Predicting breeding values (BVs) for traits like feed intake or egg production, which are difficult or expensive to measure individually, is a challenge in animal breeding. Group records, where measurements are taken on groups of individuals, offer a potential solution. Previous research has shown that using full-sib group records can provide accurate BV estimations, comparable to individual records, particularly with small group sizes. Methods have been developed to handle multiple effects (litter and pen effects) in estimating variance components and predicting BVs from group records with varying sizes. However, individual recording remains challenging for some traits, while correlated traits (e.g., daily gain in relation to feed intake) are more easily measured. This suggests a potential improvement in accuracy by incorporating individual records of correlated traits into a bivariate analysis. Genomic prediction, using single-step genomic BLUP (ssGBLUP), offers an accurate approach to estimate BVs by integrating information from phenotypes, pedigree, and markers. Studies using individual records have demonstrated the significant improvement in EBV accuracy using genomic information. The potential benefit of genomic information for group records remains largely unexplored, although its ability to capture Mendelian sampling error suggests it could be even more important in such scenarios. This study aimed to quantify the improvement in the accuracy of genetic evaluation for group-recorded traits by using (1) varying proportions of genotyped animals (0%, 30%, and 100%) and (2) incorporating a correlated individually-recorded trait in a bivariate model.

Several studies have explored the use of group records for genetic evaluation. Nurgiartiningsih et al. (2004) and Simianer and Gjerde (1991) found negligible differences in variance components and consistent BV rankings between full-sib group records and individual records for fish and laying hens. Olson et al. (2006) proposed a model for predicting individual BVs using pooled group records, demonstrating its effectiveness, especially with small group sizes. Su et al. (2018) introduced a method to handle multiple fixed and random effects for BV prediction from group records with varying sizes, showing consistent variance component estimates compared to individual records, albeit with larger standard errors. The accuracy of EBVs from group records of 12 individuals reached up to 70% of the accuracy obtained from individual records. Genomic prediction, using single-step genomic BLUP (ssGBLUP) has emerged as a powerful tool to increase the accuracy of EBVs for complex traits by simultaneously using phenotypic, pedigree, and marker information. This approach has shown significant benefits for traits with individual records.

This study utilized simulated data mimicking a pig nucleus population generated by QMSim. A historical population was simulated for 300 generations, followed by the creation of a recent population with eight generations. Phenotypic and genomic data were generated for two traits (e.g., feed intake and daily gain), with a genetic correlation of 0.8. The simulated genome comprised 18 chromosomes, with markers and QTLs randomly distributed. Three group record scenarios were investigated: S12L2x3 (group size 12, individuals from four sublitters), S12Lran (group size up to 12, random assignment), and S24L2x3 (group size 24, individuals from eight sublitters). Variance components were estimated using the average information-restricted maximum likelihood (AIREML) approach. Breeding values were predicted using PBLUP (no genomic information), GBLUP (all animals genotyped), and ssGBLUP (30% of animals genotyped). Univariate and bivariate models were used for analyzing the traits. The accuracy of BV prediction was assessed using the correlation between predicted and true BVs. Bias was evaluated by the regression coefficient of true BV on predicted BV. Accuracies were calculated for all validation animals, genotyped animals (Group I), and non-genotyped animals (Group II). For the bivariate model, trait 1 (group-recorded) and trait 2 (individually-recorded) were modeled jointly. The residual covariance between the group measurement of trait 1 and the individual measurement of trait 2 was set to zero. All analyses were performed using DMU.

The study revealed that both genomic information and including a correlated trait significantly improved the accuracy of EBVs for group-recorded traits. Univariate analysis showed that the accuracy of EBVs increased considerably with increasing proportions of genotyped animals (0%, 30%, 100%). The increase was more pronounced for animals with genomic information (Group I) compared to those without (Group II). The accuracy of EBVs was significantly lower in scenario S12Lran (random animal assignment to pens) than in S12L2x3 (litters divided into sublitters), highlighting the importance of within-group relationships. Increasing the group size (Scenario S24L2x3) also led to a reduction in EBV accuracy. Compared to individual records, the accuracy of EBVs from group records ranged from 71-73% (Valid_R) and 58-66% (Valid_nR) in S12L2x3, decreasing in other scenarios. Bivariate analysis incorporating a correlated individually-recorded trait greatly improved the accuracy of variance component estimates for trait 1, compared to the univariate model. Although setting the residual covariance to zero in the bivariate model with group records for trait 1 led to a slight overestimation of pen covariance, the accuracy of EBVs for trait 1 was substantially enhanced, particularly when compared to the univariate approach with only group records. Regression coefficients close to 1 indicated unbiased BV prediction across all scenarios.

The findings demonstrate the value of using group records for traits challenging to measure individually. Incorporating genomic information consistently increased the accuracy of EBV prediction, surpassing the accuracy improvements from solely utilizing a correlated trait. The results underscore the importance of group structure. Random assignment of animals to pens decreased accuracy, emphasizing the benefit of maintaining within-group relationships, such as those present in siblings within a litter. The significant improvement in accuracy observed with the bivariate model highlights the complementary nature of group and individual records, allowing for more precise BV estimations. The study's findings are relevant to animal breeding programs aiming to improve the efficiency of genetic evaluation, particularly for complex traits that are expensive or difficult to measure directly on each individual. The successful use of a bivariate approach suggests that integrating information from easily measurable correlated traits can significantly enhance the accuracy of group-based assessments.

This study confirms the feasibility and benefits of using group records for genetic evaluation, especially when combined with genomic information and data from correlated traits. Smaller group sizes with close relationships between individuals within groups are recommended. The substantial increase in accuracy using a bivariate model incorporating an individually-recorded correlated trait makes this approach promising for efficient genetic evaluation. Future research could explore the optimal group size and structure for different traits and populations, investigate alternative methods for handling residual covariances in bivariate analysis with group records, and apply the developed methods to real-world datasets.

The study relied on simulated data, which might not perfectly reflect the complexities of real-world livestock populations. The assumption of zero residual covariance in the bivariate analysis might have affected the accuracy of the variance component estimation. Further investigations with real datasets are needed to validate the findings. The model used for the ssGBLUP assumes a certain level of linkage disequilibrium, which might not be representative for all livestock breeds or populations.