Clusters and Clines

This paper

Front Genet. 2016; 7: 22.
Published online 2016 Feb 17. doi: 10.3389/fgene.2016.00022
PMCID: PMC4756148
Population Genomics and the Statistical Values of Race: An Interdisciplinary Perspective on the Biological Classification of Human Populations and Implications for Clinical Genetic Epidemiological Research
Koffi N. Maglo, Tesfaye B. Mersha, and Lisa J. Martin

From the abstract: "...contrasts the scientific status of the “cluster” and “cline” constructs in human population genomics, and shows how cluster may be instrumentally produced."

To be frank, I do not yet fully understand the paper.  But this is intriguing (you'll have to read the paper to get the context):

Furthermore, it has been shown that the rate of individuals having membership in multiple clusters increases with the inclusion of admixed populations in studies. This does not however negate the computational possibility of clustering admixed individuals. But under this scenario, many individuals will typically have mixed membership in different clusters (Pritchard et al., 2007; Bryc et al., 2010; Maglo, 2011; Jin et al., 2012). As mentioned above, the correlated allele model was specifically designed to resolve “subtle admixture problems.” Curiously, some researchers perform cluster analysis on admixed populations by bypassing this model (Tang et al., 2005), raising questions about their findings (Graves, 2011). Yet the user guide of Structure states that “Admixture is a common feature of real data, and you probably won't find it if you use the no-admixture model” (Pritchard et al., 2000; Elhaik, 2012).

In a word, computational success does not by itself alone entail the natural reality of clustered entities in evolutionary classification.