search this blog

Monday, July 11, 2016

Genome-wide variants of Eurasian facial shape differentiation

Very interesting new preprint at bioRxiv:

Abstract: It is a long standing question as which genes define the characteristic facial features among different ethnic groups. In this study, we use Uyghurs, an ancient admixed population to query the genetic bases why Europeans and Han Chinese look different. Facial trait variations were analyzed based on high dense 3D facial images; numerous biometric spaces were examined for divergent facial features between European and Han Chinese, ranging from inner-landmarks to dense shape geometrics. A series of genome-wide association analyses were conducted on a discovery panel of Uyghurs. Six significant loci were identified and four of which, rs1868752, rs118078182, rs60159418 at or near UBASH3B, COL23A1, PCDH7 and rs17868256 were replicated in two independent cohorts of Uyghurs or Southern Han Chinese. We further developed a quantitative model to predict 3D faces based on 277 top GWAS SNPs. In hypothetic forensic scenarios, this model was found to significantly enhance the rate of suspect verification, suggesting a practical potential of related research.

Lu Qiao et al., Detecting Genome-wide Variants of Eurasian Facial Shape Differentiation: DNA based Face Prediction Tested in Forensic Scenario, bioRxiv, posted July 11, 2016, doi:


Davidski said...

Here's a PCA done with 105 of the markers, including present-day and a few ancient samples.

Matt said...

Cool. Ancient samples mostly overlap with present day West Eurasians, not out beyond them on the PC1 separating East Asian from Europeans. Suggests the ancients may have genetic variants which distinguish their facial shape from moderns but if so these are orthogonal to the East Asian->European facial genetic dimension. PC2 separates out Papuans from the intermediate Uyghur like position and PC3 separates Americans from East Asians (particularly some SE Asian samples), even though this panel of markers was not selected with Papuan or American populations in mind. Kostenki looks to have a PC2 shift away from Papuans, then not related to whatever shifts the East and West Eurasians in PC1.

If you do have any chance to rerun those PCA, with Africans and Uyghurs, included? I think that might show whether the Papuans are falling at the position they have because they are falling at an ancestral African-like position on these markers, or whether they've partly taken part in the Asian shift, but only partially.

(This is all assuming their GWAS captures the face structure more than it is just neutral AIMs).

Earlier papers on the same concept in African-European and Latin American groups: - use of those identified SNPs in predictive models

Davidski said...

Yeah, pretty much.

Gui S said...

Fascinating stuff!
Worth noting there are other studies like this one (unlike what is claimed in their PDF).
or (same stuff)

If you compile those three papers, plus the classic pigmentation-genes. You can use the following genes to create phenotype-PCAs:

(on top of the ones from the new paper).

Gui S said...

Ah I see Matt's beaten me to it.
Still, would be interesting to create admixture runs using only a combination of all of these genes (rather than the single associated SNPs).

Davidski said...

Yeah, true. I'll try and do a more involved PCA of all of the available markers from the face shape genes.

Matt said...

Thanks David. Yeah, Papuans don't have any overlap with Africans really on those, except for both being intermediate on the East-West Eurasian dimension. Population averages from those two PCA:

(With that many markers you might need a lot of samples before populations become closest to their closest relatives, or populations may just really not be closest to their closest relatives on these markers, even if they're closest to their closest relatives on the overall set of variants that cause facial shape.)

Seinundzeit said...

This is pretty interesting.

I wonder how MA1 and Afontova Gora3 would plot, considering that anthropologists used to construe them as broadly East Asian in terms of phenotypic affinity (supposedly, due to high cheekbones, flattish midface, non-prominent nasal bones, etc)?

I mean, even K14 seems very East Asian-like, compared to living West Eurasians.

Also, I wonder if scheduled caste South Indians would deviate in the direction of Papuans/Australians, on the Eurasian plot?

In terms of cranio-facial morphology, even scheduled caste South Indians are closest to Europeans/West Asians/South Central Asians, but as per the old physical anthropology, they supposedly deviate from Europeans/West Asians/South Central Asians in the direction of Australasian-looking people.

Not sure if that's a product of confusion on the part of those old anthropologists, or a real physical similarity. Perhaps plotting them on something like this could be informative about that.

Davidski said...

Here's an Admixture run based on ~750 SNPs from most of the genes listed above by Gui.

I'm PL1, Gui is FR7, and Sein in Pathan1.

I can't run other people at the moment. Maybe one day.

Seinundzeit said...


This is pretty awesome, thanks for including me!

My percentages make sense, as outside South Central Asia it seems my closest match is an individual Chechen. Nice, since that's how I look IRL, like a cross between an Uzbek from Central Asia and a Dagestani from the Caucasus.

(My combo seems rare though, a similar ratio of southern to northern is hard to find without an African component, which I lack in this context. A lot of West Eurasians are getting those African percentages, I wonder what sort of phenotype complexes that component maps unto, in non-Africans)

Gui S said...

Excellent David, by the way, I am sorry not to have included the genes associated in the study you posted in my list. But if you can make a new version including UBASH3B, COL23A1, PCDH7 from the new study as well, it would be even more amazing!


Davidski said...

Yep, that run includes SNPs from those genes as well. I was only able to get 768 SNPs from all of those genes, including the 277 SNPs listed in the study.

A PCA based on the Admixture results looks a lot like a low res global plot.

Davidski said...

Actually, here's a PCA datasheet.

Sgt said...

Very interesting!
RE: Pheno.4.Q - Admix
Sorting clines, the Northern seems to have great latitude with a Bosnian(modal individual), Sephardic-Jew, Ashkenazi-Jew, Tuscan, Greek, Spanish, Armenian, Columbian in the top 1%

Here's a couple of population averages:
Southern Northern Sub-Saharan East_Asian
AshkenazJ 40.83% 45.38% 2.98% 10.82%
So.Italian 34.47% 50.13% 7.62% 7.79%
Palestinian 44.19% 37.34% 13.35% 5.12%
Norwegian 26.76% 57.40% 3.87% 11.97%
Lithuanian 28.56% 59.30% 3.64% 8.50%

Gui S said...

Impressive results Davidski, you must look like a paleolithic European fossil! You are closest to Loschbour (and then Lviv).
I made a little gradient map for my own results here: (I am most similar to Peloponnese Greeks, South Italians, Sephardi and Moroccans!!!)
I'll make you one for Sein, Davidski, Andronovo, Loschbour and maybe more if I have time.

Davidski said...

Yeah, I was certainly taken aback by my results. A map like yours above would be great.

Chad Rohlfsen said...


Could I get a list of those SNPs please? Thanks!

MfA said...

Interesting run,

This is mine

Southern 34%
Northern 43%
SSA 0%
East Asian 23%

Matt said...

Took those ADMIXTURE results, made group averages, put in a PCA and neighbour joining clusters:

Neighbour joining:

It seems like the Southern component loads most on both South Asian ancestry and "other" West Eurasian, while Northern on European of all sorts, but more so to "Euro HG" (with Loschbour and Stuttgart both seem a bit outlying).

Seems to make some sense in that these are SNPs which are mostly ascertained to distinguish European phenotypic variation vs African and East Asian. To some extent vs Native American in the admixed Latinos study (if you were able to include anything from that David), but that might end up weighted more weakly than the other two I think.

Populations that IRC probably show some divergence phenotypically don't end up as diverged in the tree for those kind of reasons that the SNPs aren't as informative for them - like Kharia sitting on a branch with Papuans or Karitiana with Burmese and Malays or Columbians sitting in the Afghan-Pakistan part of the tree.

Might've been interesting to see if slightly different SNPs came together if they'd been able to use an Iranian-Han model for Uyghur, in face shape and genes, since that's maybe slightly closer to the real populations.

Matt said...

Same results for group averages with the PCA:

Neighbour joining:

PC1 vs PC2 looks quite the classical for world PCA, while PC3 is very strongly marked by a contrast between Loschbour and to a lesser extent Europeans and Siberians vs Amerindians and Papuans.

Gui S said... Davidski Seinundzeit
Andronovo and Loschbour up next.

Davidski said...

Thanks. More or less correct, I'd say.

Seinundzeit said...

Gui S,


Yeah, it seems affinity to myself peaks in Afghanistan/northern Pakistan and in the Caucasus, so this does make sense.

Ryan said...

I'm not sure how much work it requires, so I'm not sure how feasible it is, but could this be turned into a DIY calculator? I think it would be quite fascinating.

Davidski said...

I think Chad and Vincent are working on something like that for GEDmatch.

If they don't get anything up by next week I'll post some DIY files here.

Jack Rusher said...

Do the publicly available Neanderthal and Denisovan samples have the SNPs used here? It would be interesting to see where they fall using this measure. Also, I'd be interested in the SNP list. :)

Gui S said... Andronov.
I didn't end up doing one for Loschbour, because it was basically be the same as Davidski.

Davidski said...

I didn't end up doing one for Loschbour, because it was basically the same as Davidski.

Haha. Love it.

Gui S said...

Perhaps that's a bit too creepy or a bit of a privacy issue. But it could be interesting to get closest IBS matches on these SNPs on gedmatch to "find your doppelganger" in the database.

(you could also setup a matrimonial system based on MHC compatibility).

Karl_K said...


"Do the publicly available Neanderthal and Denisovan samples have the SNPs used here? It would be interesting to see where they fall using this measure."

They are high quality genome sequences, so the SNPs are known.

However, they probably had other SNPs that effect face shape that didn't make it into modern gene pools.

Still, it would be interesting to see if any of these SNPs were directly inherited from archaics.

Gui S said...

Made some simulations for the components (using substractive morphing and my large personal collection of averages [ie. to get Northern out of Lviv 200% vectorial morph translation from Kalash42%-Saudi68% Hybrid to Lviv, repeated with 5 different groups per component]).
Got the following: [Southern] [Northern] [Subsaharan African] [East Asian]

I then made some predicted faces [Andronovo] [Loschbour] [Mota] [Ust-Ishim]

Caveats, it didn't do a very good job of replicating my own face (, but that's understandable considering the methods used. It didn't do a very good job at coming up with a Papuan-looking Papuan-simulation (, which does indicate we need to uncover quite a few more-markers.

Grey said...

Fascinating stuff - especially Ust-Ishim

rozenfag said...

New paper about Zagros farmers:

We sequenced Early Neolithic genomes from the Zagros region of Iran (eastern Fertile Crescent), where some of the earliest evidence for farming is found, and identify a previously uncharacterized population that is neither ancestral to the first European farmers nor has contributed significantly to the ancestry of modern Europeans. These people are estimated to have separated from Early Neolithic farmers in Anatolia some 46-77,000 years ago and show affinities to modern day Pakistani and Afghan populations, but particularly to Iranian Zoroastrians. We conclude that multiple, genetically differentiated hunter-gatherer populations adopted farming in SW-Asia, that components of pre-Neolithic population structure were preserved as farming spread into neighboring regions, and that the Zagros region was the cradle of eastward expansion.

bellbeakerblogger said...

Wow, Zargos Iron Age F38 is R1b1a2a2-CTS1078/Z2103

jparada said...

can't see the ust-ishim sim, for some reason. Papuan looks rather Central Asian.

Grey said...


i had to right click and select "reload" to see some of them - don't know why

Seinundzeit said...

Gui S,

Your IBS idea is pretty interesting. Kurd at Anthrogenica has given it a try, and he was kind enough to run my data. My top 8 IBS matches for SNPs affecting phenotype:

Kurdish individual 1: 77.26%
Kurdish individual 2: 75.72%
Pakistani individual: 75.65%
Pashtun_Afghan Pashtun10_17Af: 75.26%
Turkish individual: 75.26%
North Indian Brahmin individual: 75.11%
Pashtun_Afghan Pashtun2_8Af: 75.08%
Tajik Individual: 74.56%

The whole list is at Anthrogenica, and includes many populations/individuals of diverse background (ancient populations, South Asians, West Asians, Europeans, etc).

It's pretty amazing that one can now attain such accuracy concerning a person's phenotype, using genetic data. This reflects how I look IRL very well, as I can pass as a local in places as far west as Turkey, as far east as the Pashtun portion of Pakistan, as far north as the northern Caucasus + Central Asia (Tajikistan and Uzbekistan), and as far south as Balochistan.

We live in interesting times.

Fun sidenote, I am MA1's closest match, in the data-set he ran, which is pretty cool. My percentage with MA1 is 71.75%.

Titane said...

To Davidski
What happened to the idea of having a DYI?
To all
All simulations are of men. Does it work for women?

Davidski said...

I'll throw something on GEDmatch this week.

Why wouldn't it work for women?

Titane said...

Thanks, I can't wait!!!
. About women, it's just I have never seen an example. How is the hair length decided - that is not a genetic trait...

Samuel Andrews said...


An ADMIXTURE test won't give a prediction of how you look. Hair SNPs aren't gonna be in the test anyways.

Davidski said...

This test will take a while. It's at the back of the to do queue at the moment.

Titane said...

@samuel Andrews
I don't see why not, since Lochsbourg looks like he could be my cousin.
I know the fun stuff always has to wait...

Simon_W said...

Anyway, two West Eurasian phenotypic components (a northern and a southern one) are not enough to capture the phenotypic variation between populations, because different populations combine traits differently. E.g. not all southerners have high rooted noses, e.g. the EEF had rather broad shaped noses. And not all northerners have typically broad shaped faces, e.g. Scandinavians have rather narrow shaped averages. You could model this as being partly southern, but their pigmentation on the other hand is clearly northern.

Davidski said...

Scandinavians as a rule have long and broad faces; generally quite large and unlike southern Europeans.

If you're referring to the so called Nordid anthro type, that's a circular argument; since Nordids are basically Northern Europeans with narrow faces, then Nordids and thus the archetypal physical type of Northern Europe has a narrow face.