GENEALOGY-DNA-L Archives

Archiver > GENEALOGY-DNA > 2005-06 > 1118171439


From: (John Chandler)
Subject: Re: [DNA] DNAPrint strikes again!
Date: Tue, 7 Jun 2005 15:10:39 -0400 (EDT)
References: <1d8.3e62b9a6.2fd6fd5c@aol.com>
In-Reply-To: <1d8.3e62b9a6.2fd6fd5c@aol.com> (DNACousins@aol.com)


Ann wrote:
> wondering whether there would be any utility in merging all the results into one
> "gene pool" and creating a mythical "average" person based on the overall
> frequencies of the various alleles. Might that create a stronger signal of NA
> ancestry?

There are at least two different methods for estimating ancestry from a
list of frequencies and a genotype. DNAprint uses a "brute-force"
maximum-likelihood method -- meaning that they calculate the likelihood
of producing the given genotype from all possible mixtures of the
reference populations and then pick the highest. They could generalize
this method by doing all 100 people in parallel and multiplying together
all 100 likelihoods for each mix and again picking the highest. This
procedure would get around the problem of negative percentages for
some individuals.

The other method is a least-squares fit of the observed allele frequency
of the test subject (either 0%, 50%, or 100% of the mutated allele at
each marker). This method could also be generalized by substituting the
average allele frequency for all 100 people in the fit. Here, too, the
negative percentages for some individuals would automatically be folded
into the ensemble before the fitting takes place.

In both methods, the result from pooling before the fit should be
nearly the same as the result from taking an average of individual
fits, as long as negative percentages are allowed.

John Chandler


This thread: