Site icon DNAeXplained – Genetic Genealogy

23andMe Automatically Creates Tree Using New Technology – Relationship Triangulation

Roberta Estes

23andMe has introduced a new tree feature that automatically creates a “tree” for you based on your predicted relationships to others and their predicted relationships to each other. 23andMe hasn’t coined a term for this, but I’m calling it Relationship Triangulation.

Let’s look at traditional forms of triangulation and how Relationship Triangulation is different.

Triangulation

Segment Triangulation – In traditional triangulation for genetic genealogists, we match the same (reasonably sized) segments between a minimum of 3 not-closely-related people to assign that segment to a common ancestor. Of course, in this scenario, you need to know who your ancestors are – at least some of them.

Once enough people test and match on that segment, hopefully at least a few will be able to identify a common ancestor or minimally, an ancestral line. You can read about this type of triangulation here and a more detailed article here.

Tree Triangulation – Genealogists use what I refer to as “tree triangulation” when segment data isn’t available, like at Ancestry, or when they are seeking to determine parentage.

Adoptees use this method of triangulation where they look at the trees of their closest matches, hoping to discover a common ancestor in their close matches’ trees – because that points the way to a descendant of those ancestors who is their biological parent.

I’ve described this technique in the article, Identifying Unknown Parents and Individuals Using DNA Matching.

Relationship Triangulation

What 23andMe is doing with relationship triangulation is different yet.

They are using the same techniques used by genetic genealogists manually to try to place their matches in their trees. In essence, we perform the following steps:

  1. Look at the predicted relationship provided by the testing company in order to get an estimation of where that person might fit in our tree – in other words, how far back in the tree the common ancestor will be found. First cousins share grandparents, second cousins share great-grandparents, etc. Additionally, I utilize the relationship table at DNAPainter to view alternative relationships based on total shared DNA.
  2. Look at who we and our match matches in common. For example, if the match also matches my first cousin on my father’s side, there’s a high probability that’s where the person fits on my tree. If they also match with my second and third cousin on my father’s side, especially if they match on the same segment (traditional segment triangulation) then I not only know which “side” they match on, I can push the match back several generations to the common ancestor or ancestral couple I share with the third cousin.
  3. If I also have detailed haplogroup information about the ancestors in my family tree, and my match also has haplogroup information, I can use that to rule out or include lines as potential placements. 23andMe provides haplogroup estimates at a high level, not at a detailed level, but they can still serve to rule out various lineage placements on the tree when compared with other people. For example, if a person’s mitochondrial haplogroup is J1c, they can’t be the child of a female ancestor whose haplogroup is H1a. This is exactly why I have constructed a DNA Pedigree Chart.

What is 23andMe Doing?

I don’t have any inside information about exactly how 23andMe is constructing this tree, but they published the following. The difference, of course, is that they have automated the process and they don’t have customers’ trees to work with.

This information is found on your account at 23andMe under the 3 dots at the upper right-hand side of the tree itself, which I’ll show you how to access in a moment.

23andMe will probably provide additional information and make adjustments as this feature comes out of beta – including the ability to modify relationships which are inaccurate – and several are.

Beta is Required

In order to utilize the new tree feature, you must enable Beta for your account, which you can find under your name in the Setting area.

If you need help, I wrote detailed instructions for enabling Beta in this article.

Accessing Your Tree

You can access the new tree feature in 2 ways.

At the top, under Ancestry, select DNA Relatives

Scroll to the very bottom.

To see your tree, click on the Family Tree link.

The second way to access your tree is under DNA Relatives. When you click on any specific person, if they are in your tree, you’ll see the following:

The “View your Family Tree” button takes you to your family tree.

Your Squiggly Relationship Triangulation Tree

Please click to enlarge this tree in the image below. The font for names is small, but at 23andMe, you can zoom in on various sections.

Click to enlarge

When you first see your tree, it looks a bit odd and takes a some getting used to because it looks different than the trees genealogists are used to working with.

Here’s what the blue box contents say:

I tried to see if there was a way to add people in my tree. When you click, the node enlarges, but there isn’t anything to be done yet.

How Is 23andMe Creating Relationship Trees?

23andMe is constructing my family tree based on how I’m related to people, and how they are related to each other as well. I wouldn’t be surprised if they are also using their basic haplogroup information.

Let’s look at how this works.

My two children have tested. I did not tell 23andMe that they are my children, but because both of their mitochondrial haplogroups are the same, and I’m related to both of them in a parent/child relationship, and they are related to each other as siblings, I have to be their mother. That’s the only combination of relationships that works for the three of us.

If one of them was my parent, they would not be related to each other as siblings.

One could be my parent, and one my child, but then, again, they would not be related to each other as siblings.

Furthermore, my son would carry my mitochondrial haplogroup, but I would not carry my father’s mitochondrial haplogroup, so my relationship to him cannot be that he is my father. It would be very unlikely for a father’s mitochondrial haplogroup to match a child by happenstance although that’s not impossible, especially with only partial haplogroups with no full sequence testing. If you’re interested in the difference, you can read more about that here.

Lastly, my children would be related to a subset of the people I’m related to, but (in general) sharing less DNA, so a generation removed. If one of them was my parent, I would be related to a subset of the people they are related to.

23andMe is also looking at who shares DNA with other relatives.

In this DNA Relatives chart comparing me with cousin Laura, you can see that she and I share a common segment of DNA with my twin and with “J”, meaning we have segment triangulation, although my V4 kit doesn’t count for triangulation because it’s a duplicate. Additionally, Laura and I both match Trista, but Laura, Trista and I don’t share a common DNA segment between the three of us.

Laura, “J,” Trista and my V4 test are all shown in my tree. We’ll analyze those matches momentarily.

My Two Results

You’ll notice that I have two tests at 23andMe. I took a V3 test and another V4 test when it was introduced to see if my results were the same. 23andMe accurately calculated this as a “twin” relationship.

In general, the results are close, but not exactly the same. Interestingly, my V3 tree has 2 more cousins than my V4 tree. Of course, the need to “fit people in” on the tree means the layout looks different between the two trees, so I’m only going to work with one set of results in order to reduce confusion.

It’s confusing to look at both trees side by side, because these trees are auto-generated to “best fit” the branches with the available screen space – so the branches are not in the same place on both trees. As more cousins are added to your tree, the layout will change.

Paths

For each person in your created tree, you can click on the person you match and the path from them to you highlights.

I have to laugh, because there is a quilt pattern named “Drunkard’s Path,” and it’s much straighter than this path from my second cousin, Patricia, to me.

Of course, today there are no names on these “?” ancestor nodes but I know who they are.

Let’s take a look to see if these placements are accurate.

Identifying Nodes on the Tree

Based on where my matches are placed, I determined the identity of the “connecting couple” nodes on my tree and utilized Snagit to mark up the 23andMe tree, below.

You can see that I’ve labeled my mother and father, based on what I know about the various people that I match. For example, I know where my Ferverda cousins reside on my tree, so those cousins must be on my mother’s side.

Conversely, my Vannoy cousins must be on my father’s side.

In my case, my mother is on the left and my father on the right, which is backwards to the normal genealogist pedigree way of thinking. However, without any tree information at all, 23andMe is flying completely blind.

Click to enlarge

I analyzed each of my 18 matches that 23andMe placed on my tree, of my 1553 total matches at 23andMe.

Accurate Placements

The gold stars indicate accurate placements. There are a total of 7 gold stars, BUT, of those 7, one is my own second kit which clearly is my “identical twin.” Two are my children whose accounts I manage. Of the balance, there are:

There are a total of 4 accurate placements that are not immediate family members whose kits I manage.

Why might a placement not be accurate? 23andMe says the following:

This is exactly why I utilize the DNAPainter segment tool to evaluate different possibilities for the matching total centiMorgans.

Unknown Placements

Green boxes on my tree are people who are unknown, but based on their common matches with known cousins, they are in the right general area of my tree. I wish I could tell more.

With 23andMe’s historic lack of tree support, and few people having added surnames or recently, ancestors through Family Search, it’s impossible to determine how these matches descend from common ancestors. The best I can do based on common matches and shared DNA is to determine and assign “sides” and sometimes roughly how many generations back to the common ancestor.

Hopefully, once people can actually identify ancestors on their trees, this may well improve. I’m concerned that so many years have passed with no tree support that many people aren’t signing in anymore, will never notice the tree feature and won’t add anything. I hope I’m wrong.

There are only 4 people in my tree whose relationship to me is unknown. Perhaps I should send them a message. I do know approximately where they belong on the tree, so I could at least ask some leading questions with surnames I think they might recognize.

Uncertain Placement

The gold box is an uncertain placement. Based on common matches, this person descends from Ollie Bolton’s parents’ line, not from Joel Vannoy and Phebe Crumley. However, Diane also matches Patricia M who descends from my mother’s line, so she could be doubly related to me. Those lines are not even from the same state, so this person is a bit of an anomaly.

This causes me to wonder how 23andMe will handle situations where a match does descend from both sides of a person’s tree.

I should send her a message too.

Inaccurate Placements

The red boxes indicate placements that are known to be incorrect, and why they are incorrect. When the placement needs to be moved to another connecting node couple, I’ve drawn a red arrow.

I’ve numbered the red boxes so we can discuss each one.

In box #1, Cheryl is identified correctly as to her relationship with me, as is “J,” but their relationship to each other is inaccurate. They are half siblings, not first cousins.

In boxes #2 and #3, these people are shown descending from my grandparents, when in reality, they descend from my great-grandparents. Of course, this makes their relationship to me inaccurate too.

In box #4, my half-grand-niece, meaning my half-sister’s granddaughter is shown as descending through another child of my grandparents, when in fact, she descends from my father and his first wife. My father needs another spouse in the chart and the relationship needs to be calculated accordingly.

In box #5, both individuals are actually one more generation further down the tree than they are shown, so the relationship needs to be recalculated.

Summary

How did 23andMe do with their automated Relationship Triangulation tree construction?

Given that this tree was entirely constructed by 23andMe without any genealogical foreknowledge, based only upon genetic relationships to me along with genetic relationships of my matches to each other – this isn’t bad at all. It’s certainly a start.

Adoptees must think they’ve died and gone to the happy hereafter, because all other vendors’ tree support requires you to actually HAVE a tree of some description. Of course, adoptees and people seeking an unknown parent don’t have trees for their unknown parents.

23andMe creates a tree from scratch for you based only on genetic relationships – Relationship Triangulation.

While only one third of these matches are accurately placed today, all were at least on the correct side of the tree, with one confusing possible exception. Genealogists already know that things like pedigree collapse and endogamy will complicate efforts like this.

My mother’s side was accurately identified to appropriate great-grandparents, but my father’s side wasn’t as clean, especially where “half” relationships are involved.

It would be interesting to combine the Leeds Method or Genetic Affairs clustering with the 23andMe technology and see what those results look like.

23andMe would be in a much better position had they never obsoleted tree support years ago. They have never been a “genealogy” company, with their focus always having been on medical genetics. Genealogists, with their incessant need to know were the perfect people to attract to test. 23andMe has always given us “just enough,” but never trees or the heavy duty tools we need.

Perhaps the tide has turned and this is their way of reintroducing a hybrid genetic+genealogy tree.

The Future

In the short-term future, 23andMe is going to add the ability to define and modify relationships on this tree, which will help them refine and improve their machine learning tools. This will also help with their medical research initiatives because clearly, the companies they partner with want to know specifically, not generally, about the heritability of medical traits.

Trees encourage genealogists to provide 23andMe with family information for their medical research, while trees clearly benefit genealogists too.

In the longer term, for genealogists, let’s hope that the marriage of genetics, machine learning, trees and technology produces helpful tools.

This Relationship Triangulation tree is interesting and fascinating, if not yet terribly useful. This technology holds a lot of promise and every innovation begins with a imperfect tenuous first step.

Who knows what the future will bring, either at 23andMe or perhaps as other vendors integrate this same type of technology.

You can’t play if you don’t test and 23andMe does not accept uploads.

If you would like to take the 23andMe genealogy only test, click here or if you want the 23andMe genealogy + health test, click here.

If you have already tested, check your account for your new tree and tell me what you think!

____________________________________________________________

Disclosure

I receive a small contribution when you click on some of the links to vendors in my articles. This does NOT increase the price you pay but helps me to keep the lights on and this informational blog free for everyone. Please click on the links in the articles or to the vendors below if you are purchasing products or DNA testing.

Thank you so much.

DNA Purchases and Free Transfers

Genealogy Services

Genealogy Research

Exit mobile version