Bayesian Data Analysis for Sets of Data Sets: Towards Populations of Virtual Chimeras for In-Silico Trials

该项目旨在创新主要依赖于小规模传统临床试验的医疗方法的开发。这些试验可能成本高，成功率低，阻碍了应对日益增长的医疗保健需求所需的发展。受试者自然变异性的代表性不足会破坏试验的结果和可信度。该项目的目标是用多样化和大规模的虚拟受试者或代理人替代人类参与者，进行相对便宜的基于计算机的模拟，对应于结合统计建模和领域知识的项目研究人员的研究专业知识的计算机模拟试验。利兹大学及其相关研究机构是英国为人类健康和福祉开发计算机模拟试验的先驱之一，为分享知识和推进最先进的发展提供了良好的研究环境。成功的计算机模拟试验依赖于虚拟群体的质量，主要挑战是建立一个模型来生成模拟的合成数据（初始值）。该项目建议建立现实的生成人口模型，以有意义的方式捕捉自然和丰富的异质性/变异性，保留真实的人口的属性，重要的是，生成/预测高质量的数据条件下的描述性统计的虚拟主题。根据英国生物银行提供的数据，该项目的研究人员已经可以访问这些数据，该项目的重点是以数据驱动的方式构建此类模型，通过结合来自不同人群的多个数据集的信息，为开展计算机模拟试验提供必要的研究。关键的想法是，多个数据集可以相互补充，并加强它们之间潜在的弱共享信息，以建立集体和整体的统计人口模型，提供似是而非的和现实的替代真实的主题。该项目通过复制现有临床试验结果的能力得到了验证，这些临床试验具有真实的人群，为未来的计算机模拟试验奠定了基础。

This project proposes to innovate development of medical treatments that mainly relies on small-scale traditional clinical trials. The trials may have high cost and low success rate preventing the development required to face the growing demand on healthcare. Under-representation of natural variability of subjects undermines the outcome and credibility of the trials. The project goal is to substitute human participants with a diverse and large-scale population of virtual subjects or surrogates carrying out relatively inexpensive computer-based simulations corresponding to in-silico trials combining research expertise of the project investigators of statistical modelling and domain knowledge. The University of Leeds and its associated research institutes are among the UK forerunners on developing in-silico trials for human health and wellbeing providing an excellent research environment for sharing knowledge and advancing state-of-the-art developments. Successful in-silico trials rely on the quality of the virtual population and the main challenge is to build a model for generating synthetic data (initial values) for the simulations. The project proposes to build realistic generative population models that capture natural and rich heterogeneity/variability in a meaningful manner, preserve the properties of the real population and, importantly, generate/predict high-quality data conditioning on descriptive statistics of the virtual subjects. Based on the data provided by the UK Biobank, already accessible to the project investigators, this project focuses on building such models in a data-driven manner delivering necessary research for carrying out in-silico trials by combining information from multiple data sets over a diverse population. The key idea is that the multiple data sets may complement each other and enforce potentially weakly shared information between them to build collective and holistic statistical population models providing plausible and realistic surrogates of real subjects. The project is validated by the ability to replicate the outcome of existing clinical trials with real populations creating the basis for future in-silico trials.