STR genetic difference vs known SNP mutation

[JeremyGlen2017]

Hello,

I am hoping to get some insight as to how one can have a match at a GD of say 7 at Y111(99% confidence MCRA @ 22 generations), but not be positive for a mutation that occurred around 22-24 generations back?

My closest match, at GD of 4, has a GD 3 match that does not share the mutation we share, and is a surname match as well. The mutation, R-S781, is confirmed to have occurred in a man born in the mid 13th century, which is well within the 99% confidence level of 14 generations at that GD of 3.

Can anyone shed light on how this occurs?

[TwiddlingThumbs]

I suspect it is because that, per my recent post on a different thread, TMRCA probabilities are highly inaccurate and seem to frequently grossly overstate probabilities of relatedness.

It may be that the estimated time of SNP mutation is way off as well. I'm not sure what the basis for your statement is that the SNP "is confirmed to have occurred in a man born in the mid 13th century" Taking a look at YFull, they give an estimate for how long ago a SNP arose, but it is just an estimate. For R-S781, YFull says there is a 95% chance that it arose between 500-750 years ago (which you can see if you put your cursor over the "formed 600 years ago part). https://www.yfull.com/tree/R-S781/ It may well be the case, however, this these SNP dating calculations are just as crude as those for STR TMRCA. Lots of math, but not a lot of accuracy and with unreliable confidence probabilities.

[JeremyGlen2017]

The R-S781 mutation has been confirmed to have happened in John Stewart de Bonkyll, born in 1245 in Scotland. Since the Royal Stewart lines have been so heavily tested, and have great paper trails from current living descendants back through all the High Stewards, this is a known mutation date.

What I find interesting is the relations in regards to STR values of those I, and others, match to that are very close matches, but not within the known time frame of this particular mutation.

I did discuss this briefly with Ian McDonald, and he said he is working towards being able to use the information from the Stewart lines to further refine his SNP aging process, which definitely lends credibility to your statement about the probabilities being inaccurate.

[TwiddlingThumbs]

The R-S781 mutation has been confirmed to have happened in John Stewart de Bonkyll, born in 1245 in Scotland. Since the Royal Stewart lines have been so heavily tested, and have great paper trails from current living descendants back through all the High Stewards, this is a known mutation date.

Triangulation. Got it. So mutation actually occurred about 775 years ago, when YFull estimated that it occurred 600 years ago, with a 95% chance that it occurred 500-750 years ago. So the estimate was not too terrible -- just outside the range, but does make the confidence interval/probability look a bit inaccurate. You'd like to see it at least fall within the range. Kind of reduces one's confidence in the confidence interval ;-) Probably still a lot better than the STR TMRCA estimates.

[benowicz]

Not all the people giving the DNA (i.e., people alive today) were born on the same day. Probably some people giving the DNA were already old men before others were even born. So the thing is probably not a problem with the formula, but the range of birth days for the people giving the DNA. 25 years is not so long a time.

[TwiddlingThumbs]

Not all the people giving the DNA (i.e., people alive today) were born on the same day. Probably some people giving the DNA were already old men before others were even born. So the thing is probably not a problem with the formula, but the range of birth days for the people giving the DNA. 25 years is not so long a time.

Good point about different ages of test takers. But their confidence interval should take all stuff like that account. If it doesn't, that's a problem with their algorithm.

[benowicz]

Good point about different ages of test takers. But their confidence interval should take all stuff like that account. If it doesn't, that's a problem with their algorithm.

No, that's not what the confidence interval does. It measures the variability of SNP mutations, not the starting point from which you begin counting back.

The problem is that 750 years ago is not from today. It is from the date of conception of the donors, who obviously were not all conceived on the same day. Most likely the person who started this thread forgot that the typical donor for these tests is probably in his 40s, placing the mutation in a specific range of years consistent with this Stuart of Bonkyll person.

[JeremyGlen2017]

I still haven't figured out how someone can be at a 99% confidence of a shared ancestor in 14 generations, when they are proven by SNP testing to not have a connection in at least 20 generations.

My closest match and I are waiting for one of his "closer" matches, GD of 2 @Y111, testing results to see if he has the same mutation. Everything is pointing to him not having it, according to their paper genealogy trail, and points to the connection being an additional generation back past that, with the man being born in 1225.

We are possibly looking at a GD 2 match being related in a minimum of 21 generations(assuming a starting point of 1950 and 33yrs per generation average).

I'm starting to lean towards STRs not really being very accurate at all.

[benowicz]

I'm used to the idea that accuracy is increased with sample size. It's so elementary. But I have never seen anyone even try to quantify the precise level of accuracy in the context of matching STR profiles. That is surprising. How much more accurate is a 111 marker match vs. a 67 marker match vs. a 37 marker match at a similar predicted TMRCA? How is it that nobody has even attempted such a calculation yet?

The same theory goes for SNPs. In theory, an incredibly rare back mutation of a single SNP could produce anomalous results, so it should be helpful to provide context for the total number of SNPs compared--and whether or not both donors have tested to the same level.

At least you're lucky that you belong to a common SNP status with a lot of research behind it. People in the far reaches of DF27 don't even have enough SNPs identified in the literature to even contemplate such a comparison.

[JeremyGlen2017]

If you really want to throw a wrench into the mix, I have no GD 0 matches at Y67, where my closest match at Y111, GD of 4, has 16 GD 0 matches at Y67. I have a variation in the STR profile at DYS448, with a value of 20, and all my matches have a 19. Since I have the variation so early on in the panel, it appears that I should not be as close of a relation as I am, since we share a common ancestor born in 1797.

Yes, I was definitely blessed with a very heavily tested branch of the tree. My results just came in from FGC, and my closest match and I are hoping to add a new twig.

With the possibility of FTDNA going to a Y500 STR test, I'm not sure if that will muddy the waters further, or help clear things up more.

[benowicz]

The problem with STRs is the possibility that a back mutation could obscure from view an original mutation. If, say DYS458, went from 19 to 20 and back again to 19, you would have no way of knowing, since you have only one donor as a reference point, not time series data, like a sample from each ancestor in the direct line of descent.

One crude way of estimating the relative accuracy of each marker array--not absolute, but in comparison to each other--would be to sum the mutation rates for all markers in the array. I did that using the FTDNA numbers reported in the McGee Utility--no idea whether they're considered up to date estimates--and this is what I found:

37 markers is about 80% more accurate than 25 markers
67 markers is about 47% more accurate than 37 markers, or about 165% more accurate than 25 markers

I haven't tested for 111 markers, so I didn't bother to calculate that out. But I think all the new markers have very low rates, around 0.2%. So I guess that

111 markers is about 243% more accurate than 25 markers, 91% more accurate than 37 markers, 30% more accurate than 67 markers.

Assuming the other markers in the proposed 500 marker array are also about 0.2%, I'd guess

500 markers would be about 950% more accurate than 25 markers, 480% more accurate than 37 markers, 295% more accurate than 67 markers, and 200% more accurate than 111 markers.

With each new STR marker tested, you become less susceptible to distortion through a weird back mutation on any one of them.

It's probably also a good deal for people in poorly studied subclades like DF27. An extremely high resolution STR test may not necessarily be more accurate than an SNP match, but for a lot of people it could be a good proxy. Or, indeed, the only option. There are a vast number of people, like me, who live in under-studied subclades. SNP testing doesn't do much for us because they haven't even identified the relevant SNPs.

[benowicz]

Assuming that people are willing to pay for a high resolution test, that is. The problem is there are all these people running around with useless 12 and 25 marker STR profiles.