Improving the robustness of animals has become a priority in breeding due to climate change, new societal demands and the agroecological transition. Components of animal robustness can be extracted from the analysis of the adaptive response of an animal to disturbance using longitudinal data. Nonetheless, this response is a function of animal robustness as well as of disturbance characteristics (intensity and duration). To correctly assess an animal's robustness potential, it is therefore useful to know the characteristics of the disturbances it faces. The UpDown method, which detects and characterizes unknown disturbances at different levels of organization of the population (e.g., individual, pen, batch disturbances), has been proposed for this purpose. Furthermore, using the outputs of the method, it is possible to extract proxies of the robustness of animals. In this context, the objective of the study was to evaluate the performances of the UpDown method to detect and characterize disturbances and to quantify the robustness of animals in a genetic framework using different sets of simulation, and to apply this method to real pig longitudinal data recorded during the fattening period (body weight, cumulative feed intake and feeding rate). Based on the simulations, the specificity of the UpDown method was high (> 0.95). Its sensitivity increased with the level of organization exposed (from 0.23 to 0.32 for individual disturbances, from 0.45 to 0.59 for pen disturbances, and from 0.77 to 0.88 for batch disturbances). The UpDown method also showed a good ability to characterize detected disturbances. The average time interval between the estimated and true start date or duration of the disturbance was lower than three days. The correlation between the true and estimated intensity of the disturbance increased with the hierarchical level of organization (on average, 0.41, 0.78 and 0.83 for individual, pen and batch disturbance, respectively). The accuracy of the estimated breeding values of the proxies for robustness extracted from the analysis of individual trajectories over time were moderate (lower than 0.33). Applied to real data, the UpDown method detected different disturbances depending on the phenotype analyzed. The heritability of the proxies of robustness were low to moderate (ranging from 0.11 to 0.20).

中文翻译：

---

使用高通量表型数据和鲁棒性测量来识别和表征结构化群体中的未知干扰：在生长猪中的应用

由于气候变化、新的社会需求和农业生态转型，提高动物的健壮性已成为育种的首要任务。动物稳健性的组成部分可以通过使用纵向数据分析动物对干扰的适应性反应来提取。尽管如此，这种反应是动物稳健性以及干扰特征（强度和持续时间）的函数。因此，为了正确评估动物的稳健潜力，了解其面临的干扰的特征是有用的。为此目的，提出了 UpDown 方法，该方法检测并表征不同群体组织级别的未知干扰（例如，个体、围栏、批次干扰）。此外，使用该方法的输出，可以提取动物稳健性的代理。在此背景下，本研究的目的是评估 UpDown 方法检测和表征干扰的性能，并使用不同的模拟组量化遗传框架中动物的鲁棒性，并将该方法应用于真实的猪纵向数据育肥期间记录（体重、累计采食量和饲喂率）。基于模拟，UpDown方法的特异性很高(＞0.95)。其灵敏度随着组织暴露程度的增加而增加（个体干扰从 0.23 增加到 0.32，笔干扰从 0.45 增加到 0.59，批量干扰从 0.77 增加到 0.88）。UpDown 方法还表现出良好的表征检测到的干扰的能力。预计干扰开始日期或真实开始日期或持续时间之间的平均时间间隔小于三天。干扰的真实强度和估计强度之间的相关性随着组织层级的增加而增加（个体、围栏和批次干扰的平均值分别为 0.41、0.78 和 0.83）。从个体轨迹随时间的分析中提取的鲁棒性代理的估计育种值的准确性是中等的（低于 0.33）。应用于实际数据时，UpDown 方法根据分析的表型检测到不同的干扰。稳健性代理的遗传力为低至中等（范围为 0.11 至 0.20）。