My father is born of unknown father and did a 23 and me DNA autosomal test. I thought I could use this test to potentially identify some relatives on this side but only very distant ones (at best 3rd cousin) could be matched.

In order to split the problem in smaller chunks, I was then thinking about trying to group the DNA segments into 4 clusters belong to each of his grand-parents. For that, I downloaded all his matches (more than 2000) from MyHeritage. I got regions with significant overlap with some with more than 400 matches.


The plan was to select some of these regions with overlap and see how much each match has an overlap with them. So I would get a matrix with the regions as column and the matches as rows and containing the overlap in CMs as value.

Then I would just need to calculate the correlation between the columns/regions, use some kind of clustering method and I would see clean clusters showing up. But of course that didn't happen, I'm just getting some kind of mess where the regions cannot be cleanly clustered.


I tried to filter the regions by number of overlap with a minimum and maximum value, minimum to have some overlap and maximum because I thought that there could some false-positive regions. But with little success.

What would you recommend as a way forward to do this grand-parents level clustering?

I think I understand why it does not work, because I have no way to tell on which chromosome (paternal or maternal) a segment is matching so I could have 2 matching segments from 2 people around the same position that wouldn't match with each other.

Is there a way around this problem?

I was thinking about matching on the start or end position since it is unlikely that two segments would have the same start or end without being related.

Or am I again missing something?

3 Answers 3


Interesting question and nice analysis that you did. I actually also looked into segment analysis, quite recently. Up until then, the clustering approach I created were based on shared match data. If you have enough high cM matches, between the range 400-90 cM, you can get around 4 clusters. Most people don't get them and by using segment data and lower cM matches, you are looking at many clusters.

Anyway, like I said, if you want to perform clustering using segment data, I also have something for that. It's called AutoSegment: https://www.geneticaffairs.com/features-autosegment.html

Here is how it works. It compares all segments and keeps overlapping segment that overlap enough (you can select how much cM). Based on these segment overlaps, I create segment clusters. These are used to cluster your matches. Since they are universal, you can then cluster matches from 23andme, FTDNA, MyHeritage and GEDmatch into one clustering (using the hybrid AutoSegment).

Here is our FB group: https://www.facebook.com/groups/GeneticAffairs, lots of folks there that can help. This YouTube also goes into detail about AutoSegment: https://www.youtube.com/watch?v=LST8jqY7C2A

  • 1
    Hi Evert-Jan, thanks for your response, you definitely have interesting features on your website, maybe it's a better idea to use existing solutions instead of trying (and probably failing) to redevelop them myself ;)
    – C. V.
    Jan 5, 2021 at 22:38

I went this same route trying to identify where my great-great-grandfather came from in Germany. Few matches seemed to come from that part of my ancestry, but I identified two who did. Both, however, had similar problems, and didn't know where that part of their ancestry had come from either.

Rather than focusing on the cm overlaps, clustering by which matches match each other works far better. That's what Genetic Affairs does; they automated what has become known as the "Leeds method", after Dana Leeds who came up with it. She gives an introduction at https://www.danaleeds.com/dna-color-clustering-the-leeds-method-for-easily-visualizing-matches/

Her method only describes splitting using grandparents, and hints at taking it to the great-grandparent level, but I found that, over time, if you slowly work through your matches such that a sufficient number are identified out to the great-grandparent level, it can even work at the the great-great-grandparent level, assuming you don't have endogamy in your ancestry.

Side note: though I've now managed to discover and document that particular line out to between the 6gg and 9gg level, I should note that I've still yet to find the connection to those original two matches I'd started out focusing too closely on.

  • Hi Brian, Thanks for you answer, I'll look at that.
    – C. V.
    Dec 16, 2020 at 19:31

Try clustering by which matches match each other (like at Genetic Affairs) instead of by which matches overlap with each other. Start and end positions can be clues, but I think you would miss a lot of valid triangulated segments if you tried to sort that way.

  • Hi Tanya, Thanks for your answer. I actually tried before but I was getting the same kind of messy correlation matrices. But again, I was associating the matches using their overlap regions which ends in the same problem as mentioned above. I'll try to see if I can find the real relation between matches.
    – C. V.
    Dec 9, 2020 at 19:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.