Development of a Bayesian network approach for analysing social science and biological data: the case of antibiotic resistance.

健康方面的大数据是一种日益增长的资源，也是一项复杂的跨学科挑战。大型异质数据集将不同领域的数据联系起来--如精神健康指标和微生物基因组学--的可用性，有可能解开个人生命过程中的无数元素和整个社会之间的复杂相互作用2，3。为了最大限度地发挥这种潜力，迫切需要先进的方法来分析这些复杂的数据。然而，对这种复杂的关联数据的分析带来了挑战，传统上应用的统计分析往往难以处理数据本身。数据可能是有噪声的，并且噪声是非随机的。问卷的回答往往是无系统的偏颇的，对某些方面的自我报告容易受到回忆能力差、可取性差、偏见或自我估计不准确的影响。数据也常常以分类的形式出现，而不是数字形式，这对某些形式的统计建模可能是有问题的。最后，遗漏数据在遗漏项目响应和纵向自然减损方面很常见。因此，我们需要更灵活和更先进的技术来解决这些问题。在这个博士和MRE中，我们将开发一种创新的跨学科解决方案，通过应用基于网络的方法，通常用于分析生物系统，以分析综合社会科学和生物医学数据。贝叶斯网络有潜力解决社会科学数据面临的许多技术建模问题，但尚未在社会科学中广泛应用。这是因为在它们成为真正有用的方法之前，有必要发展实践方法和具体的理论进步。在这里，我们将发展这些进展(更多细节如下)。这个项目将应用BN方法来应对一个紧迫而复杂的全球健康挑战：东非三个国家的抗菌素耐药性(ABR)。ABR水平上升对人类健康的潜在危害是巨大的。非洲是世界上最容易受到抗生素耐药性增加影响的地区之一，与世界其他地区相比，非洲的传染病负担最重。有人提出了一种“同一健康”的跨学科方法来理解这一问题，该方法强调社会和生物领域的相互联系，并跨越人类、动物和环境领域。然而，如果没有先进的工具，很难分析这样一个复杂的系统。可能有多个不同规模的驱动因素：个人行为、家庭、社区和医疗保健系统水平。然而，这些驱动因素在预测个人和社区层面的AMR水平方面的相对重要性和相互关系尚未得到很好的了解，而且很可能是高度上下文特定的研究问题本博士的研究问题是双重的，方法论和实质性的：1.方法论：发展BN结构学习，以实现跨社会科学和健康的跨学科大数据的整合，特别是解决：a.缺少数据。范畴变量c。有偏见的噪音2。实质性的重点是传染病的案例研究：使用HATUA的数据，调查东非尿路感染中抗菌素耐药性的驱动因素。

Big data' on health is a growing resource and complex interdisciplinary challenge. The availability of large heterogeneous datasets which link data across diverse domains - such as mental health indicators and microbiological genomics - has the potential to unravel complex interplay between myriad elements in the individual life course and society in general2,3. To maximise this potential, there is pressing need for advanced methods for analysing these complex data. However, the analysis of such complex linked data, presents challenges, and traditionally applied statistical analyses often struggle with the data itself. Data can be noisy, and the noise non-random. Questionnaires are often answered in asystematically biased way, and self-reporting of certain aspects are subject to poor recall, desirability, bias or inaccurate self-estimation. Data also often comes in categorical, rather than numerical forms, which can be problematic for some forms of statistical modelling. Finally, missing data, in terms of missing item responses and longitudinal attrition is common. Thus, we need more flexible and advanced techniques to tackle these issues. In this PhD and MRes we will develop an innovative interdisciplinary solution by applying a network-based approach, typically using in analysing biological systems, to analyse integrated social science and biomedical data. Bayesian networks have potential to address many technical modelling issues that social science data faces but have yet to be widely applied in social sciences. This is because it is necessary to develop both practical methodology and specific theoretical advances before they are a truly usable approach. Here, we will develop these advances (further details below).This project will apply BN approaches to a pressing and complex global health challenge: antibacterial resistance (ABR) in three countries in East Africa. The potential harm that increasing levels of ABR will have on human health is vast. One of the world regions most vulnerable to the increase in antibiotic resistance is Africa where, in comparison to other regions of the world, the burden of infectious diseases is highest. A 'one health', interdisciplinary approach to understanding the problem, which stresses the interconnectedness of social and biological domains and spans the human, animal and environmental spheres has been proposed.However such a complex system is difficult to analyse without advanced tools. There may be multiple drivers are varying scales: individual behavioural, household, community and healthcare system levels. The relative importance and inter-relationships between these drivers in predicting levels of AMR at individual and community level, however, not well understood, and moreover are likely to be highly context specificResearch Questions The research questions for this PhD are twofold, methodological and substantive:1. Methodological: To develop BN structure learning to enable integrating of interdisciplinary big data across social sciences and health, specifically addressing:a. Missing datab. Categorical variablesc. Biased noise2. Substantive focussed on a case study of communicable disease: using data from HATUA, to investigate the drivers of antibacterial resistance in urinary tract infections in East Africa.