Towards a Standard Threshold for Genome Wide Significance in Dogs

Genome-wide association studies (GWAS) are a foundational step in tying phenotype to genotype, relying on statistical significance thresholds to distinguish true- from false-positive signals of association. Dog genomics has long relied on per-study Bonferroni thresholds of significance, basing these on SNP chip levels of markers (~100 k to > 14 M variable sites). However, as the field progresses into whole genome imputation analyses and more powerful meta-analyses, there is a clear need to develop a standard significance threshold for common-variant GWAS. Using 1591 dogs from the broad-ancestry Dog10K dataset, we performed permutation analysis and developed GWAS thresholds for datasets using either 1% or 5% minor allele frequencies. The resultant p-values, 4.2 × 10^-7 and 5.0 × 10^-7 respectively, are similar to previous Bonferroni levels (p-value ~6 × 10^-7), but less restrictive than the standard human p-value, 5 × 10^-8, which is sometimes used in dog studies. Given the diverse haplotypes from the > 320 breeds in the Dog10K input dataset, we suggest a p-value of 4 × 10^-7 as a standard significance threshold that could be applied to any dog GWAS.