First Application of a Distance-based Outlier Approach to Detect Highly Differentiated Genomic Regions across Human Populations

Genomic scans for positive selection or population differentiation are often used in evolutionary genetics to shortlist genetic loci with
potentially adaptive biological functions. However, the vast majority of
such tests relies on empirical ranking methods, which suffer from high
false positive rates. In this work we computed a modified genetic distance on a 10 , 000 bp sliding window between sets of three samples each
from CHB, CEU and YRI samples from the 1000 Genomes Project. We
applied SolvingSet, a distance-based outlier detection method capable
of mining hundreds of thousands of multivariate entries in a computationally efficient manner, to the average pairwise distances obtained from
each window for each CHB-CEU, CHB-YRI and CEU-YRI to compute
the top- n genic windows exhibiting the highest scores for the three distances. The outliers detected by this approach were screened for their
biological significance, showing good overlap with previously known targets of differentiation and positive selection in human populations.