Doug McDonald Ancestry Analysis

As I noted last time, I was in a situation where I needed some help into ascertaining my genetic ancestry. Fortunately, there are people willing to do that sort of analysis for you. One of these is Doug McDonald. So I sent him my data and within an hour I had an analysis.

The PCA plot below shows me as a large cross in relation to different reference populations (like Europeans, Africans, East Asians etc).

Doug McDonald Ancestry Plot for me

Here’s what he said:

We also do quantitative tests. These come in three flavors, first without South Asia (represented by Pakistan) and the Mideast, second with South Asia, and finally with all three, as comparison panels.

The typical random error in the data (standard deviation) is 1%, meaning that numbers less than about 2% are not highly significant. There are also systematic errors. In particular, there is cross-coupling of values for Europe, the Mideast and S. Asia. For example, on the middle panel, a pure, northwestern European measures about 9% S. Asian, and on the third panel they typically measure 4.5% Mideastern and 8% S. Asian. Actual people from South Asia or the Mideast always test at least 15% European.

His first panel:

Europe 71.1%
East Asia 12.3%
Africa 8.2%
Oceania 4.7%
America 3.3%

When South Asia is added:

South Asia 48.7%
Europe 36.8%
Africa 5.8%
East Asia 5.0%
Oceania 2.4%
America 1.2%

And finally when Middle East is added to the list:

South Asia 46.9%
Europe 29.0%
Mideast 11.0%
East Asia 5.1%
Africa 4.3%
Oceania 2.3%
America 1.4%

And here is his analysis of these results:

This is basically a person from somewhere in region from say Iraq to Pakistan, with a substantial African contribution. The East Asian is probably not real. The African could be a few percent direct recent admixture, or it could be in input from a previously mixed population like the Makrani of Pakistan. My techniques can’t tell them apart.

The most interesting thing here for me was the African percentage.

Genome Similarity

23andme has a feature where you can find out how similar your genes are to your friends and family (who you are sharing with at the site). The result is a bar list with percentages showing similarity.

The number next to each person in the bar list is a measure of similarity. Specifically, it is the percentage of matching genotypes for all of the SNPs on our chip that are located in the genes or regions of interest.

If person 1 has AA and person 2 has AA at a particular position, they are 100% similar. If person 1 has AA and person 2 has AG, they are 50% similar at that position. If person 1 has AA and person 2 has GG, they are 0% similar at that position. We then average the percent similarity over all positions included in that comparison.

You can also calculate these similarity measures (IBS, Identical by State, distances) using plink if you have the genetic data for someone.

Discussing the expected similarity percentages, I figured that siblings and parents generally have similarity measures around the mid-80s. Usually for South Asians, it seems like their similarity percentage with other unrelated South Asians is close to 74%, especially for similar ethnic or geographic groups. Please note that this specific number 74% is valid for the specific set of SNPs included in the 23andme v2 chip. For v3 chip, people are getting higher numbers.

However, I looked far and wide and shared data with 81 people. My highest similarity percentage came out to be 73.22% with Amber, followed closely by a Bihari guy at 73.2% and a couple of other Punjabis. While most of my top matches are South Asian, with a large number of Punjabis, there is no particular pattern with several South Indians and Biharis matching highly too. My top non-South Asian matches are Iranians.

I expected my similarity percentages to be lower like they turned out to be due to my quarter non-South Asian ancestry. So that wasn’t a surprise.

I asked my parents and uncles and aunts about my great-grandmother’s ancestry. I knew she was from Egypt, but I found out that her ancestors had arrived in Egypt with Muhammad Ali Pasha. Since Muhammad Ali Pasha was an Albanian who worked for the Ottoman Sultanate, my relatives deduced that we have some Turk and/or Balkan ancestry.

So I asked a number of Turks, Southern Europeans and North Africans to share. I expected my similarity with them to be less than my similarity with South Asians, since 3/4 ancestry is more than 1/4. But I found something strange.

Let’s take any random person I am comparing my genes to who is not South Asian. If I compare how similar that person is to me against how similar he is to any South Asian, it turns out he’s more similar to the South Asian. This turns out to be true for similarity measures between me and all non-South Asians vs similarity measures between that non-South Asian and all South Asians. In brief, all East Africans, North Africans, Southern Europeans, Northern Europeans, Turks and Iranians (that are in my friends list) are more similar to all the South Asians among my friends list than they are to me.

I found this weird. If I had someone in my friends list who belonged to my great grandmother’s ethnicity or a closely related one, then I should be more similar to that friend than any random South Asian, instead of being the least similar like I was in all cases.

Clearly there was a need for better ancestry analysis in my case.

My Biogeographical Ancestry

There are several different ways to figure out your genetic ancestry. One way that 23andme shows your ancestry is by comparison with reference populations of the HGDP (Human Genome Diversity Project) dataset. I have listed how similar I am to various groups in the table below:

Reference Population Similarity Groups Included
Central & South Asians 67.14 Pathan, Makrani, Kalash, Hazara, Balochi, Sindhi, Brahui and Burusho
Northern Europeans 66.94 western Russia, France, Orkney Islands
Southern Europeans 66.93 northern Italy, Tuscany, Sardinia, French Basque
Near Easterners 66.82 Palestinian, Druze, Bedouin
Siberians 66.55 Yakut
Eastern Asians 66.48 Japan, Cambodia, China (Dai, Daur, Han, Hezhen, Lahu, Miaozu, Mongola, Naxi, Oroqen, She, Tu, Tujia, Uygu, Xibo, Yizu)
North Americans 66.47 Pima, Maya
South Americans 66.43 Surui, Karitiana, Piapoco, Curripaco
Oceanians 66.38 Papuans, Melanesians
Northern Africans 66.16 Mozabite
Eastern Africans 64.13 Kenya
Southern Africans 64.04 San, Bantu speaking South Africans
Central Africans 64.01 Biaka, Mbuti Pygmies
Western Africans 63.98 Mandenka, Yoruba

My numbers are not too different from anyone from the northwestern part of the South Asian subcontinent.

One thing to consider over here is that you are being compared to a specific set of populations. As you can see, there is no Indian references here. Similarly, Near Easterners are represented only by samples from Israel and North Africans by one Algerian population. I wonder what the case would be if they had Egyptians or Ethiopians etc in their reference.

Another way to look at your genetic ancestry is with a PCA (Principal Component Analysis) plot. With the same reference populations mentioned above, 23andme calculated the two dimensions of largest variation among that data. These two axes don’t completely describe the variation across the samples, but being the two largest components they can be used to project your genetic data in that space. At the world level, I am the green marker in the middle of the Central/South Asian cluster.

In the South Asian PCA plot, I am in the middle of the Pathan cluster and right at the top edge of the Sindhi one.

Now this doesn’t make me a Pathan. For one thing, 23andme’s reference populations do not have any Punjabis. I am sharing with a number of North Indians and Pakistanis, including several Punjabis, and they all lie around me in the plot.

There is another problem with a PCA plot though. We are looking at the two most significant dimensions, but there are other dimensions too and they combined together could account for a lot of the variation among people’s genomes. Also, let’s say we have someone who is a child of a European and an East Asian parent. Now that person, who is 50% East Asian and 50% European, would be placed about midway between the East Asian and European clusters. That’s where the Uygur and Hazara clusters are. So we can’t say that someone is Uygur just because they are placed in the Uygur cluster in a PCA plot.

There are other ways to look at your genetic ancestry and I have been exploring a bunch of them. We’ll talk about them next.

Ancestry Painting

23andme has a feature called ancestry painting which gives you the percentages of different populations you are admixed from. In my case, I got the following:

European 91.22%
Asian 8.69%
African 0.09%

Ignore the precision (to 2 decimal places). I showed that because I wanted to highlight the nonzero African ancestry percentage which I will talk about in more detail some other day.

The ancestry painting also shows you the segments on your chromosomes and which ancestral group you inherited them from. Here’s an image showing mine:

My 23andme ancestry painting

Does this mean I am 91% European and 9% Asian? Not quite! My results are about typical for someone from Punjab.

Also, the results depend on which reference populations were used as the exemplar European, Asian and African populations.

23andMe takes advantage of publicly available data for four populations studied extensively via the International HapMap project (hapmap.org). That project obtained the genotypes for 60 individuals of western European descent from Utah, 60 western African individuals from Nigeria, and 90 eastern Asian individuals, 45 from each of Japan and China. Because the two eastern Asian populations are geographically near one another and relatively similar at the genetic level, 23andMe combines these to form a single eastern Asian reference population. For more information on why these regions were used, please see (Why are these three populations used?)

So they are comparing your DNA segments to those of the three populations from the HapMap dataset. Using more reference populations would give you more fine-grained results (which is something I plan to do in my Harappa Ancestry Project).

Using the technique described by Eurogenes, you can check which chromosomal segments are classified as European (C), Asian (A) or African (Y). Here are my results for chromosome 9:

Chromosome Segment Ancestry
9 36587, 97974029 CC
9 97976425, 99363907 AC
9 99367419, 100530260 CC
9 100536329, 104679442 AC
9 104680472, 106598880 CC
9 106602625, 108990980 AC
9 108993234, 133447401 CC
9 133447580, 138437690 AC
9 138443022, 140147760 CC

The number of Asian segments on my homozygous chromosome 9 makes me doubt that it comes from my Egyptian great-grandmother. May be it’s from my great-grandfather.

The African segments are on chromosome 8:

Chromosome Segment Ancestry
8 154984, 4074371 CY
8 140917074, 142173290 CY

I need to do ancestry painting on my own in more detail.

Harappa Ancestry Project Update

I have got 25 participants to the Harappa Ancestry Project now. But we still need more especially from the Hindi belt.

I have been detailing the datasets I am using:

I have also started admixture analysis of the reference populations and first batch of project participants.

Paternal and Maternal Lines

We men inherit the Y chromosome from our fathers who got it from their fathers. So the Y chromosome can be used to trace your paternal lineage. Different sequences of alleles and mutations can be assigned to haplogroups where a haplogroup signifies common descent on the uniparental line.

According to my 23andme results, I belong to the paternal haplogroup R1a1a. This group is very common in Eastern Europe as well as South Asia. The distribution of R1a1a can be seen in the map below.

Similarly, we all inherit mitochondrial DNA from our mothers. The sequence of alleles and mutations on the mitochondrial DNA (mtDNA) is also organized into phylogenetic tree.

I can trace my maternal line to Egypt (my great-grandmother) and thus I expected a maternal haplogroup common in the eastern Mediterranean. It turns out I belong to haplogroup H, which everyone and their mother belong to in Europe as can be seen in this map.

According to Wikipedia,

Haplogroup H is the most common mtDNA haplogroup in Europe. About one half of Europeans are of mtDNA haplogroup H. The haplogroup is also common in North Africa and the Middle East. The majority of the European populations have an overall haplogroup H frequency of 40%–50%. Frequencies decrease in the southeast of the continent, reaching 20% in the Near East and Caucasus, 17% in Iran, and <10% in the Persian Gulf, Northern India and Central Asia.

Since 23andme didn’t tell me which subgroup of H I belonged to, I used mthap by James Lick:

Your rCRS differences found:

HVR2: 263G
CR: 750G 1438G 4769G 15326G
HVR1: (16519C)

Best mtDNA Haplogroup Matches:

1) H
2) H26
2) H(16192)
2) H35
2) H24
2) H10
2) H25
2) H(195)
2) H33
3) H19

Amber’s maternal haplogroup is M4a, which is mainly found in South Asia.

You can see the Y-DNA haplogroup tree and the mtDNA tree online.

Inbreeding

I have found out I am actually from West Virginia. Ok, I am just joking.

I knew that my family had a history of marriages among relatives. After all I have only 10 great-great-grandparents instead of the usual 16. With my genome in hand, I set about to quantify the inbreeding.

First, I used David Pike’s Homozygosity tool. It analyzes your genome to find significant runs where the same haplotype is inherited from both parents. Large portions of the human genome are like that. The length of these homozygous regions, however, varies depending on the relation of your parents. If your parents are closely related (first cousins in my case), then you will have longer runs. If your parents are distantly related, then over the generations those genes have had a chance to recombine and so you will have shorter runs that are homozygous.

Overall, the percentage of my autosomal (i.e. on chromosomes 1-22) SNPs that are homozygous is 71.767 and I have 41 runs of homozygosity (ROH) of length at least 200. Here are some of my longest runs:

  • Chr 1 has a ROH of length 6009 (30.95 Mb)
  • Chr 8 has a ROH of length 5819 (33.00 Mb)
  • Chr 9 has a ROH of length 5877 (57.81 Mb)
  • Chr 9 has a ROH of length 5941 (24.38 Mb)

Let’s look at my homozygosity percentage by chromosome.

Chr 1: 71.734 %
Chr 2: 69.952 %
Chr 3: 65.741 %
Chr 4: 71.563 %
Chr 5: 69.270 %
Chr 6: 76.025 %
Chr 7: 69.445 %
Chr 8: 72.690 %
Chr 9: 93.323 %
Chr 10: 69.765 %
Chr 11: 71.866 %
Chr 12: 68.443 %
Chr 13: 74.184 %
Chr 14: 68.571 %
Chr 15: 73.087 %
Chr 16: 66.541 %
Chr 17: 77.555 %
Chr 18: 67.763 %
Chr 19: 66.267 %
Chr 20: 66.228 %
Chr 21: 79.902 %
Chr 22: 69.896 %

A majority of chromosomes seem to have reasonable percentages while chromosomes 4, 6, 8, 11, 13, 15, 17 and 21 are high. However, chromosome 9 is really weird: It is 93.323% homozygous.

David Pike writes that:

So far the largest ROHs in 23andMe V2 data that I am aware of consist of:

  • 9191 consecutive tested SNPs, corresponding to a DNA segment of length 49.99 Mb.
  • 6129 consecutive tested SNPs, corresponding to a DNA segment of length 39.05 Mb.
  • 5594 consecutive tested SNPs, corresponding to a DNA segment of length 28.95 Mb.
  • 4644 consecutive tested SNPs, corresponding to a DNA segment of length 27.71 Mb.

The highest percentage for overall autosomal homozygosity that I have so far seen from 23andMe V2 data is 71.763%.

As you can see, I am an extreme case.

A number of members at DNA Forums reported their homozygous percentage. Of all those listed, mine is the 2nd highest.

According to the paper Genomic Runs of Homozygosity Record Population History and Consanguinity:

South/Central Asians and West Asians have more than three times as many ROH in all categories over 4 Mb long than sub-Saharan Africans and other Eurasians. 19% of individuals from these populations have ROH over 16 Mb in length, consistent with the high prevalence of consanguineous marriage (marriage between individuals who are second cousins or closer) in these populations.

My total ROH length (segments > 0.5Mb) is about 282Mb which is about 1.2 standard deviations above the Central/South Asian sample mean in that paper. But I am more than 1.7 standard deviations above the mean for longer segments (>5Mb).

Let’s take a look at a graph from the paper’s supplemental material which plots total ROH length versus number of homozygous segments:

My inbreeding coefficient based on the length of long (>5Mb) runs of homozygosity in my genome (fROH5) is about 0.11 while the average in the Central and South Asian sample for the HGDP dataset is 0.015 (not directly comparable due to different number of SNPs used to calculate).

Finally, I used Plink to calculate my inbreeding coefficient F using all the South Asians from my reference datasets. That coefficient comes out to be 0.1184.

Harappa Project New Site

As several people had asked, I have set up a separate website for the Harappa Ancestry Project at http://www.harappadna.org/.

I am keeping a link to the new site on the top menu bar here titled Harappa DNA.

I might also crosspost some items from the project here.

I have also set up a Facebook page for the Harappa Ancestry Project. Please like it on Facebook so I can get a nice short name for the Facebook page URL.

I have received several samples and will be reporting some analysis results soon. However, I do need lots of participants, so please spread the word.

Cross-posted at Harappa Ancestry Project.

Harappa Ancestry Project

I have become interested (some would say obsessed) with genetics recently. I wrote about getting my DNA test done and there’s a lot more about my own results that I plan to bore you with.

One fun application of genetic testing is inferring ancestry: Which ancestral group are you descended from? Can we estimate the admixture of the different population groups you are descended from?

Most DNA testing companies provide information about ancestry and genetic genealogy has taken off. With several genome databases (HapMap, HGDP, etc) and software (like plink, admixture, Structure) publicly available, the days of the genome bloggers are here. And I am trying to be the latest one.

In starting this project, I have been inspired by the Dodecad Ancestry Project by Dienekes Pontikos and Eurogenes Ancestry Project by David Wesolowski. The catalyst for this project was my friend Razib who I bug whenever I need to talk genetics.

What is Harappa Ancestry Project?
It is a project to analyze (autosomal) genetic data of participants of South Asian origin for the purpose of providing detailed ancestry information. So the focus of the project is on South Asians: Indians, Pakistanis, Bangladeshis and Sri Lankans.

The project will collect 23andme raw genetic data from participants to better understand the ancestry relationships of different South Asian ethnicities.

I have named it after Harappa, an archaeological site of the Indus Valley Civilization in Punjab, Pakistan.

Participation
People of South Asian origin, or from neighboring countries, are eligible to participate. The list of countries of origin I am accepting are as follows:

  • Afghanistan
  • Bangladesh
  • Bhutan
  • Burma
  • India
  • Iran
  • Maldives
  • Nepal
  • Pakistan
  • Sri Lanka
  • Tibet

Right now, I am only accepting raw data samples from people who have tested with 23andme.

Please do not send samples from close relatives. I define close relatives as 2nd cousins or closer. If you have data from yourself and your parents, it might be better to send the samples from your parents (assuming they are not related to each other) and not send your own sample.

If you are unsure if you are eligible to participate, please send me an email (harappa@zackvision.com) to inquire about it before sending off your raw data.

What to send?
Please send your All DNA raw data text file (zipped is better) downloaded from 23andme to harappa@zackvision.com along with ancestral background information about you and all four of your grandparents. Background information would include where they were born, mother tongue, caste/community to which they belonged, etc. Please provide as much ancestry information as possible and try to be specific. Do especially include information about any ancestry from outside South Asia.

Data Privacy
The raw genetic data and ancestry information that you send me will not be shared with anyone.

Your data will be used only for ancestry analysis. No analysis of physical or health/medical traits will be performed.

The individual ancestry analysis published on this blog will be done using an ID of the form HRPnnnn known to only you and me.

What do you get?
All results of ancestry analysis (individual and group) will be posted on this blog under the Harappa Ancestry Project category. This will include admixture analysis as well as clustering into population groups etc.

I suggest you read about Dienekes’ analysis on South Asians for an idea about what to expect.

You can access all blog posts related to this project from the Harappa Ancestry Project link on the navigation menu on every page of my website. You can also subscribe to the project feed.