MACHINE LEARNING TO FORECAST ZOONOTIC DISEASE EMERGENCE

描述（由申请人提供）：由于超过70%的新发传染病是由从动物传播给人类的寄生虫或病原体引起的（导致“人畜共患”感染），公共卫生的一个基本问题是确定导致人类人畜共患疾病的驱动因素。传染因子的跨物种传播取决于宿主的许多特征，它们的传染因子，以及确定疾病外部环境的环境因素。先前确定跨物种传播预测因子的研究受到限制，主要集中在有限空间尺度上的单一传染病（如狂犬病、莱姆病），部分原因是由于多变量、高维数据中常见的许多复杂相互作用、自相关性和抽样偏差，导致无法对众多宿主物种和传染因子进行大规模分析。提出的研究通过机器学习算法的创新应用来面对这些计算限制。具体而言，分析将解决全球卫生中三个突出且相互关联的问题：(1)哪些特征表明哺乳动物宿主物种容易成为人畜共患疾病的宿主？(2)传染因子的哪些特征预示其可能引起人畜共患感染？(3)全球人畜共患病暴发最重要的环境和人为预测因素是什么？分析将应用一系列监督、无监督和半监督机器学习算法，以新的全球规模数据库，其中包含三组哺乳动物宿主（灵长类动物、食肉动物和有蹄类动物）及其人畜共患病病原体的生物、生态、环境和人为数据。这项研究的一个长期目标是通过强调哺乳动物宿主的关键特征、传染因子以及描述近代史上人畜共患疾病暴发的环境和人为因素，从经验上制定关于人畜共患疾病的“经验法则”。最终，本文提出的研究将为预测未来人畜共患病的地理位置、感染源和动物宿主提供基础。

DESCRIPTION (provided by applicant): As over 70% of emerging infectious diseases are caused by parasites or pathogens transmitted from animals to humans (leading to 'zoonotic' infections), a fundamental issues for public health is identifying the drivers leading to zoonotic diseases in humans. Cross-species transmission of infectious agents depends on numerous traits of hosts, their infectious agents, and environmental factors defining the external context of disease. Previous studies identifying predictors of cross-species transmission have been limited by a focus on single infectious diseases (e.g., rabies, Lyme disease) at restricted spatial scales, in part because large-scale analyses spanning numerous host species and infectious agents are precluded by the many complex interactions, autocorrelations, and sampling biases common in multivariate, high-dimensional data. The proposed research confronts these computational limitations through the innovative application of machine learning algorithms. Specifically, analyses will address three outstanding and interrelated questions in global health: (1) What characteristics signal a predisposition of mammalian host species to be reservoirs of zoonotic disease?; (2) What traits among infectious agents predict their potential to cause zoonotic infection?; (3) What are the most important environmental and anthropogenic predictors of zoonotic outbreaks globally? Analyses will apply a series of supervised, unsupervised and semi-supervised machine learning algorithms to new, global-scale databases containing biological, ecological, environmental, and anthropogenic data for three groups of mammalian hosts (primates, carnivores, and ungulates) and their zoonotic infectious agents. A long-term goal of this research is to empirically develop "rules of thumb" about zoonotic diseases by highlighting the key traits of mammalian hosts, infectious agents, and the environmental and human factors describing zoonotic outbreaks in recent history. Ultimately, research proposed herein will provide a basis for predicting the geographic locations, infectious agents, and animal reservoirs from which future zoonoses will emerge. PUBLIC HEALTH RELEVANCE: This project proposes to investigate the factors driving zoonotic disease outbreaks and cross-species transmission from wild mammals into humans through the innovative application of machine learning algorithms to newly published data describing hundreds of infectious agents, their mammalian host species, human populations, and the global environment. Ultimately, this project aims to predict the locations and species from which future diseases will emerge, and is therefore directly relevant for the improvement of human health.