This is our Friday rubric: every week a new Science Page from the Bob Morrison’s Swine Health Monitoring Project. The previous editions of the science page are available on our website.
This Science Page is a joint effort of researchers from around the globe! Saif Agha, Eric Psota, Simon P. Turner, Craig R. G. Lewis, Juan P Steibel, and Andrea Doeschl-Wilson share a summary of their paper looking at swine social structures.
Key points:
- This study demonstrates the potential of integrating social network analysis (SNA) with automated monitoring data and AI tools to construct informative social contact networks.
- SNA provided novel insights into the social structures that can be employed to simultaneously improve the performance, health, and welfare of farm animals.
Introduction:
Social interactions play a critical role in the welfare, health, and productivity of livestock. In pigs, group housing can lead to both beneficial social bonds and harmful behaviours like aggression. Understanding the structure of these interactions is essential for improving animal management. Traditional observation methods are time-consuming and subject to bias, making automated and AI-driven tools an attractive alternative. Social network analysis (SNA) offers a powerful framework for characterizing group dynamics and individual social roles. This study aimed to: (1) assess the feasibility of constructing social networks from data collected via automated monitoring systems, (2) analyse the social structure and its dynamics within pens, and (3) identify behavioural phenotypes relevant to social interactions in growing pigs.
Methods:
Over 1,000 pigs were monitored on commercial farms operated by PIC using an automated system that collected continuous 2D camera data. Each pig was identified via electronic ear tags, and the system recorded posture and 2D X-Y coordinates of the front and rear torso in real-time. Deep learning algorithms were used to extract movement and proximity data. Using the R packages spatsoc, asnipe, sna, and igraph, researchers identified when pigs were within 0.5 meters of each other while standing—used as a proxy for social contact. Six pens, each containing 16–19 pigs, were monitored for 12 hours per day (06:00–18:00), and proximity-based social networks were constructed. These networks were weighted by the duration of contact between each pair of pigs. Community detection and clique analysis were applied to characterize subgroup structures and identify tightly connected clusters. Centrality metrics (e.g., degree, betweenness, closeness, eigenvector centrality) were calculated at both the individual and group levels.
Results:
Social networks were successfully constructed based on the proximity between animals in each pen (Figure 1). For each identified group, SNA traits at the group-level were calculated e.g., group-degree, group-density, group-closeness, group-betweenness, group-eigenvector centrality and largest clique size within each identified group.
Discussion:
The results of this study have demonstrated that integrating AI-assisted automated monitoring technologies with SNA provides an efficient, real-time method for analysing animal social interactions, advancing beyond traditional manual observation techniques. By identifying community structures, analysing cliques, and assessing individual and group-level SNA traits, we gained novel insights into the social interactions of group-housed pigs under commercial farm conditions. These novel insights highlight the promise that SNA offers for optimizing management practices and improving animal performance, health, and welfare of farm animals.
Full Paper: https://doi.org/10.3390/ani15070996
References:
- Foister et al. (2018). https://doi.org/10.1371/journal.pone.0205122
- Agha et al. (2020). https://doi.org/10.3390/genes13040561
- Agha et al (2022). https://doi.org/10.3390/genes13040561
- Agha et al. (2022b). https://doi.org/10.3390/genes13091616