Low-level contamination confounds population genomic analysis

Genome sequence contamination has a variety of causes and can originate from within or between species. Previous research focused on contamination between distantly related species or on prokaryotes. Here, we test for intraspecies contamination by mapping short read genome data to a reference and visualizing the frequency of reads with single nucleotide differences from the reference. Out of 1,298 publicly available genome sequences investigated for Saccharomyces cerevisiae, a small number (eight genomes) show at least 5% contamination. Contamination rates differed however among sequencing centers: one unusually large study had a low contamination rate (below 0.2%) but the contamination rate was higher for other studies (2% or 15% of genomes). Using genome data contaminated in silico to known degrees, we showed that contamination is recognizable in plots with unexpected secondary allele (B-allele) frequencies of at least 5% and measured contamination effects on admixture and phylogenetic analysis in two fungal species. With a standard base calling pipeline, we found that contaminated genomes superficially appeared to produce good quality genome data. Yet as little as 5-10% genome contamination was enough to change phylogenetic tree topologies and make contaminated strains appear as hybrids between lineages (genetically admixed). We recommend the use of B-allele frequency plots to screen genome resequencing data for intraspecies contamination.