Deep Poisson process pathogen phylodynamics to accelerate understanding in disease transmission

破译人类遗传密码的发现是一项里程碑式的科学成就，导致了个性化医学、基因疗法或现代疫苗的发展。今天，主要生物体的遗传密码以及危害人类健康的病毒或细菌病原体被以低成本和工业规模识别(或测序)，目的包括重建传染病在人类群体中的传播方式，以及如何阻止传播。病原体变异的遗传关系提供了关于谁感染了世卫组织的客观数据，否则很难获得这些信息。对这些数据的分析所依据的数学和统计理论被称为“系统动力学”。这一理论使重建和量化耐药病原体如何在全球范围内、在社区或医院病房传播，或如何出现新的新冠肺炎变体并相互取代成为可能。这个EPSRC项目旨在开发一种新的统计系统动力学理论，基于深度泊松点过程，比现有方法更灵活，计算速度更快。我们的初步发现表明，这种方法有可能解锁对年龄、行为特征、地点、流动模式或作为病原体传播来源的人口群体的其他特征的重要问题的分析，这些问题迄今非常具有挑战性或不可能解决。我们将发展统计理论，并为灵活和可重复的分析提供开放获取和计算可伸缩的代码。该项目将受益于与机器学习和全球健康网络(开发深度非参数方法)、国际盘古病毒-艾滋病毒联盟(获取过去10年收集的大规模、丰富和具有全球重要性的数据)、英国健康安全局(旨在在英国使用我们的方法)和牛津纳米孔(过渡性行业影响)的密切联系。

The discovery of deciphering the human genetic code has been a landmark scientific achievement, leading to the development of personalized medicine, gene therapies, or modern vaccines. Today, the genetic codes of major organisms, and viral or bacterial pathogens that compromise human health are identified (or sequenced) at low cost and at industrial scale, including for the purpose of reconstructing how infectious diseases spread in human populations, and how to stop spread. The genetic relationships of pathogen variants provide objective data about who infected who, information that is otherwise hard to obtain. The mathematical and statistical theory that underlies the analysis of these data is called 'phylodynamics'. This theory has made possible to reconstruct and quantify how anti-microbial resistant pathogens have spread worldwide, in communities, or in hospital wards, or how novel COVID-19 variants emerge and replace each other. This EPSRC project aims to develop a novel class of statistical phylodynamic theory, grounded in deep Poisson point processes, that are substantially more flexible and computationally faster than existing methods. Our preliminary findings indicate this approach has the potential to unlock the analysis of important questions about the age, behavioural characteristics, locations, mobility patterns or other characteristics of population groups that are the sources of pathogenic spread, and which to date are very challenging or impossible to address. We will develop the statistical theory and provide open-access and computationally scalable code for flexible and reproducible analyses. This project will benefit from close ties to the Machine Learning & Global Health network (development of deep non-parametric methods), the international PANGEA-HIV consortium (access to large-scale, rich and globally important data collected over the past 10 years), the UK Health Security Agency (aiming to use our methods in the UK) and to Oxford Nanopore (transitional industry impact).