Minor allele frequency thresholds dramatically affect population structure inference with genomic datasets

Abstract

Abstract.-Across the genome, the e↵ects of di↵erent evolutionary processes and historical events can result in di↵erent classes of genetic variants (or alleles) characterized by their relative frequency in a given population. As a result, population genetic inference can be strongly a↵ected by biases in laboratory and bioinformatics treatments that a↵ect the site frequence spectrum, or SFS. Yet despite the widespread use of reduced-representation genomic datasets with nonmodel organisms, the potential consequences of these biases for downstream analyses remain poorly examined. Here, we assess the influence of minor allele frequency (MAF) thresholds implemented during variant detection on inference of population structure. We use simulated and empirical datasets to evaluate the e↵ect of . CC-BY-NC-ND 4.0 International license peer-reviewed) is the author/funder. It is made available under a The copyright holder for this preprint (which was not . http://dx.doi.org/10.1101/188623 doi: bioRxiv preprint first posted online Sep. 14, 2017; MAF thresholds on the ability to discriminate among populations and quantify admixture with both model-based and non-model-based clustering methods. We find model-based inference of population structure is highly sensitive to choice of MAF, and may be confounded by either including singletons or excluding all rare alleles. In contrast, non-model-based clustering is largely robust to MAF choice. Our results suggest that model-based inference of population structure can fail due to either natural demographic processes or assembly artifacts, with broad consequences for phylogeographic and population genetic studies. We propose a simple hypothesis to explain this behavior and recommend a set of best practices for researchers seeking to describe population structure using reduced-representation libraries.(Keywords: MAF; population structure; population genetic inference; structure; clustering; genomics; bioinformatics)The distribution of genetic variation within and among individuals is the crucial to understanding the organization of biological diversity and its underlying causes. Across the genome, the e↵ects of di↵erent evolutionary processes and historical events can result in di↵erent classes of genetic variants (or alleles) characterized by their relative frequency in a given population. An excess of common alleles may reflect the signature of population bottlenecks (Marth et al. 2004), purifying selection (Fay et al. 2001), or the absence of population subdivision (Pritchard et al. 2000). Alternatively, high frequencies of rare alleles can provide evidence of population expansion (Marth et al. 2004), detailed information on mutation rates and gene flow (Slatkin 1985), and reveal geographically localized population subdivision (Barton and Slatkin 1986;Gompert et al. 2014). Because the distribution of allele frequencies (also known as the site frequency spectrum, or SFS). CC-BY-NC-ND 4.0 International license peer-reviewed) is the author/funder. It is made available under a The co...

Conservation genetics of the pond bat (Myotis dasycneme) with special focus on the populations in northwestern Germany and in Jutland, Denmark

“…Locus G9 was omitted due to the observed significant genotypic linkage to H29 , and G25 and H19 were omitted due to occurrence of observed rare alleles and low variability. It has been shown by Linck and Battey () that occurrences of rare alleles introduce noise when estimating population structure, blurring the population inference. Consequently, STRUCTURE analysis together with all the data analysis based on microsatellite markers were performed based on the remaining seven loci and the aggregated Danish samples.…”

The comparative biogeography of Philippine geckos challenges predictions from a paradigm of climate‐driven vicariant diversification across an island archipelago

“…Furthermore, when analyzing the RADseq data from three of the Gekko population pairs we are analyzing here, Oaks (2019) found biologically unrealistic estimates of divergence times and population sizes when only unlinked variable sites (i.e., SNPs) were analyzed. Using additional simulations, Oaks (2019) found these unrealistic estimates were likely due to data-acquisition biases, which are known to be common in alignments from reduced-representation genomic libraries (Harvey et al 2015;Linck and Battey 2019). Oaks (2019) found that using all of the sites, rather than only SNPs, greatly improved the robustness of these parameter estimates to such acquisition biases.…”

Vanishing refuge? Testing the forest refuge hypothesis in coastal East Africa using genome‐wide sequence data for seven amphibians

“…For A. xenodactyloides , dapc and admixture results were also incongruent ( k = 4 and k = 8, respectively). As dapc is known to be less sensitive to minor allele frequency thresholds than other model‐based clustering methods (Linck & Battey, ), we based subsequent A. fornasini analyses on the dapc results given their congruence with phylogenetic and phylogenomic inferences. Similarly, for A. xenodactyloides , subsequent analyses were based on k = 3 following consistently lowest BIC and CV scores in dapc and admixture analyses (Supporting Information Table ), and consistency with phylogenetic clustering.…”

“…We investigated population structure per taxon using discriminant analysis of principal components ( dapc ) in the adegenet r package (Jombart & Ahmed, ), after converting stacks output files into Fstat format using pgdspider 2.1.0.3 (Lischer & Excoffier, ). Unlike model‐based clustering methods, the dapc method is free of assumptions regarding Hardy–Weinberg equilibrium (Jeffries et al., ; Jombart & Ahmed, ) and less sensitive to minor allele frequency thresholds (Linck & Battey, ). We defined values of k between 1 (i.e., a single panmictic population) and 8, using Bayesian information criterion (BIC, Schwarz, ) scores across tested k values to infer the number of populations.…”