Application of machine learning to genomic selection of dairy cattle through improved feed efficiency complex prediction

奶牛畜牧业的可持续性和盈利能力取决于生产能够满足消费者期望的产品、维护动物健康和福祉以及有效管理资源和投入之间的平衡相互作用。动物的饲料效率(FE)是一个关键的生产力指标，据估计，奶牛生产总成本的60%与饲料有关，其定义是使用较少的饲料资源生产相同数量的乳制品。鉴于目前国际上对温室气体排放、养分损失和水质的关注，促进改善FE复合体是以乳制品为基础的生产系统的重要重点。铁复合体受许多因素的影响，包括系谱、生理、环境和基因组变异，因此传统的基于线性方法的铁复合体定量方法的统计能力和准确性有限。另一种更全面的方法，如基于机器学习(ML)的方法，提供了整合不同变量并更有效地利用高维数据的潜力。ML技术可以突出影响FE复合体的最重要的因素，这些选定的变量可以用来开发改进的模型，可以更准确地预测FE复合体，并提供关于FE复合体潜在生物过程的新知识。该项目将寻求利用数据驱动的方法，通过结合系谱、生理、环境和基因组信息来使用ML，以提高FE复合体中的预测精度。这项工作将通过北爱尔兰农场动物生物库(NIFAB)获取过去20年在农业-食品和生物科学研究所(AFBI)整理的高精度数据库中捕获的高精度数据集。这些数据集包括系谱、每周采食量和饲料分析、管理和其他环境数据、每日生产记录和基因组信息，为更好地了解奶牛FE复合体提供了独特的机会，并通过识别关键基因和相关的生物途径提高了我们对FE复合体的基因组影响的了解。这些信息最终将使英国的奶农能够从牛群中选择效率最高的奶牛来繁殖替代品，并选择最高效的公牛来饲养这些奶牛。因此，饲料成本将降低，盈利能力将提高，从而减轻生产系统的低效和养分损失。作为个案学生，该项目将提供一个独特的机会，直接与美国联邦调查局合作，了解农业食品部门的需求和研究差距。

The sustainability and profitability of the dairy livestock sector is dependent on a balanced interplay between the production of produce that can meet consumer expectations, the maintenance of animal health and wellbeing, and the efficient management of resources and inputs. Feed efficiency (FE) complex in animals, as defined by the production of the same quantity of dairy product using fewer feeding resources, is a key productivity indicator with an estimated 60% of the total costs of dairy production being associated with feeding. Given current international concerns regarding greenhouse gas emissions, nutrient losses, and water quality, facilitating improvements to FE complex is an important focus of dairy-based production systems. FE complex is affected by many factors, including pedigree, physiological, environmental, and genomics variations, and therefore traditional methods based on linear approaches that quantify FE complex have limited statistical power and accuracy. Alternative more comprehensive approaches such as machine learning (ML)-based methods offer the potential to integrate heterogeneous variables and utilise high dimensional data more efficiently. ML techniques can highlight the most important factors influencing FE complex and these selected variables can be used to develop improved models that can more accurately predict FE complex and provide new knowledge on the biological processes underlying the FE complex. This project will seek to utilise a data-driven approach using ML by combining pedigree, physiological, environmental, and genomics information to provide improved prediction accuracy in the FE complex. High precision datasets, captured within a specially constructed database, collated at the Agri-Food and Bioscience Institute (AFBI) over the last 20 years will be made accessible to this work through the Northern Ireland Farm Animal Biobank (NIFAB). These datasets, containing pedigree, weekly feed intake and feed analysis, management and other environmental data, daily production records, and genomics information, present a unique opportunity to better understand the FE complex in dairy cattle, and improve our knowledge on genomic influence on FE complex by identifying key genes and associated biological pathways. This information will ultimately allow dairy farmers in the UK to select the most efficient cows within the herd from which to breed replacements, and the most food efficient sires to use on these cows. As a consequence, feed costs will be reduced and profitability improved, mitigating inefficiencies in production systems and nutrient losses. As a CASE studentship, this project will provide a unique opportunity to work directly with AFBI and learn about the needs and research gaps of the agri-food sector.