Admixture estimate Vs. Population-of-origin estimate

It is important to distinguish between classification of an individual vs. admixture analysis.

The classification problem assumes that each individual has a single origin, e.g., "NW European" or "SE European" and attempts to guess this origin.

For example, EURO-DNA-CALC 1.02 is a classifier, and tries to guess an individual's origin out of the set {"NW EUropean", "SE European", "Ashkenazi Jewish"}.

The admixture estimate problem assumes that each individual is a mixture of groups, and attempts to guess the relative strength of the components in the mixture.

EURO-DNA-CALC 1.1.1 is an admixture estimator, and tries to guess an individual's admixture from the groups "NW EUropean", "SE European", "Ashkenazi Jewish".

In many of the testees that have communicated their results to me, the classification provided by 1.0.2 is equal to the most important component identified by 1.1.1. But this does not need to be the case.

To understand why, let's consider some real-life examples.

The Proto-Uralics

Using skull measurements, scientists have identified traits that seem to have belonged to a Proto-Uralic human group. Presently, individuals of Northern Eurasia, especially Uralic speakers are a mix of Western Eurasian (Caucasoid) and Eastern Eurasian (Mongoloid) elements, but they also possess a small and varying component of this old Proto-Uralic group.

So, an individual could be guessed as a Uralic speaker with some success, and yet his Uralic admixture would be generally low.

African Americans

It is well-known that African Americans have European admixture which varies a lot individually. Indeed, some African Americans are more European genetically than they are African. On the other hand, European Americans generally tend to have insignificant African admixture. This assymetry is due to the fact that individuals of partial African ancestry were reckoned to be "Black", no matter what the exact percentage of their African ancestry.

Now, a classifier that would distinguish between Whites and African Americans would have no trouble distinguishing between the two groups easily. But this classification would not need to be concordant with their admixture estimate:

A person of 25% African ancestry is extremely more likely to be (in the US) an African American than a White. And yet, the high classification probability that he is African American is discordant with their most important component, i.e., European.

The "Brown" Cluster

In the recent paper on human genetic variation using 650K SNPs, a "Brown" cluster was identified. Middle Eastern populations seemed to have a high fraction of their ancestry from this cluster:


If we built a classifier to distinguish between Europeans and non-Europeans, the presence of the "Brown" cluster would give us near certainty that an individual is non-European and yet there are non-European individuals (e.g., the Druze), where the "European" component (average for the group 52%) is greater than the "Brown" one (34%).