Thursday, January 22, 2009

To test or not to test...DNA half-siblingship tests

Please see 11/16/09 update about siblingship DNA tests here: "Even more about DNA half-siblingship tests"


In response to several questions over the last few weeks from some distraught donor conceived adults, I'm going to discuss DNA testing, specifically half-siblingship DNA tests, and how they can help donor conceived persons find paternal relatives.

In short, a siblingship DNA test is an autosomal (across the entire genome) test of typically 15 markers [see: Whatcha gonna do with all that junk, all that junk in your...DNA?? for more information about these markers] that two potential siblings do to determine HOW LIKELY it is that they are siblings. There is also half-siblingship tests that determine how likely it is that both individuals share one common parent. The latter is quite handy for donor conceived offspring to test whether or not they have found a sibling, especially if neither individual has a donor number.

Typically, at least one mother must be tested as well for this to be accurate. Since siblings only share 50% of their DNA, half-siblings only share about 25% of their DNA, which means that often many of the markers do not match. The mother can determine which allele at a certain marker the child inherited from her, and thus by process of elimination, which allele came from the father. If one mother is tested, this is usually enough information to determine if two individuals are related or not, but for a more conclusive result both mothers can be tested.

Siblingship DNA tests are inherently very different from paternity tests. Paternity tests are cut-and-dry. Either he's the father or he's not, with 99.999% accuracy (this has some exceptions, but not for today's discussion), because for each marker one of the child's MUST match one of his. Half-siblings are different, because for a given marker, child #1 could have alleles A and B and child #2 could have alleles C and D. While this looks like they couldn't be siblings, in reality they very well could be. The father, in fact, could have carried alleles A and C, and each child inherited a different allele.

Here's another example, which shows the importance of testing mom. Suppose child #1 carried alleles A and B, and child #2 carried alleles A and D. If child #1's mom was tested and they found that she carried alleles B and C, then these two children could be siblings. If both moms were tested and found that child #2's mom carried alleles D and E, then the possibility of the children being siblings is greatly increased (because it shows that allele A MUST have come from the father).

If mom carried alleles A and C though, then the A allele from child #1 must have come from mom, and therefore the two children do not share a match at that marker. It does not mean, though, that they're not siblings, since the father could have had alleles A and B, thus accounting for both children, even though it's not apparent on the DNA test.

This leads me to my next point. Siblingship DNA tests are not 100% for sure. For each marker, matches between the two children are analyzed with high-tech mathematical equations by professionals, using the frequency of the allele in the general population. That is, for each allele at a marker, millions of individuals in the world also carry that same allele, but some alleles are much more rare than others. If it is one of those rare alleles, the points assigned to that match are significantly higher than if the allele is a common on in the population. The more points that accumulate the higher the relationship index.

A relationship index that is below 1.0 means that these two individuals are not related at all - this is a very conclusive (but not 100%) answer that the two people are not siblings. A relationship index that is between 1.0 and 50 generally means that it is more likely that the two individuals are siblings, but that the test was inconclusive (this can especially be the case if neither or only 1 mother is tested). A relationship index between 50 and 100 means that it's not conclusive but that it's somewhat supported, and a relationship index above 100 is considered conclusive. Most half-siblingship tests (where at least 1 mom was tested) end up somewhere between 50 and 70, but I know of several donor conceived adults who found siblings that tested well into the 90s.

For older female offspring who may be unable to test one or both mothers, testing services at CaBRI are unable to determine siblingship status, because they look solely at the X chromosome, and in order to determine which X chromosome was inherited by the donor (and thus shared with a female sibling), mom needs to be tested to eliminate her X chromosome. However, all is not lost. Most DNA testing companies can do a half-siblingship DNA test with only one mom's DNA, and it's possible, but not recommended that they could do a test of only the two alleged siblings. However, the likelihood of getting anything above a somewhat supported relationship index is practically impossible, thus it would only give you an idea. It would tell you that there's "no way in hell" that you could be siblings, but it could not tell you with much accuracy that you are siblings.

For males who have found a potential male sibling, CaBRI is the easiest and cheapest solution (but it takes quite a few weeks to get results).

For offspring of opposite sexes a DNA testing company must be used to determine siblingship status.


Anonymous said...

Great post. Have you been able to find the algorithm that any company uses to assign the points based on the frequency of a specific allele in the population? I have tried various formulations but I've been unsatisfied with the results.

On that same note, do you know of anyone who is using, has tried a clustering approach (i.e. showing that a group of DC people being tested match at more sites to each other than to random groupings of the same size). I'm not sure it would work and the idea is half-baked.

Thanks again!

Lindsay said...

I have not been able to find the algorithm - I assume that because of the complexity of the tests the software to perform the algorithms must be expensive. I spent some time last spring attempting to piece together the frequency points, but much of the numbers I found were ethnically specific, which wouldn't work, mainly from anthropological data. I also tried to decipher what the calculations were based on DNA tests I had already taken, as well as results from several others, but that turned up not to be fruitful either, because all of the results I was looking at were negative paternity tests (so not enough of the markers were matches).

I've never thought of a cluster approach, but I suppose if there were a handful of DC people who thought they might all be siblings, it would be possible. I'm not sure if I understand your logic on this though.

mia said...

I added you to my links Lindsay. I'm sorry it took me so long. I have a LOT of work to do with those links and I am an impossibly bad procrastinator.

I wish you the best on your journey of self discovery.

Anonymous said...

Yeah, it sounds like you went down some of the paths I tried, too. I also tried back calculating from some sibling results people sent me and all I really learned was that I didn't understand all of the parameters. I don't think the calculations are likely to be anything that Excel can't handle; it's just that there are a lot of parameters (which ethnic tables are used and how they are weighted AND WHY) that we aren't told.

The clustering idea is definitely not fully thought through. The logic (such as it is) is something along the line of taking a large dataset of DC people's alleles then grouping them by relatedness. That gives you groupings that might be sibs and you can tweak the data to fit test groups where you know the relationships independently. The main problem is that I can imagine the groupings might be too crude and end up being things like ethnic ancestry rather than family. Frankly, I don't understand statistics deeply enough to even know if I'm being silly. My mathematical kung fu isn't up to the task.

I also have thought about an approach like the one you describe (basing the calculations on group probabilities rather than pairwise probabilities). I think it might be less statistically "noisy"... but, again the math brain is weak.

Lindsay said...

I agree....I love the genetics, but I tend to get lost in the statistics --- math was NEVER my strong point!!

If I wanted to end up back in the ER with a migraine again, I could try to tackle those numbers again, but I spent weeks on it last spring and felt like I was running around in circles.

I see what you're trying to say with the cluster approach, but I still don't know how accurate it would be, since there would also have to be a way to eliminate mom's alleles. At best, ethnicity might be determined, but again, without eliminating mom's alleles, it wouldn't be much help. What you're really describing is what CaBRI is trying to do - except they are able to find siblings based on X and Y chromosomes.

Honestly, what I would LOVE to see is something like the UKDonorLink, except on a worldwide scale...considering sperm banks now ship sperm around the world, so it's likely (especially from large banks like CCB, Cryos, Fairfax, Xytex, etc) that siblings could be found on several continents! Especially with parents in countries like Australia, Netherlands, and Britain who are trying to avoid telling their children about their conception.

yagmurunsesi said...
This comment has been removed by a blog administrator.
Anonymous said...

I wonder how a lab could go to court with this? Most labs have said they would not but there is one lab that said it can be used in court. If the test said it was possible the girl was related by this test could this be fought in court?

Lindsay said...

See update about the combined sibship index here:

We've discovered the math to determine relatedness.

Anonymous said...

My father, 87, and a woman in GA think they might have the same father. Of course the father is deceased so we did a home half siblingship test on the 2 of them. It came back >1%. Does that mean they are definitly NOT related? We were so disappointed as we thought we had found the wereabouts of my grandfather.