Evaluating methods to visualize patterns of genetic differentiation on a landscape

Abstract

With advances in sequencing technology, research in the field of landscape genetics can now be conducted at unprecedented spatial and genomic scales. This has been especially evident when using sequence data to visualize patterns of genetic differentiation across a landscape due to demographic history, including changes in migration. Two recent model-based visualization methods that can highlight unusual patterns of genetic differentiation across a landscape, SpaceMix and EEMS, are increasingly used. While SpaceMix's model can infer long-distance migration, EEMS' model is more sensitive to short-distance changes in genetic differentiation, and it is unclear how these differences may affect their results in various situations. Here, we compare SpaceMix and EEMS side by side using landscape genetics simulations representing different migration scenarios. While both methods excel when patterns of simulated migration closely match their underlying models, they can produce either un-intuitive or misleading results when the simulated migration patterns match their models less well, and this may be difficult to assess in empirical data sets. We also introduce unbundled principal components (un-PC), a fast, model-free method to visualize patterns of genetic differentiation by combining principal components analysis (PCA), which is already used in many landscape genetics studies, with the locations of sampled individuals. Un-PC has characteristics of both SpaceMix and EEMS and works well with simulated and empirical data. Finally, we introduce msLandscape, a collection of tools that streamline the creation of customizable landscape-scale simulations using the popular coalescent simulator ms and conversion of the simulated data for use with un-PC, SpaceMix and EEMS.

Reshaping theHexagone: the genetic landscape of modern France

“…Furthermore, the barrier in the northwestern corner is consistent with the lower levels of allele sharing we found between the Breton departments and the southeastern corner of France. Apart from some possible artifacts due to the presence of unsampled regions (House et al, 2018), we found the EEMS results consistent with the rest of the data collected in our study. Indeed, another group with a higher internal migration rate according to the EEMS results is represented by the north and the northeastern departments.…”

Demographic and genetic approaches to study dispersal in wild animal populations: A methodological review

“…Boundary‐based methods include edge detection techniques (e.g., Cercueil, François, & Manel, ; House & Hahn, ; Jombart, Devillard, & Balloux, ; Jombart, Devillard, Dufour, & Pontier, ; Monmonier, ; Piry et al., ), and Bayesian clustering algorithms (reviewed in François & Durand, ) aim to delineate discrete or admixed populations in space (Wagner & Fortin, ). They allow identifying spatial genetic boundaries that are visually (e.g., Frantz et al., ; Prunier et al., ) or statistically (e.g., Balkenhol et al., ; Jay et al., ; Murphy, Evans, Cushman, & Storfer, ) compared to landscape patterns.…”

ResDisMapper: An r package for fine‐scale mapping of resistance to dispersal

“…Although intuitive, there are only a few such programs available and all of them were published very recently. Un‐PC (House & Hahn, ) is an r package that projects principal components analysis (PCA)‐based genetic distances of pairs of populations (the “ un‐PC ” values) as ellipses to highlight areas of extreme genetic differentiation and similarity, and therefore areas of high and low resistance. Un‐PC is model‐free in the sense that it does not specify a particular mechanistic model underlying the observed genetic pattern.…”

“…However, mapi directly projects ellipses representing genetic distances between pairs of individuals, which ignores population‐level correlations between genetic distance and environmental features. Moreover, mapi is implemented in the open source database postgresql (Momjian, ), which requires relatively complicated database installation and configuration (House & Hahn, ). In contrast, a third program, DResD , is presented in the form of an r script and calculates resistance using population‐level information (Keis et al, ).…”