I hope that all of my readers realize that you are literally watching science hatch. We are on the leading, and sometimes bleeding edge, of this new science of genetic genealogy. Because many of these things have never been done before, we have to learn by doing and experimenting. Because I blog about this, these experiments are “in public,” so there is no option of a private “oops.” Fortunately, I’m not sensitive about these kinds of things. Plus, I think people really enjoy coming along for the ride of discovery. I mean, where else can you do that? It’s really difficult to get a ride-along on the space shuttle!
One of the best pieces of advice I ever got was from someone who was taken from my life far too early. I had made a mistake of some sort…don’t even remember what…and he gave me a card that said, “The only people who don’t make mistakes are the people who don’t try.”
This isn’t an “oops” moment. More like an “aha” moment. Or more precisely, a “huh” moment. It falls in the “Houston, we’ve got a problem” category.
So, this week’s new discovery is that there seems to be some inconsistency in the matching to the Anzick kit at GedMatch. Before I go any further, I want to say very clearly that this is in no way a criticism of anyone or any tool. Every person involved is a volunteer and we would not be making any of these steps forward, including a few backwards, without these wonderful volunteers and tools.
I have reached out to the people involved and asked for their help to unravel this mystery, and I’m sharing the story with you, partly so you can understand what is involved, and the process, partly so that you don’t inadvertently encounter the same kinds of issues and draw unrealistic or incorrect conclusions, and partly so you can help. If there has been any common theme in all of my articles in the past week or so about the ancient DNA articles, it has been that we really don’t understand what conclusions to draw yet…and we still don’t. So don’t.
Let’s introduce the players here.
Felix Chandrakumar has very graciously prepared the various ancient DNA files and uploaded them to GedMatch. Felix has written a number of DNA analysis tools as well.
John Olson is one of the two volunteers who created and does everything at Gedmatch, plus works a full time job. By the way, in case you’re not aware, this is a contribution site, meaning they depend on your financial contributions to function, purchase hardware, servers, etc. If you use this site, periodically scroll down and click on the donate button. We, as a community, would be lost without John and his partner.
David Pike is a long time genetic genealogist who I have had the pleasure of working with on a number of Native American and related topics over the years. He also has created several genetic genealogy tools to deal with autosomal DNA. David prepared the Anzick files for some private work we were doing several months ago, so he has experience with this DNA as well. Dr. Pike has a great deal of experience analyzing the endogamous population of Newfoundland, which is also admixed with Native Americans.
Marie Rundquist, also a long time genetic genealogist who specializes in both technology and Acadian history along with genetic genealogy. Acadians are proven to be admixed with Native Americans. Marie shares my deep interest and commitment to Native American study and genetics. Furthermore, Marie and I also share ancestors and co-administer several related projects. As you might imagine, Marie and I took this opportunity immediately to see if she and her mother share any of Anzick’s segments with me and my mother.
So, a big thank you to all of these people.
When Felix originally e-mailed me about the Anzick kit being uploaded to GedMatch, as you might imagine, I stopped doing whatever I was doing and immediately went to study Anzick and the other ancient DNA kits.
I wrote about this experience in the article, “Utilizing Ancient DNA at GedMatch.”
As part of that process, I not only ran Anzick’s kit utilizing the “one to many” option, I also compared my own kit to Anzick’s. My proven Native lines descend through my mother, so I ran her kit against Anzick’s as well, at the same thresholds, and I combined the two results to see where mother and I overlapped.
I showed these overlaps in the article, along with which genealogy lines they matched by utilizing my ancestor matching spreadsheet.
Everything was hunky dory…for then.
The next day, I received a note from Felix that the Anzick kit may not have been fully tokenized at GedMatch previously, so I reran the Anzick “one to all” comparison and wrote about those results in the second article, “Analyzing the Native American Clovis Ancient Results.” Because it wasn’t yet fully processed originally, the second results produced more matches, not fewer.
I wasn’t worried about the one to one comparison of Anzick to my own kit, because one to one comparisons are available immediately, while one to many comparisons are not, per the GedMatch instructions.
“Once you have loaded your data, you will be able to use some features of the site within a minute or so. Additional batch processing, which usually takes a couple of days, must complete before you can use some of the tools comparing you to everyone in the data pool.”
So, everything was stlll hunky dory.
The next day, Marie and I had a few minutes, sometime between 2 and 3AM, and no, I’m not kidding. We decided to compare results. I decided it would be quicker to run the match again at GedMatch than to sort through my Master spreadsheet, into which I had copied the results and added other information. So, I did a second download of the Anzick comparison, utilizing the exact same thresholds (200 SNPs, 2cM, and the rest left at the default,) and added them to a spreadsheet that Marie and I were passing back and forth, and sent them to Marie. I noticed that there seemed to be fewer matches, but by then it was after 3AM and I decided to follow up on that later.
Not so hunky dory…but I didn’t know it yet.
The following day, Dr. Ann Turner (MD), also a long-time genetic genealogist, posted the following comment on the article.
“These results, finding “what appear to be contemporary matches for the Anzick child”, seemed very counter-intuitive to me, so I asked John Olson of GEDMatch to look under the hood a bit more. It turns out the ancient DNA sequence has many no-calls, which are treated as universal matches for segment analysis. Another factor which should be examined is whether some of the matching alleles are simply the variants with the highest frequency in all populations. If so, that would also lead to spurious matching segments. It may not be appropriate to apply tools developed for genetic genealogy to ancient DNA sequences like this without a more thorough examination of the underlying data.”
I had been aware of the no-calls due to the work that Dr. David Pike did back in March with the Anzick raw data files, but according to David, that shouldn’t affect the results.
Here’s what Dr. Pike, a Professor of Mathematics, had to say:
“Yes, these forensic samples have very high No-Call rates, which may give rise to more false matches than we would normally experience. Also, be aware that false matches are more prone to occur when using reduced thresholds (such as 100 SNPs and 1 cM) and unphased data. In this case I don’t think there’s any way around using low thresholds, simply because we’re looking for very small blocks of DNA (probably nobody alive today will have any large matching blocks with the Anzick child).
On the assumption that there will be a nearly constant noise ratio, meaning that most people will have about the same number of false matches with the Anzick child, those who are from the same gene pool should have an increased number of real matches. So by comparing the total amount of matching DNA, it ought to be possible to gauge people’s affinity with Anzick’s gene pool.”
Here are Felix’s comments about no-calls as well:
“Personally, no calls are fine as long as there are more SNPs matching above the threshold level because the possibility of errors occurring exactly on no-call positions for all the matches in all their matching segments is impossible.”
Courtesy of Felix, we’ll see an example of how no calls intersperse in a few minutes.
If no-calls were causing spurious matches in the Anzick kit, you’d expect to see the same for the other ancient DNA kits. I know that the Denisovan and Neanderthal kits also have many no-calls, and based on the nature of ancient DNA, I’m sure all of them do. So, if no calls are the culprit, they should be affecting matches to the other kits in the same way, and they aren’t.
Hunky-doryness is being replaced by a nonspecific nagging feeling…same one I used to get when my teenagers were up to something.
A day or so later, Felix uploaded file F999913 to replace F999912 with the complete SNPs from all of the companies. The original 999912 kit only included the SNP locations utilized by Family Tree DNA. Felix added the SNPs utilized by 23and Me not utilized at Family Tree DNA, and the ones from Ancestry as well. This is great news for anyone who tested at those two companies, but I had utilized my kit from Family Tree DNA, so for me, there should be no difference at all.
I later asked Felix if he had changed anything else in the file, and he said that he had not. He provided extensive documentation about what he had done.
I waited until kit F999912 was deleted to be sure tokenizing was complete for F999913 and re-compared the data again. As expected, Anzick’s one to all had more matches than before, because additional people were included due to the added SNPs from 23andMe and Ancestry.
Some of Anzick’s matches are in the contemporary range, at 3.1 estimated generations, with the largest cM segment of 22.8 and total cMs of 202.8.
These relatively large matches cause Felix to question whether the sample is actually ancient, based on these relatively large segments. I addressed my feelings on this in the article, Ancient DNA Matches – What Do They Mean?
Marie and Dr. Pike, both with extensive experience with admixed populations addressed this as well. Marie commented,
“Native DNA found in the Anzick sample hasn’t changed all of that much and may still be found in modern, Native American populations, and that if people have Native American ancestry, they’ll match to it.”
Dr. Pike says:
“I agree with Marie on this… within endogamous populations, there is an increased likelihood of blocks of DNA being preserved over lengthy time frames. Moreover, even if a block of DNA gets cut up via recombination, within an endogamous population the odds of some parts of the block later reuniting in a person’s DNA are higher than otherwise. And it exaggerates the closeness of [the] relationship that gets predicted when comparing people.
I have seen something similar within the Newfoundland & Labrador Family Finder Project, whereby lots of people are sharing small blocks of DNA, likely as a result of DNA from the early colonists still circulating among the modern gene pool.
As an anecdotal example, I have a semi-distant relative (with ancestry from Newfoundland) at 23andMe who shares 3 blocks of DNA with my father, 2 with my mother and 5 five me. As you can imagine, the relative is predicted to be a closer cousin to me than she is to either of my parents!
It doesn’t take an endogamous or isolated population to see this effect.
It can also happen in families involving cousin marriages too, although that would be more pronounced and not quite the same thing as we’re discussing with respect to ancient DNA.”
This addition of other companies SNPs should not affect my matches with Anzick because my kits are both from FTDNA and won’t utilize the added SNPs.
However, I ran my and my mother’s matches again, and we had a significantly different outcome than either of the previous times.
I utilized the same threshold for all downloads and those are the only values I changed – 200 SNPs and 2cM, leaving the other values at default, for all Anzick comparisons to my mother and my kits.
I am not hunky-dory anymore.
These matches, which should be the same in all three downloads, produced significantly different results.
Here are the number of matches at the same threshold comparing me and Mom to the Anzick file:
Me and Anzick
- original download 999912 – 47 matches
- second download 999912 – 21 matches
- 999913 – 35 matches
Mom and Anzick
- original download 999912 – 63
- second download 999912 – 37
- 999913 – 36
And no, the 36 /35 that mom and I have for 999913 are not all the same.
||Matches Between Me, Mother and Anzick
|#1-F999912 original download
|#2-F999912 second download
Of those various downloads, the following grid shows which ones matched each other.
||#1 to #2
||#2 to #3
||#1 to #3
|# of Matches
So, comparing the first download to the last download, of the 19 original matches, we lost 16 matches. In the third download, we gained 8 matches and only 3 remained as common matches. So of 30 total matches between my mother, myself and Anzick, in two downloads that should have been exactly the same, only 3 matches held, or 10%.
Obviously, something is wrong, but what, and where? At that point, I asked Marie to download her and her mother’s results again too, and she experienced the same issue.
Clearly a problem exists someplace. That’s the question I asked Felix, John and David to help answer.
I realize that this spreadsheet it very long, and I apologize, but I think this issue is much easier to see visually. I’ve compiled the matches by color and shade to make looking at them relatively easy.
My matches to the Anzick kit are in shades of pink – the first match download being the lightest and the last one to kit F999913 being the darkest. Mother is green, same shading scheme.
The three columns to the right show the matching segments for each download – shaded in green. You can easily see which ones line up, meaning which ones match consistently across all three downloads. There aren’t many. They should all match.
Obviously this led to many questions that I asked of the various players involved.
My first thought was that perhaps a matching algorithm change occurred in GedMatch, but John assured me that he had made no changes.
Next question was whether or not Felix changed something other than adding the 23andMe and Ancestry SNPs. He had not.
Felix was kind enough to explain about bunching and to do some analysis on the files.
“When you have low thresholds, make sure you don’t allow errors. For example, at 200 SNPs, the default ‘Mismatch Evaluation window’ and in GEDMatch is same as SNP threshold and ‘Mismatch-Bunching limit’ is half of mismatch evaluation window. So, at 200 cM, you are allowing 1 error every 100 SNPs apart from no-calls.
I did some analysis on your phased mother’s kit, PF6656M1 so that at least we know that it is an IBD for one generation. The spreadsheet (below) are segments I found at 2 cM/200 SNPs threshold without allowing any errors.”
Kit PF6656M1 is one single kit created by phasing my data against my mother’s so that we don’t have to run both kits. I had not utilized the phased kit previously, so I was interested in his results.
The results above confirm chromosome matches, 2, 17, 19 and 21, but introduce a new match on chromosome 4. This match was present in the original download, but not in the second or third download, so once again, we have disparate data, except the thresholds Felix used were at a different level.
One of the more interesting things that Felix included is the no-call match information, the three columns to the right. I want to show what the no-calls look like. There are not huge segments that are blank and are being called as matches because they are no-calls, when they shouldn’t be. No calls are scattered like salt and pepper. In fact, no calls happen in every kit and they are called as matches so they don’t in fact disrupt a valid match string, potentially making it too small to be considered a match. Of course, ancient DNA has more no-calls that contemporary DNA kits.
Below are the first few match positions from chromosome 2 where mother, Anzick and I have a confirmed match across all downloads. The genotype shows you that both kits match.
For consistency, I ran the same kits that Felix ran, PF6656M1 and F999913, with the original thresholds I had used, and found the following:
This introduces chromosomes 1 and 5, not shown above. The chromosome 1 match was shown in the first and second download, but not the third, and the chromosome 5 match was shown in the first download only, but not the second or third.
Can you see me beating my head against the wall yet??
In a fit of apparent insanity, I decided to try, once again, an individual download of Anzick compared to my mother and to me, but not utilizing the phased kit – the original F6656 and F9141, and at the original thresholds, for consistency. I wanted to see if the matches were the same now as they were a day or so ago. They should be exact. This first one is mine.
What you should see are two identical downloads. I have color coded the rows so you can see easily – and what you should see are candy-cane stripes – one red and one white for every match location.
That’s not what we’re seeing. The kits are the same, the match parameters are the same, but the results are not. Once again, the downloads don’t match.
I did another match on mother and Anzick, and her results were consistent between the first and second match to kit F999913.
The begs the next question. Have mother’s results always been consistent, suggesting a problem with my kit?
I sorted all of her downloads, and no, they are not consistent, except for the first and second download matches to kit F999913, shown above. The inconsistencies show up in both mother and my kits, although not in the same locations. Recall also that Marie had the same issue.
Something is wrong, someplace. I know that sounds intuitively obvious – NOW. But it wasn’t initially and I wouldn’t even have suspected a problem without running the second and third downloads, quite unintentionally. Most people never do that, because once you’ve done the match, you have no reason to ever match to that particular person again. Given that, you’ll never know if a problem exists.
So, the only Anzick GedMatch matches I have any confidence in at all, at this point, are the few that are consistent between all of the downloads, and I didn’t add the fourth download into the mix. I don’t’ see any point because I’ve pretty much concluded that until we determine where the issue resides, that I won’t have confidence in the results.
The next question that comes to mind, and that I can’t answer, is whether or not this issue is present in contemporary matching kits – or if this is somehow an ancient DNA problem – although I don’t know quite how that could be – since matching is matching.
I haven’t saved any matches that I’ve run to other people in spreadsheets, so I can’t go back and see if a GedMatch match today produces the exact same results as a previous match.
Clearly there is no diagnosis or solution in this summary. We are not yet hunky dory.
What You Can Do
- Run your Anzick and ancient DNA matches multiple times, at the same exact thresholds, on different days, to see if your results are consistent or inconsistent. Same kit, same thresholds, the results should be identical.
- If you have some saved GedMatch matches with contemporary people, and you are positive of the match thresholds used, please run them again to see if the results are identical. They should be.
- No drawing of or jumping to conclusions, please, especially about ancient DNA:) It’s a journey and we are fellow pilgrims!
If your results are not consistent, please document the problem and let the appropriate person know. I don’t want to overwhelm John at GedMatch but I’m concerned at this point that the problem may not be isolated to ancient DNA matching since the issue seems to extend to Marie’s results as well.
If your results, especially to Anzick, from previous matches to now are consistent, that’s worth knowing too. Please add a comment to that effect.
Thoughts and ideas are welcome.