Brad Ideas - The odds of knowing your cousins: 23andme Part 1 - Comments

It's complicated

brad — Mon, 29 Mar 2010 18:22:50 -0700

Presumably more will be learned the more samples they get with real connections behind them. Though in theory a lot of that should have been learned from the icelandic studies. And full sequencing will tell more.

The issue with SNPs is that in order to spot two strands as identical by descent, they have to amass a large string of SNPs that could be the same. For each location, you have two values, and we don’t know (unless they are very close I presume) which strand any given value is from.

So if you have say AG and the other person has CT, you can say for sure this section is not identical by descent. If you have AG and the other person has AT, you have something which might be identical, but might not be. If you both have the same pair it doesn’t tell you much more, you only learn from the negatives.

If you get a long string with no negatives, it starts to be such that it is not likely to be from chance. Two unrelated people should not “not mismatch” — yes two “nots” here for a long string.

But who knows, maybe there are patterns that stay together for various reasons and make fairly unrelated people keep them intact.

Ancient Connections

Anonymous — Mon, 29 Mar 2010 17:38:35 -0700

This is exactly what I was wondering about the main blog, the cousins did not find out or figure out a real common ancestor...thus the possibility of error reporting, or the possibility of really distant cousin calculations (say neolithic, could this explain jewish and non jewish connections in some cases?) For example, there are hispanics looking or claiming a crypto jewish history and are resulting in 5th cousins or greater to database verified jews, could this be a neolithic ancient connection?

This is not surprising

brad — Mon, 08 Mar 2010 22:41:19 -0800

You and your brothers are all different mixes of your parents (and thus your grandparents.) You are all related to this cousin through either mom or dad, or more to the point through one of your grandparents. All 3 of you got different mixes of your 4 grandparents and you obviously got more of the segments in question from that grandparent than your brothers did.

So these numbers make sense, but this is also why they both say “predicted 3rd cousin” and also “Could be 3rd to 4th” and also say that the odds of error can be high. However, with a lot of DNA matching (like 80cm) the theory is you have to be close, and could be even closer. Your brother B is of course just as close but he shares less, so he might get a prediction of more distant cousin but “could be closer.”

But yes, they don’t account well for the idea of being a 4th cousin through 4 different paths, which would create the same DNA sharing as a 3rd cousin would have. If it turns out, for example, than JZ’s parents are both my 3rd cousins, then he is my 3rd cousin once removed, but has as much DNA shared as a 3rd cousin.

I found these earlier

brad — Mon, 08 Mar 2010 15:50:00 -0800

But did not have a great deal of luck. Their surnames got anglicized on arrival in England. My g-grandmother was Rebecca Pozin but may have uses Sarah or Sura as a Hebrew name, and the spellings probably changed. Same for the g-grandfather.

Better experiments are likely

brad — Mon, 08 Mar 2010 15:41:00 -0800

And may disprove these algorithms. However, the stretches are not a couple of shared SNPs.

JZ and I share 5 segments of DNA declared “Identical by descent” by 23andMe’s algorithms. It is a total of 54 “centimorgans” covering 36 million base pairs and 6275 SNPs.

However, within a few years, with cheap full-genome sequencing available, there will be no doubt due to the limitations of the genechips.

Here is their FAQ on it:

Relative Finder looks for segments of DNA in pairs of individuals that have come from a common ancestor (i.e. the two segments are Identical By Descent, or IBD). It does this by taking advantage of the fact that if two people have “opposite homozygotes” at given SNP, where say one person is an AA at this SNP and the other person is a GG, then we know that at this SNP location it is not possible for these two people to have inherited either letter (allele) from a common ancestor. Stretches of DNA that contain hundreds of SNPs that lack opposite homozygotes, then, are evidence for IBD. We report only matches that have a longest segment of at least 7 cM (centiMorgans, a measure of genetic distance) and at least 700 SNPs. Additional segments need to be at least 5 cM and have at least 700 SNPs.

Vitebsk Jewish records

Brooke Schreier Ganz — Mon, 08 Mar 2010 15:27:56 -0800

You know, a LOT of 19th and early 20th Century Jewish vital records did survive for Vitebsk: census records, birth records, marriage records, death records, school records, police/KGB records, voter records, etc.

Here is the latest list of every Jewish record that is known to have survived in the archives for Vitebsk city, Vitebsk Guberniya (an administrative subdivision), and Vitebsk Uezd (also an administrative subdivision), respectively:

http://www.rtrfoundation.org/results.php?townName=VITEBSK
http://www.rtrfoundation.org/results.php?townName=VITEBSK%20GUBERNIYA
http://www.rtrfoundation.org/results.php?townName=VITEBSK%20UEZD

A small amount of this information is already online at JewishGen -- try their "All-Belarus Database" here, which is totally free and has 150,000+ records online:
http://www.jewishgen.org/databases/belarus/

And the "Belarus-SIG" (Special Interest Group) has even more information:
http://www.jewishgen.org/belarus/

But a lot of Vitebsk's vital records information is still in the stacks of the various Belarus archives. So if you (or your newfound cousins) would like to do some genealogical digging, or hire a genealogist to do the digging for you, you all might happily surprised at just how much family information you would be able to find.

technical question

Mike — Mon, 08 Mar 2010 15:08:47 -0800

All the posts seem to share a strong faith in 23nme's algorithm. Why so? 600,000 common SNPs is pretty crap coverage, and I certainly would not go on faith that anyone with a couple of shared SNPs was any kind of cousin. These old-timey arrays were not designed for this purpose and don't genotype the most informative loci for this sort of thing. Aren't you reading a bit too much into the data? Has anyone done a controlled experiment?

right

Brooke Schreier Ganz — Mon, 08 Mar 2010 15:04:50 -0800

"somehow medium distance relatives found one another even though the two forks of the family took entirely different journeys through different countries at different times"

My husband and I met in college. He happened to see me in a comedy group show, nudged his friend who was sitting with him, and whispered to him if he knew me, could he introduce us? The friend did know me, I agreed to be set up on a (semi-)blind date, and we went out not too long afterwards. So I guess he saw something he liked, but whether it was thanks to that teensy bit of shared DNA that made me look appealing to him is unknown...

But before it sounds like those tiny bits of DNA were calling out to another from across the cosmos, let us not forget that Penn is a 30% Jewish school, and so the odds of meeting a nice Jewish guy there were pretty high, and the odds of that Jewish guy being distantly related to me are, as we have seen, more common than most people would believe.

Oddly enough, though, there is a recent study that says that couples who are third cousins are slightly more prolific (i.e. have more children together) than both more and less related couples: http://www.eurekalert.org/pub_releases/2008-02/dg-dlc020408.php

And keep in mind that the 23andMe Relative Finder algorithms were calibrated using Utah Mormons, most of whom have Northern European pedigrees, as its standard population set. Obviously, there is much less inbreeding and many fewer historical bottlenecks there than in the Jewish community, or other semi-closed groups like the Amish, Mennonites, Icelanders, Druze, etc. And speaking of bottlenecks, the post-Black Death anti-Jewish riots (in 14th Century Germany) and the the Bogdan Khmelnitsky pogroms (in 17th Century Ukraine) both did a real doozy on the number of surviving European Jewish lineages in those areas, along with many less known and less widepread events. The handful of survivors of such events may have been relatives, close or distant, who would have been forced to marry within the small remaining gene pool, which would probably increase the homozygosity of their offspring. Again, hello inbreeding!

Yes, I know it's happening

brad — Mon, 08 Mar 2010 12:49:04 -0800

But right now mostly to people who are getting involve in genetic genealogy. In a few years it’s going to happen to everybody, even those who can’t spell DNA.

Your story would not be so odd

brad — Mon, 08 Mar 2010 12:48:11 -0800

If you had married in the old country, where in small communities marrying a 4th cousin was a pretty regular event. It’s not too uncommon even today among people who don’t leave the nest and go to other states and countries.

What makes these connections surprising is they are post-diaspora, that somehow medium distance relatives found one another even though the two forks of the family took entirely different journeys through different countries at different times.

Now it does seem clear that a lot of Jews came to Canada and the USA since Europe started showing a growing amount of hostility, culminating in the horror of WWII.

These things seem so unlikely that there is reason to suspect that the cousin algorithm is either too generous, or is fooled by inbreeding. I do share as much DNA with JZ as 3rd cousins should share, but it may be it’s because we are, through different paths, a set of 4th cousins, 5th cousins and 6th cousins all at once, rather than one direct 4 generation family.

I don't think the Ashkenazi

Brooke Schreier Ganz — Mon, 08 Mar 2010 10:23:39 -0800

I don't think the Ashkenazi Jewish over-representation among customers at 23andMe is due to the medical tests 23andMe offers. Ashkenazi Jews are also very over-represented in FamilyTreeDNA.com's database, and FTDNA doesn't offer disease testing like 23andMe. And testing for the "Askenazic diseases" panel (Tay Sachs, Canavan, Gaucher's, Maple Syrup Urine Disease, etc.) is offered by Quest Diagnostics and other companies as a blood test that many people choose to take at their doctor's offices before having children. So I think the motivation behind the numbers at 23andMe is two-fold:

1) Many Ashkenazim only know their family histories for the last hundred years or so, and just in the US or Canada or wherever, and do not know their family histories within "the Old Country". They believe -- usually incorrectly! -- that all the old vital records in Poland or Hungary or wherever were destroyed during the Holocaust or under communism, or are inaccessible to laymen. So they start off their genealogical sleuthing with a DNA test, hoping to get enough clues to find their distant family.

From a genealogical perspective, this is frustratingly backwards; they should be using DNA as an adjunct or as a confirmation to standard research techniques, which can be much more fruitful than they realize. How many of them, for example, know about the online JRI-Poland vital records database, which is totally free? (http://www.jri-poland.org). How many of them know about JewishGen, the massive Jewish genealogy portal with all its mailing lists and databases and such? (http://www.jewishgen.org) How many know that all the Ellis Island immigration records are online, which can usually give them the name of their ancestor's specific hometown? (http://stevemorse.org/ellis2/ellisgold.html) From discussions I've had with matches at 23andMe, almost none of them know about these key genealogical research tools, and wanted to just jump in to genetic testing without putting in the regular legwork. Thankfully, the testees at FTDNA are usually far better informed.

2) These tests aren't cheap, and they are marketed almost exclusively in the English-speaking Western world. Most US Jews have a higher-than-average income level and live in the English-speaking Western world and therefore can make use of them in higher numbers than other ethnic groups.

And note that Sephardic, Romaniote, and Mizrahi Jews are all under-represented in the major genetic genealogy databases. The number of Persian Jews in the FTDNA database is almost nil, despite the number of Persian Jews living in the US.

a little late?

Brooke Schreier Ganz — Mon, 08 Mar 2010 09:47:11 -0800

You wrote: Next week, I will write about some of those privacy questions, and the coming (in just a few years) exposure of almost all deep family secrets (adoptions, sperm donations, and children-by-infidelity).

A few years?! Uh, I think you're a little late to the party. Check out the copious discussions of the delightfully inclusive term "NPE" -- "Non-Paternal Event" -- on the handful of genetic genealogy mailing lists and fora (such as http://archiver.rootsweb.ancestry.com/th/index/GENEALOGY-DNA , for example). Many project administrators at FamilyTreeDNA.com, which has a much more genealogically-knowledgeable userbase, have already had to deal with breaking the news to the occasional project member that their long-studied family lineage doesn't match up with their genetic realities. There have already been situations where brothers or first cousins have tested and stumbled across Mom's or Grandma's little secret the hard way.

On the other hand, both FamilyTreeDNA and 23andMe have been excellent tools for adoptees searching for their biological families. That's the flip-side.

genetic relationship vs genealogical relationship

Peter — Mon, 08 Mar 2010 07:41:49 -0800

My new "cousin" has been away, so no chance yet to compare family trees.

Here is something of interest though - he is "more related" to me than to my two brothers:

to me: HIR's are 80cM over 9 segments
to my brother "A": 58cM over 6 segments
to my other brother "B": 35cM over 5 segments

(HIR= half-identical region; cM= centiMorgans)

My brother "A" is somewhat closer to me genetically than my brother "B."

I'm not sure how the 23andme algorithm can distinguish between 3rd cousins and 4th cousins multiple times over.

A really interesting computer science project would be to combine some partial family trees with this genetic information. I suspect it would be possible to fill out the gaps in the trees. For example, once my first cousin (on my paternal side) gets her results from 23andme, I should be able to determine whether my new-found putative 3rd cousin is related on my paternal side or maternal side.

Actually, if some distant ancestor had enough descendants on the system, it might be possible to recreate big chunks of that ancestor's genes. To the extent multiple HIR's overlap, you could combine them into longer sequences, just as was done when the genome was first sequenced.

Peter

missing the point..

Ralph — Mon, 08 Mar 2010 05:15:10 -0800

the point of the birthday example is not that a specific person is looking to match their birthday, but whether you can find such a match amongst any two people in the room. Same idea here (where the room has 35,000 people). As with the birthday example, the person with the match feels as if they beat the odds, but in fact the odds were good that *someone* in the room would have that event.

overrepresention factor vs inbreeding factor

Ralph — Mon, 08 Mar 2010 03:33:31 -0800

the ashkenazis are indeed overrepresented, but that doesn't seem to be as big of a factor as the pedigree collapse. the realization is that the ratio (number of "genetic 3rd cousins")/(number of actual 3rd cousins) is much higher than expectations. It's still very rare to find someone (even in a self-selected group) with a common great-great-grandparent; but for an inbred population, it's not that uncommon to find someone you know in that group with enough common DNA segments as if you had a common great-great-grandparent.

The odds of knowing your cousins: 23andme Part 1

brad — Wed, 03 Mar 2010 15:20:18 -0800

Bizarrely, Jonathan Zittrain turns out to be my cousin — which is odd because I have known him for some time and he is also very active in the online civil rights world. How we came to learn this will be the first of my postings on the future of DNA sequencing and the company 23andMe.

(Follow the genetics for part two and other articles.)

23andMe is one of a small crop of personal genomics companies. For a cash fee (ranging from $400 to $1000, but dropping with regularity) you get a kit to send in a DNA sample. They can’t sequence your genome for that amount today, but they can read around 600,000 “single-nucleotide polymorphisms” (SNPs) which are single-letter locations in the genome that are known to vary among different people, and the subject of various research about disease. 23andMe began hoping to let their customers know about how their own DNA predicted their risk for a variety of different diseases and traits. The result is a collection of information — some of which will just make you worry (or breathe more easily) and some of which is actually useful. However, the company’s second-order goal is the real money-maker. They hope to get the sequenced people to fill out surveys and participate in studies. For example, the more people fill out their weight in surveys, the more likely they might notice, “Hey, all the fat people have this SNP, and the thin people have that SNP, maybe we’ve found something.”

However, recently they added a new feature called “Relative Finder.” With Relative Finder, they will compare your DNA with all the other customers, and see if they can find long identical stretches which are very likely to have come from a common ancestor. The more of this they find, the more closely related two people are. All of us are related, often closer than we think, but this technique, in theory, can identify closer relatives like 1st through 4th cousins. (It gets a bit noisy after this.)

Relative Finder shows you a display listing all the people you are related to in their database, and for some people, it turns out to be a lot. You don’t see the name of the person but you can send them an E-mail, and if they agree and respond, you can talk, or even compare your genomes to see where you have matching DNA.

For me it showed one third cousin, and about a dozen 4th cousins. Many people don’t get many relatives that close. A third cousin, if you were wondering, is somebody who shares a great-great-grandparent with you, or more typically a pair of them. It means that your grandparents and their grandparents were “1st” cousins (ordinary cousins.) Most people don’t have much contact with 3rd cousins or care much to. It’s not a very close relationship.

However, I was greatly shocked to see the response that this mystery cousin was Jonathan Zittrain. Jonathan and I are not close friends, more appropriately we might be called friendly colleagues in the cyberlaw field, he being a founder of the Berkman Center and I being at the EFF. But we had seen one another a few times in the prior month, and both lectured recently at the new Singularity University, so we are not distant acquaintances either. Still, it was rather shocking to see this result. I was curious to try to figure out what the odds of it are.