Phenomics

Bayesian imputation of egg number using behavioral information derived from automated nest systems in local laying hens

---

Abstract

The shift from individual cages to group housing, barn, and free-range systems has improved animal welfare but has also made recording individual egg production more challenging. In these environments, hens share multiple nest boxes and move more freely, increasing the likelihood of ambiguous or missing hen-to-egg assignments. While some facilities rely on trap nests and human observers, such systems are cost- and labour-intensive. Automated nest systems offer a potential solution by using RFID technology to identify hen-to-egg associations and additionally provide behavioural information of the hens. However, their assignment is not perfect. Moreover, some hens prefer to lay floor eggs rather than using the automated nests, introducing additional uncertainty and reducing accuracy of phenotyping for breeding evaluations.To address these challenges, we developed a Bayesian imputation approach that uses behavioural information captured by an automated nest system, the "Weihenstephaner Muldennest" (Big Dutchman), and hand-counted number of eggs on farm as reference, to estimate individual egg numbers under uncertain identification. The model updates flat priors through likelihoods derived from multiple behavioural scenarios, including temporal laying patterns, nest visits and laying sequences, and nest preference. The derived posterior probabilities are subsequently used for a probabilistic egg assignment.We applied the method to data from three German local chicken breeds - Altsteirer (ALT, N = 74), Bielefelder (BIE, N = 56), and Ramelsloher (RAM, N = 87) - all kept in pens. Each pen was equipped with four RFID-based automated nest boxes. Egg laying was monitored from 63 to 73 weeks of age. Cross-validation based on trusted egg-to-hen matches showed imputation accuracies ranging from 0.72 to 0.95, with an average of 0.87, varying by pen and breed. In total, there were 4,821 eggs assigned, including 4,196 (87.0%) trusted matches, 447 (9.3%) floor eggs and 178 (3.7%) uncertain nest eggs. The imputed egg numbers were further used to evaluate their suitability for downstream genomic analyses.Heritability was estimated using Bayesian Gibbs Sampling in the BGLR package, based on genotypes from 52,903 SNPs after quality control. Median heritability estimates across breeds ranged from 0.23 to 0.28 with standard deviation of 0.1, and the 95% quantile spanning 0.11 to 0.53. These results suggest that, despite small sample sizes, the imputed phenotypes retain meaningful genetic variance.This work presents a computational approach that leverages behavioural observations from automated recording systems to improve the precision of egg production phenotyping in modern housing environments. This idea provides a proof of concept for integrating behavioural data into phenotype reconstruction when individual identification is uncertain.

Keywords: 2026

How to Cite:

Sun, C., Reimer, C., Goetzke, N., Weigend, A. & Geibel, J., (2026) “Bayesian imputation of egg number using behavioral information derived from automated nest systems in local laying hens”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2286641. doi: https://doi.org/10.31274/wcgalp.24074

Rights: 1