Using machine learning to attribute the source of antimicrobial resistant campylobacteriosis in the United States

<strong>Background </strong>The construction of large genome datasets of infectious pathogens in combination with advanced bioinformatics methods has the potential to inform public health risk and targeted intervention strategies. In this study, we use the two most common <em>Campylobacter</em> species in human gastrointestinal infection as model organisms to test the use of machine learning methods for probabilistic assignment of genome sequenced cases of campylobacteriosis in the United States between 2009 and 2019 to potential source reservoirs. These enteric bacteria are ubiquitous in the gut of wild and domestic birds, agricultural mammals and commonly infect humans via consumption of contaminated food. Rising incidence and antimicrobial resistance (AMR) are major concerns and there is an urgent need to quantify the main routes to human infection. <strong>Methods</strong> As part of routine US national surveillance (2009 through 2019), 8,889 <em>Campylobacter </em>isolate genomes were sequenced from human infections and 15,924 from possible sources. Targeting genetic variation associated with adaptation to the most recent host, we used machine learning and probabilistic models to attribute the source of human infections and estimate the relative importance of different disease reservoirs. <strong>Findings</strong> Probabilistic attribution identified poultry as the primary infection source of human clinical isolates, responsible for an estimated 72% of cases. Most of the remaining clinical isolates were attributed to cattle (25%), with only a small contribution from wild bird (2%) and pork sources (2%). Fluoroquinolone and aminoglycoside resistant isolates increased overall and among infections attributed to chicken. <strong>Interpretation </strong>National-scale surveillance and quantification of the relative contribution of infection reservoirs can guide policy. Our study suggests that the greatest reductions in human campylobacteriosis in the US will come from interventions that focus on poultry, which may also reduce the spread of AMR strains.

**研究背景**
构建感染性病原体大型基因组数据集并结合先进生物信息学方法，可为公共卫生风险评估与精准干预策略制定提供关键支撑。本研究以引发人类胃肠道感染最为常见的两种弯曲杆菌属（Campylobacter）菌株作为模式生物，验证机器学习方法在2009至2019年美国弯曲杆菌病基因组测序病例溯源中的应用效果，将病例溯源至潜在的传染源储存宿主。这类肠道细菌广泛定植于野生鸟类、家养禽类与农业养殖哺乳动物的肠道内，通常经受污染食物传播感染人类。当前弯曲杆菌病发病率持续攀升，且抗菌药物耐药性（antimicrobial resistance, AMR）问题日益突出，亟需明确人类感染的主要传播途径。

**研究方法**
作为2009至2019年美国常规国家监测项目的组成部分，研究团队从人类感染病例中分离得到8889株弯曲杆菌并完成全基因组测序，同时从潜在传染源中获取15924株分离株的基因组序列。本研究针对与宿主适应性相关的遗传变异特征，采用机器学习与概率模型对人类感染病例进行溯源，并量化不同储存宿主的相对贡献权重。

**研究结果**
概率溯源分析显示，家禽是人类临床分离株的首要传染源，约占全部感染病例的72%。剩余多数临床分离株可归因于牛源（25%），野生鸟类与猪源的贡献占比仅分别为2%与2%。氟喹诺酮类与氨基糖苷类耐药的弯曲杆菌分离株整体占比上升，且在鸡源传播的感染病例中更为显著。

**研究阐释**
国家级规模的病原监测与传染源储存宿主相对贡献量化研究，可为公共卫生政策制定提供科学指导。本研究表明，美国人类弯曲杆菌病发病率的最大降幅将来自针对家禽的干预措施，该策略同时也可减少抗菌药物耐药菌株的传播扩散。