Minor Allele Frequency Filtering Can Strongly Bias Uniqueness Estimates Within and Across Species

Abstract

Minor allele frequency filtering is a common practice in single-nucleotide polymorphism (SNP) dataset construction. It can remove not only potential genotyping errors but also young, local, introgressed, and rare alleles critical to inferring some particular processes or variables, but the effects remain unknown for many downstream analyses. Among them, the evaluation of genetic uniqueness, a fundamental genetic parameter frequently considered in conservation planning and in breeding, may be susceptible because infrequent alleles are relatively unique. Using genome-wide SNPs of two anurans and based on between-individual genetic distances calculated by different methods, with a 0.5-degree grid, I explored the effect of global minor allele frequency filtering (1%, 2%, 5%, and 10%) on within- and across-species uniqueness estimates. At both levels, the deviation from estimates based on unfiltered data increased with the filtering strength, being notable mostly at thresholds of 5% and 10%, for example, a ~750 km shift of the grid cell with the most unique local population. Such impacts were not consistent across species. The effect may depend partly on the percentage of samples that are most genetically unique, so a low threshold does not necessarily guarantee high reliability. In addition, the comparison implies that a relatively unique population can appear to be genetically highly typical (i.e., representative) at higher thresholds. This study provides early evidence for considering a potential adverse effect of minor allele frequency filtering when obtaining and interpreting uniqueness estimates. It is among those calling for further investigation on the effects of parameters and filters involved in the creation of high-throughput sequencing datasets.