Non-parametric Analysis of Population Structure

Posted by James Bowery on June 07, 2010, 11:01 AM | #

The most general metric is minimum description length.  I suspect that in the absence of quantum “computers” (necessary for optimally choosing the MDL “code”), the closest practical approximation will be Kolmogorov complexity.  In that event I further suspect a fruitful line of research will involve SA/GP hybrid search strategies—some of which have already shown promise in other areas.

Posted by Guessedworker on June 07, 2010, 05:27 PM | #

Based on the results of this paper, in particular the nonexistence of the Kalash as a separate cluster using MGD, it is questionable as to whether structure always provides the ‘most correct’ answer.

So if structure is not the whole solution but must be augmented, on what consistent basis can you eliminate the negative results of possible augmentary methods and not be accused of simply picking the answer that is correct for you?

Posted by Dasein on June 08, 2010, 07:54 AM | #

So if structure is not the whole solution but must be augmented, on what consistent basis can you eliminate the negative results of possible augmentary methods and not be accused of simply picking the answer that is correct for you?

The ideal consistent basis would be a stranglehold control over media and academia that forces acceptance of a method politically favourable to Europeans  

Kidding aside, I think when parametric and non-parametric methods give different results (however different something needs to be to be considered noteworthy), then at the very least the assumptions of the parametric method need to be questioned.  I’m not aware that Kalash have been evaluated for levels of Hardy-Weinberg and linkage disequilibrium that would make them inappropriate for inclusion in structure analysis.  Perhaps they have been.  To make what could be a long and convoluted answer short, I don’t think we can really speak of a ‘correct’ answer.  There are trade-offs, and correct is dependent on the context.  Parametric methods are more powerful than non-parametric ones (and we see above that the parametric method, structure, works better when differentiation is very low), but one has to be careful about when to use them.  As for context dependence, if you’re looking to model historic migration patterns, then Fst/Gst might be the best solution.  If you’re looking to quantify population diversity using very heterozygous markers, like microsatellites, Fst/Gst is a bad choice.  A distance measure that accounted for allelic correlation structure would find lots of practical application to questions of interest to us, like how to quantify genetic interests.

Posted by James Bowery on June 08, 2010, 10:31 AM | #

Dasein writes: “The data set used is the HGDP microsatellite markers from Rosenberg’s 2002 paper.”

GW asks: “...not be accused of simply picking the answer that is correct for you?”

To put my answer into more concrete terms:

You take the taxonomy produced by each other the methods and use them to compress the data set.  The most compressed form of the data set (fewest total bits—including the decompression algorithm) is that based on the most generally “correct” taxonomy given what is known (the data set).  That is to say, it will more often produce correct predictions.

Posted by Fred Scrooby on June 22, 2010, 07:59 AM | #

A little off topic (though Salter is mentioned in the entry):  I consider the following new piece by Frank Salter a must-read — I cite it here because it’s just gone off the front page over at Richard Spencer’s, so guys who wander over there to browse may miss it:

http://www.alternativeright.com/main/blogs/hbd-human-biodiversity/open-borders-closed-minds/ .

It consists of a series of rejoinders to academic race-replacement advocacy.  Be careful, it’s often deceptively concise and compact, so that Prof. Salter may occasionally do stuff like condense three major show-stopping arguments in a few words casually hidden in half-a-sentence or so:  although it’s simple you have to read it carefully to be sure of getting all the meat.

No one on our side ought to be unfamiliar with any of the counter-arguments to race-replacement which Prof. Salter touches on in this extremely important piece, so simply-worded yet so jam-packed.

Posted by Fred Scrooby on June 22, 2010, 08:10 AM | #

(I’m sure I had seen this new Salter piece discussed briefly somewhere before I read it — if it was somewhere here at MR.com that I saw it forgive me for not referring back to that; I can’t recall where it was and don’t have time at the moment to hunt for it.)

Posted by Armor on June 22, 2010, 08:35 AM | #

K.MacDonald, in his blog, called attention to Salter’s article.