Originally Posted by
Guiguitargz
Let consider that MYG002 is already admixed once with a local woman, and that MYG001/3/5/6/8 are admixed once more with again local women.
Thus, we could write J-L283-related second generation samples G25 as,
MYG_G2 = MYG002 (we only have MYG002 as a proxy)
Then, third generation samples G25 reads,
MYG_G3 = (MYG001/2. + MYG003 + MYG005 + MYG006 + MYG008/2.)/4.
MYG001 and MYG008 have a half-weight because they share the same mother, and I prefer to have an homogeneous weighting on the individual four mothers of MYG001/3/5/6/8 instead of an homogeneous weighting on the children.
Then, one could provide an estimate of Mygalia tyipcal women,
MYG_M = 2*MYG_G3-MYG_G2
This G25 is fairly similar to other Myceanean peoples (slightly shifted "north").
From here, an estimation of the MYG-L283-MRCA G25 coordinates is given by,
MYG_G1 = 2*MYG_G2 - MYG_M
Whatever is this dude, it is poorly described by the populations we have in our BA "bag".
The noise level on MYG_G1 is higher than the noise level on MYG_M (easily visible from the estimators).
MYG_G1 is most likely "close" from north-western Balkan populations ... but at the same time it didn't completely fit with them.
It looks to me that the source population of MYG-L283-MRCA is something that is not sampled yet (consistently with what is shown by his haplogroup).
PS : for peoples not familiar with such methodology, don't be scared if MYG_G1 shows horrible match whatever the model you try (likely >6% difference no matter what you'll do), this effect comes from the noise augmentation due to the shape of the estimator.
For instance, if you note "e" the typical fluctuation of a G25 (including statistical noise and other processes inducing variance).
MYG_G3 roughly have a noise level of "e/2"
MYG_M have a noise level of "~sqrt(2)*e"
MYG_G1 have of noise level of "~sqrt(10)*e"
To be perfectly correct, the value of "e" depends on both the source population and the quality of the sample (but it is just to give an order of magnitude).
Thus, if in a regular sample to sample match you expect a 2.5% distance to be a decent match, for MYG_G1 it would be "sqrt(11/2)" times bigger, meaning a decent match would be around ~6% (And you likely won't be able to go below that whatever is the considered model).
Reply With Quote
