Towards genetic improvement of social behaviours in livestock using large-scale sensor data: data simulation and genetic analysis

• Published in: Genetics selection evolution (Paris) Vol. 55; no. 1; p. 67
• Main Authors: Wang, Zhuoshi, Doekes, Harmen, Bijma, Piter
• Format: Journal Article
• Language: English
• Published: London BioMed Central Ltd 28.09.2023; BioMed Central; BMC
• Subjects: Accuracy; Agent-based models; Analysis; Animal behavior; Animal breeding; Animal welfare; Animals; Availability; Behavior; Breeding of animals; Computer vision; Data analysis; Data simulation; family structure; Genetic analysis; genetic correlation; Genetic diversity; Genetic improvement; Genetic research; genetic variation; Hogs; Injury analysis; Injury prevention; Life Sciences; Livestock; Machine vision; Mathematical models; Parameters; phenotype; Phenotyping; Poultry; Quantitative genetics; Selective breeding; Sensors; Simulation; Social behavior; Social factors; Social interactions; Social network analysis; Social networks; statistical models; swine; tail; variance
• ISSN: 1297-9686; 0999-193X; 1297-9686
• DOI: 10.1186/s12711-023-00840-z

Harmful social behaviours, such as injurious feather pecking in poultry and tail biting in swine, reduce animal welfare and production efficiency. While these behaviours are heritable, selective breeding is still limited due to a lack of individual phenotyping methods for large groups and proper genetic models. In the near future, large-scale longitudinal data on social behaviours will become available, e.g. through computer vision techniques, and appropriate genetic models will be needed to analyse such data. In this paper, we investigated prospects for genetic improvement of social traits recorded in large groups by (1) developing models to simulate and analyse large-scale longitudinal data on social behaviours, and (2) investigating required sample sizes to obtain reasonable accuracies of estimated genetic parameters and breeding values (EBV). Latent traits were defined as representing tendencies of individuals to be engaged in social interactions by distinguishing between performer and recipient effects. Animal movement was assumed random and without genetic variation, and performer and recipient interaction effects were assumed constant over time. Based on the literature, observed-scale heritabilities ([formula omitted]) of performer and recipient effects were both set to 0.05, 0.1, or 0.2, and the genetic correlation ([formula omitted]) between those effects was set to - 0.5, 0, or 0.5. Using agent-based modelling, we simulated ~ 200,000 interactions for 2000 animals (~ 1000 interactions per animal) with a half-sib family structure. Variance components and breeding values were estimated with a general linear mixed model. The estimated genetic parameters did not differ significantly from the true values. When all individuals and interactions were included in the analysis, the accuracy of EBV was 0.61, 0.70, and 0.76 for [formula omitted] = 0.05, 0.1, and 0.2, respectively (for [formula omitted]= 0). Including 2000 individuals each with only ~ 100 interactions, already yielded promising accuracies of 0.47, 0.60, and 0.71 for [formula omitted] = 0.05, 0.1, and 0.2, respectively (with [formula omitted] = 0). Similar results were found with [formula omitted] of - 0.5 or 0.5. We developed models to simulate and genetically analyse social behaviours for animals that are kept in large groups, anticipating the availability of large-scale longitudinal data in the near future. We obtained promising accuracies of EBV with ~ 100 interactions per individual, which would correspond to a few weeks of recording. Therefore, we conclude that animal breeding can be a promising strategy to improve social behaviours in livestock.