely.

    But then one day they “corrected” the call on one SNP for a whole group of customers, including my father. This eliminated a discordant call between my father and four of his offspring. A bit later, without ever saying anything, they also corrected the call on the other SNP, which eliminated a discordant call between my father and his other two offspring.

    After they updated both calls, the lab *ought* to have worked. But it never did, apparently because they never made the updated data available to the lab. Eventually, they dropped the lab altogether. I’m not sure why, but I suspect it had something to do with keeping the FDA happy.

    My point is, even glitches don’t just happen. They are caused, but by what in this case is unclear to me.

  9. Here’s more of the sort of nonsense I’m seeing with the beta Family Tree.

    One of my profiles is showing a relative in his tree that is predicted to be a 2nd cousin. Only, their DNA sharing is 705 cM.

    Clicking on the relative’s circle and then “See other possible relationships” leads to the explanation that “Statistically, [relative’s name] is most likely your 2nd cousin …”

    Really? Whose statistics?

    Yes, I know the mantra. Beta features are expected to have glitches. I agree, but these are not — in my opinion — beta-level glitches. They’re alpha, and should have been worked out during in-house testing.

    1. I should have noted that with that same person mentioned above, another individual who shares 670 cM with him is identified as a 1st cousin. “See other possible relationships” once again explains that “Statistically, [relatives name] is most likely to be your 1st cousin.”

      What statistics are these that make someone sharing 705 cM most likely to be a 2nd cousin, while someone sharing 670 cM is most likely to be a 1st cousin?

      However, I think I know what the problem is in this case. The profile in this instance belongs to someone whose half sister also tested. She is in fact *confirmed* as his half sister in both her profile and his. Only, their beta Family Trees labeled them as uncle and niece.

      Now, according to the explanation given by Family Tree, both the amount of DNA shared AND self-reported ages are used to determine relationship. These two were born in 1955 and 1956, and I presume that 23andMe can do the math.

      By their own criteria, they should have been identified as most likely to be half siblings — which they are. But, since they were identified as uncle and niece, this created another problem.

      The person identified as my profile’s *2nd* cousin, despite sharing 705 cM, doesn’t share anything with the niece — who should be a FULL niece, and would therefore be related to all the people her uncle is related to. She might not share DNA with all of them, but she would definitely share DNA with her “uncle’s” 1st cousins. They would be her 1st cousins once removed — and it’s just not conceivable that someone would share no DNA with a 1st cousin once removed.

      However, a person *might* not share DNA with a 2nd cousin once removed. So by labeling the person as a 2nd cousin, instead of a 1st cousin, it works. Sort of.

      Meanwhile, the person who shares 670 cM and is labeled a 2nd cousin is on the shared side of the two half siblings. In fact, the half sister’s sharing with this person is 621 cM.

      Even more interesting, though, is the proposed “1st cousin once removed” of the “uncle”. The two share 209 cM. This same individual is listed as the “niece’s” 2nd cousin, which would make sense IF the relationship to the “uncle” is correct. How much does the “niece” share with her “2nd cousin”? 325 cM.

      1. The first sentence in the next to the last paragraph should read “who shares 670 cM and is labeled a 1st cousin”, not 2nd cousin.

      2. I would love to get a peak at the assumptions they’re making on the back end. I’m really excited at the potential for this tool, but you and I both agree it needs work.

  10. I’m pleased to note that my maternal 2nd cousin and his daughter and 1st cousin’s son — who are my 2nd cousins once removed — are no longer being shown as attached through my tested and linked father. He shares no DNA with these maternal cousins, let alone an amount that would be consistent with being 1st cousin once removed or 2nd cousin.

    Unfortunately, the problem is now showing up in my youngest sister’s tree. My father is also linked to her as her father, yet Family Tree shows her as connected to these maternal cousins through our father.

    But, at least they have corrected my tree — which shows that they *can* get it right. The more troubling issue is that they *already* had it right for my sister. Apparently, fixing my tree messed hers up … and I wish I knew why. (Our trees actually should be virtually identical, except for descendants on either side.)

    Still … I’m guardedly optimistic about the future usefulness of this feature.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.