Genetic markers in the playground of multivariate analysis

Abstract

Multivariate analyses such as principal component analysis were among the first statistical methods employed to extract information from genetic markers. From their early applications to current innovations, these approaches have proven to be efficient for the analysis of the genetic variability in various contexts such as human genetics, conservation and adaptation studies. However, because multivariate analysis is a wide and diversified area of statistics, choosing a method appropriate to both the data and to the question being asked can be difficult. Moreover, some particularities of genetic markers need to be taken into account when using multivariate methods. As a consequence, multivariate analyses are often used as black boxes, which results in frequent mistakes in the literature. In this review, we provide a critical analysis of the application of multivariate methods to genetic markers, using a general framework that unifies all these methods for the sake of clarity. First, we focus on some common mistakes in these applications and ways to avoid these pitfalls. We then detail the most critical particularities of allele frequencies that demand adaptations of multivariate methods, and we propose solutions to the subsequent problems. Finally, we tackle several questions of interest in which multivariate analysis has a great role to play, such as the study of the typological coherence of different genetic markers, or the investigation of spatial genetic patterns.

Multiple introductions of divergent genetic lineages in an invasive fungal pathogen, Cryphonectria parasitica, in France

“…We tested for the existence of divergent genetic pools of C. parasitica in France using two methods: a modelbased Bayesian clustering method (Pritchard et al, 2000;Falush et al, 2003) and genetic multivariate analysis (Jombart et al, 2009) to detect genetically differentiated groups corresponding to independent introductions. These methods avoid the clustering of individuals on a priori knowledge such as geographical locations that may mix divergent genetic lineages introduced in the same area and may hinder the detection of admixture events among these lineages.…”

“…We, therefore, used a principal component analysis (PCA) to investigate the genetic structure of the C. parasitica population in France. PCA has become a standard tool to describe genetic structure (Jombart et al, 2009). As PCA is independent of any genetic hypotheses such as Hardy-Weinberg equilibrium, it is suitable for the analysis of partially clonal species.…”

Phylogeographic pattern of range expansion provides evidence for cryptic species lineages in Silene nutans in Western Europe

“…However, the assumption of panmixia does not necessarily hold in S. nutans, which can exhibit a mixed-mating system in natural populations (Dufaÿ et al, 2010). Third, to avoid this potential bias, we performed a multivariate ordination analysis that does not require any genetic assumptions: a spatial principal component analysis (sPCA, reviewed in Jombart et al, 2009). We calculated independent synthetic variables that maximise the product of the genetic variance among populations and their spatial autocorrelation (based on Moran's I).…”

Population genetic structure of serotine bats (Eptesicus serotinus) across Europe and implications for the potential spread of bat rabies (European bat lyssavirus EBLV-1)

“…In contrast to Bayesian clustering, the strength of these approaches rests in their independence from population genetic model, and inferences are therefore made only on allelic similarity (Jombart et al, 2009). The two methods were implemented in the ADEGENET 1.3-4 package (Jombart, 2008) in R (R Core Team, 2014).…”