Phenomics

Temporal Stability of Novel Social Network Phenotypes in Pigs from AI-assisted Monitoring Systems

---

Abstract

Social interactions play a crucial role in shaping livestock performance, health, and welfare. Using manually decoded behavior data, we have shown that social network analysis (SNA) can identify novel heritable social phenotypes in pigs, quantifying the direct and indirect role of each individual in pen-level aggression and skin lesions. Importantly, these phenotypes showed favorable correlations with performance traits. Recent advances in AI-assisted monitoring technologies offer an efficient method to track farm animal behavior in real time, providing a practical alternative to labor-intensive and potentially biased manual approaches. In the current study, the data were derived from automated monitoring systems that provide 2D video recordings of pigs in six pens, each containing 16-19 purebred animals, at a PIC nucleus farm in the Midwest, USA. The automated system integrates tracking-by-detection, using the Hungarian algorithm with a specialized ear tag reader, to provide real-time pig identity and 2D XY coordinates, while posture (e.g., standing) is classified using a customized DeepCut pose estimation framework based on Convolutional Neural Networks. Previously, these AI routines and automated data were subjected to extensive quality control to ensure their reliability. Furthermore, we have demonstrated that applying SNA to this AI-assisted monitoring data, based on proximity between animals, provides an efficient, real-time approach to identify SNA-derived phenotypes quantifying both direct and indirect roles of each individual and revealing novel insights into the social structure. Building on this, the aim of this study was to assess the temporal stability of the SNA-derived phenotypes across developmental stages, a key step toward their integration into genomic prediction models. Low to moderate, but significant repeatability for several SNA phenotypes were observed, with higher consistency for degree (ICC = 0.38 ± 0.06***), eigenvector centrality (ICC = 0.37 ± 0.06***), and lower repeatability for betweenness (ICC = 0.14 ± 0.07***). These results indicate that certain pigs consistently occupy central positions in the social network, reflecting stable social roles within the group over time. Furthermore, pigs were ranked by average social network scores and classified into top 25%, middle 50%, and bottom 25% for each trait, allowing assessment of temporal stability and comparison between highly and lowly connected individuals. Spearman's rank correlations between days showed moderate temporal stability for degree, betweenness, and eigenvector centrality (ρ > 0.71, p < 0.01), indicating that pigs at the top and bottom of the social hierarchy tended to maintain their relative positions over time. In conclusion, although overall repeatability values were low to moderate, the results suggest a stable social hierarchy among high- and low-ranked individuals. These stable behavioral phenotypes provide reliable measures of social behavior that can be integrated into genomic prediction models, supporting smart breeding strategies and precision livestock management to optimize individual and group performance and welfare.

Keywords: 2026

How to Cite:

Agha, S., Psota, E., Turner, S., Lewis, C., Chen, C., Steibel, J., Zefreh, M. & Doeschl-Wilson, A., (2026) “Temporal Stability of Novel Social Network Phenotypes in Pigs from AI-assisted Monitoring Systems”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2285959. doi: https://doi.org/10.31274/wcgalp.23823

Rights: 1