Contrasting whole-genome and reduced representation sequencing pipelines for population demographic inference

Genomic approaches to the study of contemporary and historic population demography often rely on accurate SNP calling and by-proxy the site frequency spectrum (SFS). Two main questions for the design of such studies remain unanswered: do reduced genomic sequencing summary statistics reflect that of whole genome sequencing, and how does these sequencing strategies and summary statistics impact downstream demographic inferences? To address those questions, we applied the ddRAD sequencing approach to 254 individuals and whole genome resequencing approach to 35 mountain goat (Oreamnos americanus) individuals across the species range. We called variants with 5 different software packages and used ANGSD to estimate the genotype likelihoods. We tested different combinations of SNP filtering by linkage disequilibrium (LD), minor allele frequency (MAF) and the genomic region. We compared the resulting SNP datasets in terms of a suite of summary statistics reflective of the SFS and quantified the relationship to population demographic inferences by estimating the contemporary effective population size, isolation-by-distance and population structure, FST, and explicit modelling of the demographic history with DADI. Both MAF and LD filtering shifted the summary statistics and considerably influenced the demographic inferences. Estimates of contemporary effective population size and isolation-by-distance patterns were largely robust to the choice of pipeline and filtering. Evaluation of population demography based on whole genome and reduced representation approaches was similar, and we argue that with lowering sequencing costs whole genome resequencing data is preferable as it can address a wider array of questions.