The Future of Autosomal DNA Testing

[djknox]

Hi

Today's autosomal testing offered by the 3 usual suspects, test +/-800,000 SNPs and can offer acurate cousin matching to the 4th cousin... and reasonable matching to the 6th cousin. However, most records from the past available to us can cover genealogical research back 200 to 250 years... then for most, it just doesn't exist. So it is logical that for autosomal testing to offer a real breakthrough, accurate cousin matching at the 6th to 10th cousin range will be necessary. Does ANYONE know enough about this stuff to advise whether i) is it theoretically possibly to match at progressively earlier generations with increased testing resolution? AND ii) if yes, how far off might we be? I love the idea of autosomal testing... but 200years isn't really a home run for most serious family history researchers! Input seeked!

[EdwardRHill]

I have some matches that have a possible Common Ancestor back to or around the 12th generation. Some families are well documented others are not but if I have one match that share the same grandparent up to 5 generation the paper trail is strong enough for me. Beyond that I would want three or more matches who all claim the same couple as me and the ability to compare Raw Data to see if we share the same segment. If they all don't I would wait and see if another match would come along and claims the same couple and shares the same segment as the match that didn't match the others. The test is only as good as the people who tested want it to be, we all need surname list and Gedcoms to help with our search but some don't or can't add the extra info for what ever reason.

[MFWare]

Autosomal DNA tests were a breakthrough that offered genetic matching outside our all-maternal and all-paternal lines. It identifies matching segments of sufficient length and number to provide confidence that matches are not by chance. Autosomal DNA also provides a "clock" to place the MRCA in time. Unfortunately, after 6 generations these matching segments become precariously short. After 7.5 generations, they are indistinguishable from random matches. If we are to extend the functionality of autosomal DNA tests beyond 7.5 generations, then new breakthroughs are required. I have no reason not to believe that the breakthrough will involve a heretofore unknown feature of DNA or protein.

This is not to say that the breakthrough won't happen. It is to say that the breakthrough has not happened yet and that there is no guarantee that it ever will.

[tomcat]

This matter has less to do with sampling capacity of retail DNA tests than the lack of comprehensive comparative data. While it would be great to be able to test more SNP's, or one's entire genome, the yield of such tests is always limited by the number of people who have been tested to the same degree.

In example, if one wanted to prove or disprove a common ancestor with another with a deep pedigree in Princeton NJ similar to one's own Princeton pedigree, one would like to sample EVERYONE with a deep Princeton pedigree, especially the dead of Princeton NJ! Only by so doing could one approach an estimation of the incidence of particular SNP'S or micro segments in an historic Princeton population upon which to base a likelihood estimate that one shared a common ancestor with that particular other and that common ancestor was a Princetonian,or not.

[mkdexter]

Hi

Today's autosomal testing offered by the 3 usual suspects, test +/-800,000 SNPs and can offer acurate cousin matching to the 4th cousin... and reasonable matching to the 6th cousin. However, most records from the past available to us can cover genealogical research back 200 to 250 years... then for most, it just doesn't exist. So it is logical that for autosomal testing to offer a real breakthrough, accurate cousin matching at the 6th to 10th cousin range will be necessary. Does ANYONE know enough about this stuff to advise whether i) is it theoretically possibly to match at progressively earlier generations with increased testing resolution? AND ii) if yes, how far off might we be? I love the idea of autosomal testing... but 200years isn't really a home run for most serious family history researchers! Input seeked!

Yes it is possible however the probability of FINDING the match is the issue. If it is there it can be found but most of the segments will not be there. That's the issue.

High resolution IBD segment matching is possible. It has been for a few years now. Ancestry is doing it. It allows the IBD segments to be a low as 3cm instead of 7cM, or 5cM.

So it is not an issue of finding the small segment matches, it is an issue of just not that many small segment matches being there in the first place.

In my case I have two branches that keep showing up more than others. I have a few 7th, 8th and 9th cousin connections via DNA by these branches. The issue is, in order for these to have persisted so long, other branches were left off during past recombination, thus making way for the ones I have to remain. So I may not be finding many Bailey ancestors but I continue to find a lot of Deal and Copeland ancestors instead.

That's the way it is for everybody. One way to work around some of this is to test your siblings but that will only get you a little farther.

The truth is, you do not have exact quanitities of each ggggg grandparent's DNA left in you. You have more of some and less of others.

Matt.

[PDHOTLEN]

What helps me most with autosomal matches are the names those people provide. I am most interested, at present, with colonial period connections. One problem with websites, such as Ancestry, is that people sometimes like to fudge and bend their lines to include famous people. That not only puts you on wrong tracks, but precludes cooperation with the perpetrators. CSI Miami-type investigations, i.e. DNA testing, is the best way to sort things out.

[djknox]

Thanks to ALL for some great feedback and insight. I acknowledge that database size is a big issue today (ie not enough participants as per Tomcat's post)... but I must believe that given time and especially if the technology's usefulness improves, a lot more people will eventually jump in. MKDexter suggests that high resolution testing is already available today... such as at Ancestry.com. However, from what I can tell, Ancestry's testing only offers 1 or 2 additional generations of high probability matching. So my question remains, does the SCIENCE EXIST to get say 95% certainty of a match with an 8th or 10th cousin, but it is just not commercially available today (due to cost, demand etc.)... or does Ancestry.com's 800,000 SNP test pretty much tap out the POTENTIAL of Autosomal Testing? Any genetic scientists out there that can answer this question? I've asked FamilyTreeDNA and the answer was a little deflecting. I think MFWARE comments are insightful about the short length of SNP segments becoming indistinguishable... but does that mean that testing more Autosomal SNPs adds little value? Maybe what is needed is more sophisticated triangularization algorithms??

I don't want to start a snowstorm of criticism here, but although I love the idea of autosomal dna cousin matching, if it can't generate high probability matches for common ancestors before the mid 1700s (even if milliions have joined the database) in a way that would allow a continuation of research from existing roadblocks, its usefulness towards extending back one's pedigree seems very limited. I would like to know from FamilyTreeDNA whether their familyfinder is just the tip of the autosomal iceberg or whether that's nearly as good as its going to get?

I'm an experienced genealogist and I know hundreds, if not thousands of family researchers that would jump into this database IF they felt they could get more out of it than just a few hundred years of generations. I think its a great tool for confirming past research... but that doesn't seem to be a big enough lure for most serious researchers... they want progression back in time on... especially on roadblocked lines.

[mkdexter]

So my question remains, does the SCIENCE EXIST to get say 95% certainty of a match with an 8th or 10th cousin, but it is just not commercially available today (due to cost, demand etc.)... or does Ancestry.com's 800,000 SNP test pretty much tap out the POTENTIAL of Autosomal Testing?

Right now people are able to find cousins up to about the 12th cousin range with any of the tests. Trying to get to 13th or 14th cousin levels has nothing to do with the test accuracy. Nor does trying to prediction a cousin is definitely an 8th and not a 7th, or a 8th and not a 9th.

I did explain one part of it, that being that the DNA itself doesn't all stay around that long.

Recombination will just about insure that you will not find too many cousins after the 10th cousin level. Some have, but usually 11th cousins, maybe 12th, but that's about it. It has nothing to do with the test resolution, whether Ancestry, 23andMe or FTDNA.

Matt.

[1_mke]

Isn't there a wildcard here though? In some families generations are much shorter than others. What this means is much older DNA may be present in some lines than others. Say family A has an average generation of twenty years over the past two hundred years. Thats ten generations to get back to the early eighteen hundreds. If family B has an average generation of thirty years over that same time period you're only talking a bit over six generations. This means "older" DNA is going to be present in family B.

I suspect the horizon is going to be pushed back further than present by a combination of inference and possibly testing of deceased relatives. Those from prolific lines will have a lot more luck if I'm right as there will be more data to work from.

[mkdexter]

The problem is not that a person no longer has DNA from their gggg-grandparents 7 generations back, but a person doesn't always SHARE segments with a different descendant of the same grandparents. That's the problem.

Recombination occurs so randomly, and is so different between two people, it is simply difficult to find SHARED segments between two people after a period of time.

So it's not the precision of the tests that will extend the tested range, it's the fact that random recombination either retains or drops the shared segments to begin with.

In other words, Cousin A might match gggg-grandparents Bob and Susie on chromosome 1, but Cousin B matches them on chromosome 2, thus Cousin A and B are not considered cousins in the FF test if that is all they have to compare to ancestors Bob and Susie with. They don't share on chromosome 1 and they don't share on chromosome 2, thus no match, yet both have DNA segments from Bob and Susie.

On the other hand, can a test detect a 3cm segment from Bob and Susie? Sure but if we don't know it is from Bob and Susie then we are stuck with the question: Where is it from? Five generations ago or fifteen? The only thing higher precision brings autosomal tests is a lot more low confidence (or distant) matches to dig through. Recombination is the problem, not the resolution of the tests.

Matt.

[Lklundin]

Recombination is the problem, not the resolution of the tests.

Right. One solution to this problem is to get tests from as many known cousins (1st, 2nd and even 3rd) as possible. It is also useful with siblings (especially sister/brother pairs since this gives both the Y and the X of their father).

That increases the number of segments which can be assumed to come from the MRCA of the testees.

I have a pair of ancestors from a part of Prussia where all records have been destroyed.
That pair had four daughters and for each of them I got an FF test from a grandchild (or two).

Although not certain it is a reasonable assumption that all sufficiently long segments shared among any of these 2nd cousin pairs come from the Prussian ancestors. With this large set of segments I hope to find FF matches that can be assumed to be related to me via my Prussian connection.

-Lars.

[T E Peterman]

By testing yourself & getting eight 3rd cousins to test, you can verify the integrity of your paper trail lines, at least back to all g-g grandparents. Some people might have genealogical issues within the last 150 years that could be resolved by this.

My approach has been to amplify the signal & enhance the resolution. This involves getting a core of closely related people tested: myself, my father & his brothers, my maternal uncle, my maternal aunt. The kinships of all of the above are, of course, confirmed, plus all matches to this set, are in fact cousin matches to me. This alone extends my paltry list of 300+ to about 700+ cousin matches. Each person is, in a sense, another listening post on the autosomal DNA signal from earlier times. This is what I mean by amplify the signal.
I enhance the resolution by getting 1st cousins of the above tested & then some of their second cousins. This breaks out the combined set of matches by "side of family". If I am discussing who a MRCA might have been with a match, it is helpful to me to be able to eliminate 3/4 or 7/8 of my family. The first & second cousins to my parents that I've tested have, of course, confirmed these fairly close genealogical relationships; plus they share matches among themselves. The entire database of "matches of interest" (ie, those found with the in common with utility) is now 1400+.

Of those that have gedcoms, a few matches have confirmed lineages extending back into the 1700s that have a weak paper trail.

Another way of looking at Family Finder is that its strengths lie at the inverse portion of the family tree to y-DNA or mtDNA testing. Although the latter are great at identifying prehistoric roots in a patriline or matriline, they might show at best a MRCA within the last 250 years or 1,000 years. But how close, really, are they? Family Finder works best at the opposite end of the spectrum; really good for close relatives, but by the time you get back more than 6 or 7 generations, the estimated kinships become harder to ascertain. You could have a y-DNA match that is 67/67 sharing all SNPs, but if he doesn't appear on FF, he is probably at least a 4th cousin or more. On the other hand, if FF finds him & you share 1,000 cms, he is likely a first cousin.

I agree that getting more people tested, essentially anyone interested in genealogy & their next of kin, is what will make this database increasingly of greater value.

Timothy Peterman

[Ann Turner]

A higher resolution test is not going to solve one intractable problem: you inherit only half of your parents' DNA. That means that some of the your ancestors' DNA is missing in you, while other ancestors gave you more than their "fair share." There's only a 54% chance that you inherited DNA from ALL 32 of your great-great-great grandparents.

http://www.genetic-inference.co.uk/blog/2009/11/how-many-ancestors-share-our-dna/

However, your siblings, aunts/uncles, and cousins will have inherited different portions of the ancestral DNA. If you don't match a cousin but your sibling does, you can still claim the common ancestor for your genealogical tree.

http://www.thegeneticgenealogist.com/2009/11/10/qa-everyone-has-two-family-trees-a-genealogical-tree-and-a-genetic-tree/

[Lklundin]

A higher resolution test is not going to solve one intractable problem: you inherit only half of your parents' DNA. That means that some of the your ancestors' DNA is missing in you, while other ancestors gave you more than their "fair share." There's only a 54% chance that you inherited DNA from ALL 32 of your great-great-great grandparents.

http://www.genetic-inference.co.uk/blog/2009/11/how-many-ancestors-share-our-dna/

The quoted simulation is interesting. Its own section with questions has one that has not been answered:
"Does your code include a minimum threshold for being identified as a relative?"

I could see no such threshold in the code (but I am also not used to a language such as R). Some numbers in the code seem to be in units of mega-bases (Mb).

I think inheriting one single SNP from an ancestor is not so interesting, so I was wondering what those 'Sharing probability' plots would look like given one of the usual thresholds, e.g. the one of 7cM and 700 SNPs from gedmatch ?

[T E Peterman]

When I talk about enhancing the resolution, I'm talking about using cousins to compare, so that you can see what part of the family a match is coming from. By testing as many close relatives as possible, focusing on those that add to the data, we can recover the DNA signal from a lot more of our ancestry than if we just test ourselves.

I have suggested to Family Tree DNA that when both the DNA suggests & the customer confirms that two or more participants are closely related, Family Finder could perhaps make better estimates of degree of kinship.

If only one in a field of participants shows someone as 5/R, I assume that he is closer to the R end of the scale; but if several siblings & cousins show someone as 5/R, I assume that he is closer to the 5 end of the scale.

Another thing that I would like to see for confirmed close relatives, is a lower centimorgan threshhold for matching if the matches are shared in common. If they do an "in common with" for the two siblings, they could report the shared matches, no matter what the cm values, as long as one of them meets Family Finder standards (total cm>20; LB>7.7). An "in common with" report for two confirmed cousins could simply lower the overall threshold, say to totalcm>10; LB>3.5; but for the cousins, I would want both matches to meet this standard.

Timothy Peterman