Genomic selection accuracy and bias using imputed genotypes on growth, welfare and fitness traits in two Pekin duck lines

Background
The study explored the accuracy and biases of genomic selection in two commercial Pekin duck lines, focusing on their performance under real-world breeding practices. A dataset of 26 K duck records with phenotype and imputed genotype information (60 K chip) was analysed for growth, welfare, and primary feather length traits. Mixed linear models with relationship matrices from pedigree (BLUP) or markers (GBLUP) were used to estimate variance components and breeding values. We then assessed selection accuracies and biases to determine the most appropriate models.

Results
Results showed high imputation accuracies of 0.93 for line A and 0.92 for line D. Heritability estimates from pedigree were generally higher than those from genomic markers. For example, juvenile weight (JW) heritability ranged from 0.22 in line A and 0.25 in line D using markers, to 0.39 and 0.50, respectively, using the pedigree. Slaughter body weight (BW) had similar trends. Gait heritability was low (0.07) using markers in both lines, while breast muscle depth (BD) also had lower estimates (0.15–0.16).

For line A, genomic prediction accuracies were higher with the G-matrix, especially for BW (r2=0.68-0.70) and JW with r2 of 0.49. Estimates for gait and foot pad dermatitis (FPD) improved using the G-matrix at 0.58 vs. 0.24 and 0.68 vs. 0.44, respectively, compared to pedigree information. Similar improvements were found for line D, with BD estimates improving from 0.50 to 0.71 using the G-matrix.

Line A showed minimal bias (0.01-0.19) with the G-matrix compared to 0.02-0.41 with the A-matrix; the highest bias was for JW. Line D had lower biases with the G-matrix (0.02-0.17) than in line A with markers, while higher biases were observed using pedigree (0.01-0.37).

Conclusions
These findings indicate that all traits were heritable with higher prediction accuracies and lower biases using GBLUP compared to BLUP. This study demonstrates the effectiveness of GBLUP in improving prediction accuracy and reducing bias in selection traits of Pekin ducks, particularly for traits with low heritability.