Data modelling for animal phenomics

影响家畜育种、生产和基因组学等多个动物科学领域的主要限制因素之一与收集、管理和有效模拟丰富、实时、高通量和高质量的表型数据有关。通过标准化和自动化表型收集协议，改进基础收集技术，并用环境、地理、基因、物理和生理数据1、2增强收集的表型信息，可以在缩小表型和基因之间的差距以及更好地理解表型和生产系统之间复杂的相互作用方面取得重大进展。研究人员和私营企业都在做出重大努力，开发系统和技术，在与特定表型特征和多个生产价值链的不同步骤相关的周期性和预先设定的时间范围内同步地收集、存储、建模和分析与动物有关的信息。鉴于动物表型的大量成分，如形态、发育、生化、生理和行为，本研究建议将重点放在与动物生长相关的表型研究上。我的研究项目的短期目标包括：(S1)使用负担得起的传感器和参考物，为猪、奶牛和肉牛开发高效、准确的半自动和自动图像采集和处理协议，(S2)使用数据挖掘、计算机视觉和机器/深度学习方法从图像中提取和分析形态测量信息(例如，身体尺寸、姿势、身体状况评分和运动模式)，(S3)调查各种形态测量、动物状态和动物发育/生长之间的相关性水平，以及它们在能够估计或预测生长的模型中的假定用途，以及(S4)基于包括具有已知尺寸的参考对象的数字图像，开发特定物种的回归和分类表型模型，用于估计由活体重和身体状况分数(以及其他相关的现有的和潜在的新的特征)表示的生长。长期目标包括：(L1)研究提高半自动和自动表型数据收集的准确性和可用性的方法，这些数据收集可以进一步与互补的牲畜信息相结合，以更高效、更可持续地生产动物，并改善农场的经济效益，(L2)收集表型结果并将其整合到模型中，这些模型可用于为加拿大行业开发预测性决策支持系统，以及(L3)招聘和培训家畜基因组学方面的高素质人员。

One of the major limiting factors impacting multiple areas of animal science such as livestock breeding, production and genomics is related to collecting, managing and effectively modeling abundant, real-time, high-throughput and high-quality phenotypic data. Significant advancements in closing the gap between phenotypes and genotypes and developing a better understanding of complex interactions between phenotypes and production systems can be achieved by standardizing and automating phenotype collection protocols, improving the underlining collection technologies and augmenting the collected phenotypic information with environmental, geographic, genotypic, physical and physiological data 1,2. Significant efforts are being made by both, researchers and private businesses, to develop systems and technologies that harvest, store, model and analyze animal-related information synchronously within periodical and pre-established time frames relevant to specific phenotypic features and various steps of multiple production value chains. Given the large number of components of an animal phenotype such as morphological, developmental, biochemical, physiological and behavioural, this research proposal will focus on the study of phenotypes related to animal growth. The short-term goals of my research project include: (S1) developing efficient and accurate semi-automatic and automatic image acquisition and processing protocols for pigs, dairy and beef cattle using affordable sensors and reference objects, (S2) extracting and analyzing morphometric information (e.g. body dimensions, posture, body condition score and motion patterns) from images using data mining, computer vision and machine/deep learning methods, (S3) investigating the level of correlation among various morphometric measurements, animal physiological states and animal development/growth and their putative use in models capable to estimate or predict growth, and (S4) developing species-specific regression and classification phenotypic models for estimation of growth represented by live body weights and body condition scores (and other relevant existing and potentially new traits) based on digital images that include reference objects with known dimensions. The long-term goals include: (L1) researching methods that increase the accuracy and usability of semi-automatic and automatic phenotypic data collection, which can be further integrated with complementary livestock information to produce animals more efficiently and more sustainably, and to improve economic outcomes on the farm, (L2) collecting and integrating phenotypic results into models that can be used to develop predictive decision support systems for the Canadian industry, and (L3) recruiting and training highly qualified personnel in livestock phenomics.