search

dipetkov / eems

Estimating Effective Migration Surfaces
GNU General Public License v2.0
102 stars 28 forks source link

EEMS on ANGSD's ibsMat #39

Closed z0on closed 5 years ago

z0on commented 5 years ago

Hi Desislava - I am trying to make EEMS work on the IBS-distance matrix from ANGSD ( http://www.popgen.dk/angsd/index.php/PCA_MDS ), and I am having a funny problem: with increasing number of individuals (>300) the determinant of the IBS matrix becomes essentially zero and runeems throws an error. Is there a way around it, do you think? cheers Misha

z0on commented 5 years ago

I think I solved it: looks like raising the IBS matrix to power 1/4 or 1/5 (until there is only a single non-negative eigenvalue from eigen(matrix)$values ) solves it. This transform emphasizes smaller genetic distances slightly, in PCoA plot it is visually imperceptible. Please let me know if this does not make any sense!

z0on commented 5 years ago

Actually, it the problem only seems to arise when I am trying to use subsetted ibsMat (after removing some individuals). An original ibsMat (as it comes out of ANGSD) is all good even with 620 individuals.