Development of high-dimensional data-adaptive causal inference methods to unravel the role of genetics in determining heart rhythm

在人类医学研究中收集的信息越来越多，正在改变生物医学研究的性质。除了年龄、性别和血压等更传统的流行病学信息外，在许多大型数据集中，如UK-Biobank，数千种遗传变异以及数百种其他详细的生物量（生物标志物、代谢物、血液电解质）也越来越多。用于分析基因组学数据的统计方法集中于将基因与特定生物结果相关联，或将生物标记与相同结果相关联。然而，通常，感兴趣的科学问题是特定的生物标志物是否介导遗传变异对结果或疾病的影响。因此，本研究的重点是提高我们对这些“OMICS”数据集的因果关系的理解。该项目旨在开发研究这种环境中的调解的方法，其动机是心脏遗传学研究，旨在确定与心律变化相关的遗传变异。通常不清楚这些基因是否通过它们对心脏离子通道的作用或通过其他途径影响心律。发现哪些电解质（例如钠，钾）位于从基因到心脏病的重要途径上，可以帮助靶向新药，以及现有药物的新适应症。大量的变量使得对这种“网络”（几个基因、几种电解质和心律之间的链）的推断更具挑战性。需要复杂的技术来解释这些变量之间的混杂。新的机器学习方法将被纳入因果推理方法中，以克服处理高维数据集的问题。虽然该项目将重点关注心脏遗传暴露，但开发的方法将广泛适用，使研究人员能够充分利用基因组学和其他大组学数据集。

The increased availability of information collected in human medical studies is transforming the nature of biomedical research. Thousands of genetic variants as well as hundreds of other detailed biological quantities (biomarkers, metabolites, blood electrolytes) are increasingly available in addition to more traditional epidemiological information such as age, gender and blood pressure in many large datasets such as UK-Biobank. Statistical methods for analysing genomics data have focused on associating genes with particular biological outcomes, or on associating biological markers with the same outcomes. Often, however, the scientific question of interest is whether a particular biomarker mediates the effect of the genetic variant on the outcome or disease. The focus of this research is therefore to improve our understanding of the causal relationships in these "OMICS" data-sets. This project aims to develop methods to study mediation in such settings, motivated by cardio-genetics studies, seeking to identify genetic variations associated with changes in heart rhythm. It is often unclear if these genes affect the heart rhythms through their effects on the heart ion channels or through other pathways. Finding which electrolytes (e.g. sodium, potassium) lie on important pathways from genes to heart disease, could help target new drugs, as well as new indications for existing drugs. Inference of such "networks" (chains between several genes, several electrolytes, and heart rhythms) is made more challenging by the large number of variables. Sophisticated techniques are required to account for confounding among these variables. Novel machine learning methods will be incorporated into the causal inference methodology, in order to overcome the problems of working with high-dimensional data sets. Although this project will focus on cardio-genetic exposures, the methods developed will be widely applicable, allowing researchers to use genomics and other big omics data sets to their full potential.