Minor allele frequency in genomic prediction for growth traits in Braunvieh cattle

The present research aimed to compare the effect of minor allele frequency (MAF) threshold during the quality control process of genotypes, using single nucleotide polymorphisms (SNPs) to predict genomic values of growth traits in Braunvieh cattle. The analysis was performed including 28,973 and 18,994 phenotypic records of birth and weaning weights, 12,835 single nucleotide polymorphisms and 300 animals. The traits were analyzed with the single-step genomic evaluation approach. The criteria for comparison of the minor allele frequency thresholds compared were: 1) ranking of genomic values: Spearman and Pearson correlation coefficient estimates were obtained for genomic values predicted with standard (0.05) and lower levels of minor allele frequency; 2) prediction ability of models: cross-validation with four replicates was performed, average of correlation between real and predicted phenotypes was obtained; and 3) regression coefficient estimates: the genomic values obtained with standard and lower levels of minor allele frequency were used as independent and dependent variables, respectively. The effect of minor allele frequency threshold on the ranking of genomic values was not important (correlation estimates greater than 0.99, <2.2x10(-16)). The prediction ability model was similar: around 0.7. the regression coefficient estimates were close to one (0.99, < 2x10(-16)). The results suggest that a minor allele frequency between 0.0 and 0.05, does not influence the genomic value prediction in small populations. However, these results should be regarded as preliminary and susceptible to changes if the study is repeated with a greater number of markers and animals.

Automated summary

The research compared the effect of different minor allele frequency thresholds during the quality control process of genotypes on the prediction of genomic values, and found that in small populations, a minor allele frequency between 0.0 and 0.05 does not significantly impact genomic value prediction. Additionally, the prediction ability model remained consistent throughout the study.