Tag Archives: hapmap

Introducing Reference 3

Having collected 12 datasets, I have gone through them and finally selected the samples and SNPs I want to include in my new dataset, which I'll call Reference 3.

It has 3,889 individuals and 217,957 SNPs. Since this is a South Asia focused blog, there are a total of 558 South Asians in this reference set (compared to 398 in my Reference I).

You can see the number of SNPs of various datasets which are common to 23andme version 2, 23andme version 3 and FTDNA Family Finder (Illumina chip).

The following datasets had more than 280,000 SNPs common with all three platforms and hence were included in Reference 3:

  1. HapMap
  2. HGDP
  3. SGVP
  4. Behar
  5. Henn (Khoisan data)
  6. Rasmussen
  7. Austroasiatic
  8. Latino
  9. 1000genomes

Reich et al had about 100,000 SNPs in common with 23andme (v2 & v3 intersection) and 137,000 with FTDNA, but there was not a great overlap. Only 59,000 Reich et al SNPs were present in all three platforms. Since I really wanted Reich et al data in Reference 3, I included it but the SNPs used for FTDNA comparisons won't be the same as for the 23andme comparisons.

Of the datasets I could not include, I am most disappointed about the Pan-Asian dataset since it has a good coverage of South and Southeast Asia. Unfortunately, it has only 19,000 SNPs in common with 23andme v2 and 23,000 with 23andme v3. I am going to have to do some analyses with the Pan-Asian data but it just can't be included in my Reference 3.

I am also interested in doing some analysis with the Henn et al African data with about 52,000 SNPs for personal reasons.

Xing et al has about 71,000 SNPs in common with 23andme v3, so some good work could be done with that, though I'll have to use only 23andme version 3 participants.

The information about the populations included in Reference 3 is in a spreadsheet as usual.

Supervised Continental Admixture

Since the version 1.1 of Admixture with supervised option came almost two months ago, I have been salivating over it.

My original use case for it is not possible (for now). I wanted to be able to assign a few of the K ancestral components to specific reference populations and let the other ancestral components fall where they may. But we can do supervised admixture only by assigning all K ancestral components.

So I decided to test this supervised option by mimicking the three continental percentages 23andme assigns you on their ancestry painting page. Mine are:

Europe 91.22%
Asia 8.69%
Africa 0.09%

You can get the extra precision (and false sense of accuracy) here.

Regarding the reference populations used for ancestry painting, 23andme says:

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.

So I dug up my reference admixture run at K=3 and found the same number of samples of these HapMap populations by looking for those samples which had the highest percentage in the respective component.

Then I combined these 210 samples from the HapMap with 74 Harappa Project participants (HRP0001 to HRP0079, excluding 5 who are related to others).

The results of the supervised admixture run are in a spreadsheet and also shown in a bar chart below.

Since I did run an unsupervised K=3 admixture analysis of the first Harappa batch with the whole reference I populations, you can compare these results to those.

HapMap Redo

As part of my effort to create one big reference dataset for my use, I have been going over all the datasets I have and make sure there's no duplicates or relatives or any other strange things that could cause issues with my analysis.

So I went back to HapMap, which you can download from their website. I am using HapMap 3 public release #3 from May 28, 2010.

I found one set of duplicates, NA21344 is identical to NA21737. And a whole bunch of pairs with high identity-by-descent values, which I calculated using Plink. You can see the samples with PI_HAT greater than 0.5 in this spreadsheet. PI_HAT is the proportion IBD estimated by plink. Notice also that all these pairs also have high IBS similarity (the DSC column), more than 85% similar in fact.

All the 41 samples I have removed as a result of this are listed in this spreadsheet.

One PED File to Rule Them All

I am interested in North African populations due to my own heritage, so when Razib alerted me that Henn et al had a paper out about South African origins of humans and their African dataset was publicly available and included populations from all over Africa, I immediately downloaded it.

I have also been considering looking into the East Asian admixture in South Asians and Iranians in some detail to see where it originates from: Southeast Asia, Chinese/Japanese/Koreans, or the Turkic/Mongolian/Siberian populations of interior northeastern Asia. At a quick glance, Razib is correct:

The eastern Asian components are enriched among Bengalis, as you’d expect, but they’re found in different proportions among many individuals who hail from the northern fringe of South Asia more generally. It seems clear that the further west you go, the more likely the “eastern” element is going to be Turk, while the further east (and to some extent south) the more likely it is to be more southernly in provenance.

To do a better job though, it would be better to have more than the Yakut as an examplar of the Siberian component as I have done till now. Therefore, I downloaded the arctic populations dataset from Rasmussen et al.

Combining Henn et al and Rasmussen et al with my previous datasets (HapMap, HGDP, SGVP, Behar et al and Xing et al), I got 3,970 samples with a total of 1,716,031 SNPs represented, though at 99% genotyping rate it gets reduced to about 27,000 SNPs.

I did not remove any populations or individuals except for any duplicates and non-founders.

Here's the information on the populations represented in this dataset.

Now I am on the lookout for more datasets that are public, have enough SNPs in common with this set and can easily be converted into the Plink PED format. So if you know of any, let me know. May be I will have the biggest and most diverse dataset with your help.

HapMap Gujaratis

Razib is wondering what's going on with the HapMap Houston Gujaratis.

As you can see, the Chinese simply do not vary much, and are a tight cluster. But, there is a somewhat equivalent Gujarati cluster too! The HapMap sample was collected from Gujaratis in Houston. To me, it looks like that Houston population can be divided into two groups: one of the tight cluster, and the rest of the population, which is all over the place. [...] What’s more interesting is to try and understanding what’s going on with Houston Gujaratis. Anyone in the audience know?

And his 3-dimensional PCA plot: (Those on the right are Gujaratis)
PCA Plot of Gujaratis and Chinese

So I thought I would share the admixture results for the Gujaratis for K=8. Here's the spreadsheet of the admixture proportions for Gujaratis. And here is the plot:

Gujaratis Admixture K=8

The ancestral components and their statistics are as follows:

Population Range Mean Median
C1 South Asian 64-89% 81.9% 85.8%
C2 West Asian 0-13% 2.3% 1.6%
C3 European 2-22% 7.6% 5.0%
C4 Southeast Asian 0-9% 4.9% 5.0%
C5 Austronesian 1-6% 2.8% 2.9%
C6 Northeast Asian 0-3% 0.4% 0.0%
C7 West African 0-1% 0.0% 0.0%
C8 East African 0-0% 0.0% 0.0%

It looks like a majority of the Gujarati samples have mostly South Asian ancestral component with small amounts of West Asian, European and Southeast Asian, but some Gujarati samples have much larger West Asian and/or European ancestral components.

Reference Dataset II

Combining my reference population with Xing et al data gets me 3,222 3,161 samples but with only about 23,000 SNPs after LD-pruning.

The good thing is that this dataset has 544 South Asian samples from 24 ethnic groups. So it'll be useful for some analyses despite the low number of SNPs. I'll try to run parallel analyses on my reference population and this dataset so we can compare the pros and cons of both.

UPDATE: I removed 61 pygmy and San samples.

Admixture: Reference Population

For regular admixture analysis, I am using HapMap, HGDP, SGVP and Behar datasets with some samples removed as I wrote earlier.

For each of these datasets,

  1. I first filtered to keep only the list of SNPs present in 23andme v2 chip.
    plink --bfile data --extract 23andmev2.snplist
  2. I also filtered for founders:
    plink --bfile data --filter-founders
  3. And excluded SNPs with missing rates greater than 1%:
    plink --bfile data --geno 0.01

Then, I merged the datasets one by one. The reason for doing it one by one was that there were conflicts of strand orientation (forward or reverse) between the different datasets. If the merge operation gave an error, I had to flip those strands in one dataset and try the merge again.

plink --bfile data1 --bmerge data2.bed data2.bim data2.fam --make-bed
plink --bfile data2 --flip plink.missnp --make-bed --out data2flip
plink --bfile data1 --bmerge data2flip.bed data2flip.bim data2flip.fam --make-bed

Once all the four datasets were merged, I processed the combined data file:

  1. Removed SNPs with a missing rate of more than 1% in the combined dataset
    plink --bfile data --geno 0.01
  2. Then i performed linkage disequilibrium based pruning using a window size of 50, a step of 5 and r^2 threshold of 0.3:
    plink --bfile data --indep-pairwise 50 5 0.3
    plink --bfile data --extract plink.prune.in --make-bed

This gave me a reference population of 2,693 2,654 individuals with each sample having about 186,000 SNPs. Out of these 2,693 2,654 individuals, we have a total of 398 South Asians belonging to 16 ethnic groups.

Finally, it's time to start having some fun!

UPDATE: I removed 39 Pygmy and San samples because they were causing some trouble with African ancestral components. Since we are not interested in detailed African ancestry and African admixture among South Asians is not likely to be pygmy or San, I decided it would be best to remove them.


I am using several datasets in the public domain for my reference population samples. HapMap is one of those datasets.

According to its website,

The goal of the International HapMap Project is to develop a haplotype map of the human genome, the HapMap, which will describe the common patterns of human DNA sequence variation. The HapMap is expected to be a key resource for researchers to use to find genes affecting health, disease, and responses to drugs and environmental factors. The information produced by the Project will be made freely available.

In the first phase, it genotyped

30 Yoruba adult-and-both-parents trios from Ibadan, Nigeria, 30 trios of U.S. (Utah) residents of northern and western European ancestry, 44 unrelated individuals from Tokyo, Japan and 45 unrelated Han Chinese individuals from Beijing, China.

In their HapMap phase 3 release #3 (NCBI build 36, dbSNP b126), there are 1,397 samples with about 1,457,897 SNPs each.

I removed related individuals as well as individuals whose genomes were too similar. This left me with a total of 1,149 samples with about 474,606 SNPs in common with 23andme's version 2 data.

Since we are not interested in Native American ancestry, I also removed 58 Mexican samples, thus leaving me with 1,091 samples.

Here are the samples I am using from the HapMap data:

Ethnicity Region Count
African Americans Africa 48
European Americans (Utahns) Europe 111
Han Chinese East Asia 137
US Chinese East Asia 106
Gujaratis South Asia 98
Japanese East Asia 113
Kenyan Luhya East Africa 101
Maasai East Africa 135
Tuscans Europe 102
Yoruba West Africa 140

The region assignments are mine to aid me in the analysis, by including/excluding samples by region or by aggregating results by region to find patterns etc.

It was easiest to use the HapMap data since it's available for download in Plink format.