Enhancing genomic selection for reproductive traits in turkeys through SNP prioritization using the fixation index

Genomic selection (GS) has revolutionized the field of animal breeding by enabling more accurate estimates of breeding values through a more precise assessment of the additive relationships between individuals. For lowly heritable traits, the increase in accuracy was in general modest at best. This is specifically the case for reproductive traits in turkeys which also suffer from a lack of normality of their phenotypic distributions. This study aimed to increase the accuracy of genomic selection for reproductive traits in turkeys by enhancing the quality of the genomic relationship matrix and by improving the normality of the phenotypic distributions. The fixation index (F_ST) was used to prioritize and select informative subsets of single nucleotide polymorphisms (SNPs) associated with genetic differentiation between extreme phenotypic groups. Concurrently, the Box-Cox transformation was applied to improve the normality of the reproductive trait phenotypic distributions. A total of 6,103, 5,564, and 5,548 records for egg production percent (PEP), egg fertility (FERT) and fertile eggs hatchability (HOF) traits, respectively were analyzed using univariate pedigree-based BLUP (BLUP), genomic BLUP (GBLUP), and single-step GBLUP (ssGBLUP) approaches. Incorporating F_ST-prioritized SNPs generally improved heritability estimates, with substantial gains observed for the lowly heritable fertility (FERT) trait when combined with the ssGBLUP approach. SNPs were prioritized using the 90th, 95th, and 99th percentiles of the F_ST distribution (F_ST90, F_ST95, and F_ST99). These thresholds retained approximately 4,801 SNPs (10 %), 2,400 SNPs (5 %), and 480 SNPs (1 %), respectively, out of the 48,008 markers available after quality control. For FERT, using all SNPs in GBLUP produced a baseline heritability of 0.15, whereas prioritizing the top 10 % F_ST markers increased the estimate to 0.29 (+93 % gain) for untransformed data. The ssGBLUP model also showed improvement, with heritability increasing from 0.12 (baseline) to 0.18 (+50 % gain) under the F_ST90 scenario. Furthermore, the Box-Cox transformation consistently resulted in higher heritability estimates across all methods compared to untransformed data, with ssGBLUP exhibiting the highest estimates after transformation. For FERT, the baseline GBLUP estimate increased from 0.15 (untransformed) to 0.19 after transformation (+27 % gain), while ssGBLUP increased from 0.12 to 0.15 (+25 % gain). When Box-Cox transformation was combined with F_ST90, heritability further increased to 0.33 for GBLUP (+74 % relative to its transformed baseline) and 0.21 for ssGBLUP (+40 % relative to its transformed baseline). The high re-ranking in the top 10 % after using SNP-prioritization suggests that marker filtering is likely to improve the accuracy of the selection decision through the correct identification of superior turkeys. These findings highlight the potential of F_ST-based SNP prioritization and data transformation techniques to enhance the accuracy of genomic selection for reproductive traits in turkeys.