Marker overlap and test accuracy

A few people are asking me about the effects of marker overlap or genotype rate on test accuracy. Logic dictates that the better the overlap, the more accurate the results, but this isn't strictly true. Here's what I've learned over the years:

- accuracy doesn't necessarily improve with higher marker overlap, it improves (up to a certain point) with more markers

- you will still see accurate results using as little as 25,000 SNPs, as long as the test doesn't suffer from any serious problems

- poorly designed tests, such as those based on less than 1000 reference samples, always produce garbage results no matter what the marker overlap

In other words, a well designed test based on 200,000 SNPs will produce very accurate results for a genotype file with a marker overlap of 50%. On the other hand, another well designed test, based on just 50,000 SNPs, is likely to produce less accurate results for a genotype file with a marker overlap of 100%.

So how can you tell a well designed test from a poorly designed one? It's easy, just have a look at the results they're producing for people with less complex ancestry. For instance, ask a Lithuanian, Swede or Pole what they're seeing at the top of their oracles. Is the Swede seeing Swedish or, say, German? If the answer is German instead of Swedish, or at least some type of Scandinavian, then the test is garbage and best ignored.

By the way, the recent Allentoft et al. paper on the ancient genomics of Eurasia includes a useful discussion on the effects of missing markers on the accuracy of both ADMIXTURE and PCA results. Refer to section 6.2 in the freely available supplementary info PDF here.

4mix: four-way mixture modeling in R

Thanks to Eurogenes project member DESEUK1. A zip file with the R script, instructions and a couple of data sheets is available here.

So let's model Poles as a bunch of ancient genomes from Central and Eastern Europe using output from my K8 analysis.

Copy & Paste: source('4mix.r')

Hit ENTER

Copy & Paste: getMix('K8avg.csv', 'target.txt', 'HungaryGamba_EN', 'HungaryGamba_HG', 'Karelia_HG', 'Corded_Ware_LN')

Hit ENTER

After a few seconds you should see the results...

Target = 19% HungaryGamba_EN + 14% HungaryGamba_HG + 2% Karelia_HG + 65% Corded_Ware_LN @ D = 0.0062

Obviously the script can use ancestry proportions and/or population averages from any test, provided they're formatted properly. The accuracy of the modeling will depend on the quality of the input.

Update 19/05/2015: A new version of the 4mix script that can run multiple targets is available here, courtesy of Open Genomes.

Short clip: The making of modern Europe

Simple but, I think, very cool animation: ten ancient genomes analyzed with the Eurogenes K15. More elaborate clips are on the way.



And this is basically the same thing, but restricted to samples from Hungary.

Model yourself as a mixture of ancient genomes

Update 12/05/2015: 4mix: four-way mixture modeling in R

...

This is really easy and should work well for most personal genomics customers (ie. those of European ancestry and with data files from 23andMe, FTDNA and AncestryDNA).

First of all, make sure you have your Eurogenes K15 ancestry proportions from GEDmatch. Then do the following:

- download the 4 Ancestors Oracle (here)

- download the Eurogenes ancient genomes datasheet (here)

- place everything into the same directory

- double click of the 4 Ancestors Oracle icon (the big red number 4)

- select the Eurogenes K15 ancient genomes datasheet

- type your Eurogenes K15 ancestry proportions into the fields provided

- hit the go button and let it rip

I'm not sure I'm allowed to upload the 4 Ancestors Oracle online, but I couldn't find the original link, so let's assume for the time being that I am. In any case, many thanks to Alexandr Burnashev for this great tool.

You'll also find some modern populations in the datasheet. They're there so that users with ancestry from outside of Europe don't end up with ridiculous results.

Obviously, you can edit the datasheet to explore more options by removing or adding individuals and populations. A spreadsheet of Eurogenes K15 population averages is available here. The oracle settings can also be tweaked in a couple of ways to fine tune the results.

If the calculator crashes, try replacing the periods with commas in both the datasheet and your ancestry proportions.

Please keep checking this post, because I'll attempt to update the datasheet at the link above every time a new ancient genome is published and has enough markers available to be tested with the Eurogenes K15. Eventually we might end up with a tool that covers most of the continents and many periods of history and prehistory.

I've done similar analyses of a variety of ancient genomes. For instance, StoraFörvar11, or SfF11, from Mesolithic Sweden came out 3/4 La Brana-1 and 1/4 MA-1, which translates to 3/4 Western European Hunter-Gatherer (WHG) and 1/4 Ancient North Eurasian (ANE), and lines up well with results reported recently for Swedish hunter-gatherers in scientific literature. You can see the full analysis StoraFörvar11 and a couple of other ancient genomes at the links below.

Analysis of Mesolithic Swedish forager StoraFörvar11

More ancient genomes from Sweden: Pitted Ware forager Ajvide58 and TRB farm girl Gokhem2

I'm still trying to answer a whole lot of e-mails so I won't be monitoring this post for a while. But please feel free to share your results and any tips you might have in the comments below.

EEF-WHG-ANE test for Europeans

This test attempts to fit you to the three inferred prehistoric European populations as described in this recent preprint. The relevant Excel file can be downloaded here, and all you have to do is stick your Eurogenes K13 results into the fields provided to get the EEF-WHG-ANE ancestry proportions. A modified version for Near Eastern and Southeast European users can be accessed here.

The test is based on correlations between the average levels of the Eurogenes K13 and the ancient components among selected European populations. Below is a brief description of each of the ancient components.

Early European Farmer (EEF): apparently this is a hybrid component, the result of mixture between "Basal Eurasians" and a WHG-like population possibly from the Balkans. It's based on a 7500 year old Linearbandkeramik (LBK) sample from Stuttgart, Germany, but today peaks at just over 80% among Sardinians.

West European Hunter-Gatherer (WHG): this ancestral component is based on an 8,000 year old forager from the Loschbour rock shelter in Luxembourg, who belonged to Y-chromosome haplogroup I2a1b. However, today the WHG component peaks among Estonians and Lithuanians, in the East Baltic region, at almost 50%.

Ancient North Eurasian (ANE): this is the twist in the tale, a component based on a 24,000 year old Upper Paleolithic forager from South Central Siberia, belonging to Y-DNA R*, and known as Mal'ta boy or MA-1. This component was very likely present in Southern Scandinavia since at least the Mesolithic, but only seems to have reached Western Europe after the Neolithic. At some point it also spread into the Americas. In Europe today it peaks among Estonians at just over 18%, and, intriguingly, reaches a similar level among Scots. However, numbers weren't given in the paper for Finns, Russians and Mordovians, who, according to one of the maps, also carry very high ANE, but their results are confounded by more recent Siberian (ENA) admixture.

It's important to note that this test is only likely to be accurate for people of European ancestry, and indeed only those who aren't outliers from the main European clines of genetic diversity. For details of what that means, please consult the aforementioned paper. However, roughly speaking, if you're of European origin and don't score more than 3% East Asian, Siberian, Amerindian, South Asian, Oceanian, Northeast African and/or Sub-Saharan admixture, then you should get a coherent result. Users from the Near East and Caucasus should run the version specifically designed for them, while those from Southeastern Europe might find it useful to run both calculators and then compare the results.

Thanks to project member DESUK1 for putting this together at such short notice, and MfA for the modified version. Please post your results in the comments section below and state your ancestry when you do. This will help us to improve the accuracy of the test. My results make perfect sense, considering my Polish ancestry.

EEF 42.012706
WHG 40.52702615
ANE 17.46026785

This is my interpretation of who these components represent. Of course, this model might change when more ancient genomes are analyzed.

WHG and WHG/ANE: indigenous European hunter-gatherers
EEF: mixed European/Near Eastern Neolithic farmers
ANE/WHG: Proto-Indo-European invaders from the Eastern European steppe
ENA/ANE: early Uralics from the Volga-Ural region
EEF/WHG/ANE: late Indo-Europeans (ie. Celts, Germanics and Slavs)

Citation...

Iosif Lazaridis, Nick Patterson, Alissa Mittnik, et al., Ancient human genomes suggest three ancestral populations for present-day Europeans, bioRxiv, Posted December 23, 2013, doi: 10.1101/001552

See also...

Ancient human genomes suggest (more than) three ancestral populations for present-day Europeans

Ancient North Eurasian (ANE) levels across Asia

Updated Eurogenes K13 now at GEDmatch

The new K13 population averages and genetic (Fst) distances between the inferred ancestral clusters are available here and here, respectively. To find this test at GEDmatch do this:

GEDmatch > Ad-Mix Utilities > Eurogenes > K13

Below is a 2D PCA based on the average K13 results of the European and Asian reference populations, courtesy of project member PL16.

I now have four tests at GEDmatch with Oracles: the Jtest, EUtest, K15 and K13. It's useful to keep in mind that these tests will differ in their interpretation of the data, and perhaps accuracy, depending on the ancestry of the user. For instance, the new K13 should be more useful for Central and South Asians than any of the others, because it features new reference samples relevant to them.

Eurogenes K15 now at GEDmatch

This new test is essentially an upgraded version of the EUtest. Unlike the original, it includes an Amerindian component and five native reference populations from North and Central America. So obviously it should be a lot more useful for users from the New World who are wondering about Amerindian admixture.

GEDmatch > Ad-Mix Utilities > Eurogenes > Eurogenes EUtestV2 K15

I just tried it myself, and have say that the 4-Ancestors Oracle results were impressive. In other words, they were very accurate based on what I know about my recent ancestry. On the other hand, I'd say the default Oracle was picking up more ancient gene flows. However, this might not be the case for everyone, so let's hear some feedback, discuss the outcomes, and perhaps tweak the settings if necessary.

One of the most important things to keep in mind is to ignore all results under 1%. These are likely to be noise.

The population averages and Fst distances between the ancestral clusters are here and here, respectively. Below are spatial maps of the main West Eurasian components courtesy of Gui (FR7): Baltic, North Sea, Atlantic, East Euro, West Med, East Med, West Asian.

See also...

Orcadians, the K15 and the calculator effect