AI-PigNet: The AI of social interactions for next gen smart animal breeding
Lead Research Organisation:
University of Edinburgh
Department Name: College of Medicine & Vet Medicine
Abstract
Rapid population growth, climate change and increasing demand for animal products and higher animal welfare provide unprecedented challenges to livestock production. Automatic monitoring systems and cutting-edge Artificial Intelligence (AI) technologies, hold great promise for the quantification of animal-animal interactions to improve livestock management and selection. Next generation animal breeding can use data from animal interactions to select animals that are better suited for modern production conditions by enabling the selection of animals with improved behavioural patterns and higher social fitness. To accomplish this, it is necessary to overcome several challenges such as the ability to accurately monitor and quantify diverse types of social interactions in large number of individuals and to incorporate these measures into novel genetic prediction models that accurately predict the genetic merit for productivity and health of animals in different social environments.
The overall aim of this project is to develop novel AI routines that accurately capture social interactions of pigs from automated on farm monitoring systems and integrate these into genetic prediction models for smart animal breeding. To achieve this, we will for the first time unite three disciplines, i.e. machine/deep learning (ML/DL), social network analysis (SNA) and quantitative genetics, to achieve the following specific objectives:
AI to identify social interactions: Construct informative social interaction measures that describe the social structure and the role of each individual for different types of social interactions of putative welfare and productivity importance.
AI to establish associations between social interactions and key phenotypes. Establish how these networks and individual social interaction measures change over time and are affected by the genetic make-up of animals, and investigate their association with key productivity, health and welfare traits.
AI to improve predictions: Incorporate social interaction measures into smart breeding for animal productivity, health and welfare and evaluate improvement in prediction accuracies and genetic gain.
To accomplish this goal, we have established a multi-disciplinary team of experts in computational genetics, AI, SNA, animal behaviour and welfare, and breeding from leading academic institutes in the UK and USA, and PIC, the world's largest pig breeding company to translate knowledge to impact. We have access to a vast amount of unique, multi-dimensional data of thousands of commercial pigs (ranging from video data generated by the first successful automated system for reliably recording animal position and posture under commercial conditions, to production and health records, and genomic data) to establish and validate the computational pipeline to achieve the above project objectives.
This pump priming project focuses on pigs due to our strong background IP, research track record and available data. However, we anticipate that the methodology developed will be applicable to other production systems. The methods and knowledge generated will be beneficial for different stakeholders, e.g. scientists, breeders and farmers. Direct involvement of the world leading pig breeding company PIC, and close connections with the farming community will ensure a swift uptake of the research findings by the industry to generate impact at scale. This will help to improve the competitiveness of the UK livestock science and industry, with significant economic impact. Furthermore, exchange visits, and training courses and workshops organized during the life-time of this project allow skill development and knowledge exchange between the UK and USA partners, and establish professional networks for long-term collaborations in AI for smart animal breeding.
The overall aim of this project is to develop novel AI routines that accurately capture social interactions of pigs from automated on farm monitoring systems and integrate these into genetic prediction models for smart animal breeding. To achieve this, we will for the first time unite three disciplines, i.e. machine/deep learning (ML/DL), social network analysis (SNA) and quantitative genetics, to achieve the following specific objectives:
AI to identify social interactions: Construct informative social interaction measures that describe the social structure and the role of each individual for different types of social interactions of putative welfare and productivity importance.
AI to establish associations between social interactions and key phenotypes. Establish how these networks and individual social interaction measures change over time and are affected by the genetic make-up of animals, and investigate their association with key productivity, health and welfare traits.
AI to improve predictions: Incorporate social interaction measures into smart breeding for animal productivity, health and welfare and evaluate improvement in prediction accuracies and genetic gain.
To accomplish this goal, we have established a multi-disciplinary team of experts in computational genetics, AI, SNA, animal behaviour and welfare, and breeding from leading academic institutes in the UK and USA, and PIC, the world's largest pig breeding company to translate knowledge to impact. We have access to a vast amount of unique, multi-dimensional data of thousands of commercial pigs (ranging from video data generated by the first successful automated system for reliably recording animal position and posture under commercial conditions, to production and health records, and genomic data) to establish and validate the computational pipeline to achieve the above project objectives.
This pump priming project focuses on pigs due to our strong background IP, research track record and available data. However, we anticipate that the methodology developed will be applicable to other production systems. The methods and knowledge generated will be beneficial for different stakeholders, e.g. scientists, breeders and farmers. Direct involvement of the world leading pig breeding company PIC, and close connections with the farming community will ensure a swift uptake of the research findings by the industry to generate impact at scale. This will help to improve the competitiveness of the UK livestock science and industry, with significant economic impact. Furthermore, exchange visits, and training courses and workshops organized during the life-time of this project allow skill development and knowledge exchange between the UK and USA partners, and establish professional networks for long-term collaborations in AI for smart animal breeding.
Description | The recent advantage in AI-automated monitoring technologies offers digital phenotypes, at low-cost, that record the animals in real-time. Our proof-of-concept study addressed, for the first time, the hypothesis that applying social network analysis (SNA) on automated data could potentially facilitate the analysis of social structures of farm animals. Data was collected using automated recording systems that captured 2D camera images and videos of pigs in six pens (16-19 animals each) on PIC breeding company farms (USA). The system provided real-time data, including ear-tag readings, elapsed time, posture (standing, lying, sitting), and XY coordinates of the shoulder and rump for each pig. Weighted SNA was performed, based on the proximity of "standing" animals, for two 3-day periods: the early growing period (first month after mixing) and the later period (60 days post-mixing). Group level degree, betweenness, and closeness centralization showed a significant increase from the early growing period to the later one (p<0.02), highlighting the pigs´ social dynamics over time. Largest clique size remained unchanged (p=0.28), but the number of maximal cliques significantly decreased from the early to late growing period (p=0.007). Individual SNA traits were stable over these periods, except for closeness centrality and clustering coefficient which significantly increased (p<0.00001). This study demonstrates that combining AI-assisted monitoring technologies with SNA offers an efficient, real-time approach to gain novel insights into animal social interactions. This approach can optimize on-farm management practices, leading to improve animal performance, health, and welfare. |
Exploitation Route | We provided proof of concept that automated real-time farm animal data, coupled with AI and social network analysis can give novel insights into social interactions of farmed animals. Others may build on these methods to develop novel digital phenotypes to improve farm animal health and welfare. |
Sectors | Agriculture Food and Drink |
Description | AI-PigNet: Insights into the social interaction of pigs through automated data and social network analysis. |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Professional Practitioners |
Results and Impact | The talk demonstrated the use of automated data and AI to study social interactions in farm animals |
Year(s) Of Engagement Activity | 2024 |
Description | Diversity and Inclusion workshop |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Third sector organisations |
Results and Impact | Over 50 conference participants attended the Power Hour on Diversity and Inclusion in the Gordon Research Conference in Quantitative Genetics and Genomics. The group discussions provided a range of ideas and suggestions on how to improve diversity and inclusion in the academic workplace |
Year(s) Of Engagement Activity | 2025 |
URL | https://www.grc.org/quantitative-genetics-and-genomics-conference/2025/ |
Description | Diversity and Inclusion workshop |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Third sector organisations |
Results and Impact | Over 50 conference participants attended the Power Hour on Diversity and Inclusion in the Gordon Research Conference in Quantitative Genetics and Genomics. The group discussions provided a range of ideas and suggestions on how to improve diversity and inclusion in the academic workplace |
Year(s) Of Engagement Activity | 2025 |
URL | https://www.grc.org/quantitative-genetics-and-genomics-conference/2025/ |
Description | Invited talk in international conference |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Third sector organisations |
Results and Impact | The talk provided novel tools and results to assess the impact of host genetics in infectious disease transmission |
Year(s) Of Engagement Activity | 2025 |
URL | https://www.grc.org/quantitative-genetics-and-genomics-conference/2025/ |
Description | Invited talk in international conference |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Third sector organisations |
Results and Impact | The talk provided novel tools and results to assess the impact of host genetics in infectious disease transmission |
Year(s) Of Engagement Activity | 2025 |
URL | https://www.grc.org/quantitative-genetics-and-genomics-conference/2025/ |