Computational modelling of disease progression and subtype discovery in Alzheimer's Disease

这项工作的动机是通过疾病进展模型的进展来更好地了解阿尔茨海默病。疾病进展模型旨在捕捉表征特定疾病的生物标志物变化的时间轨迹。这些模型提供了对疾病过程的重要见解，并可用于告知患者分层的分期系统。这对于神经退行性疾病尤其重要，如阿尔茨海默病，这种疾病具有极大的异质性，其风险因素知之甚少。先前的工作已经确定了具有共同疾病进展模式的患者亚组。然而，在混杂因素存在的情况下，子群发现是具有挑战性的，现有的建模方法不能很好地处理异常值。此外，现有的方法不能明确地模拟一个亚型内疾病轨迹的变化，这阻碍了它们进行个性化疾病预测的能力。该项目结合了伦敦大学学院POND小组开发的疾病进展建模技术，如亚型和阶段推断模型SuStaIn (Young等人，Nature Communications 2018)，以及统计无监督学习和异常值检测领域的新方法，以提高这些模型捕捉亚型景观复杂性的能力。我们将首先关注阿尔茨海默病的数据集，因为通过EuroPOND和E-DADS等项目，伦敦大学学院随时可以获得必要的数据集和临床专业知识。研究目标：将异常值检测方法纳入现有的疾病进展建模方法；使用来自统计无监督学习领域的先进技术来提高模型识别细粒度亚型和亚型内变异性的能力。将这些模型应用于大型临床数据集，以阐明与特定亚型相关的疾病进展和风险因素的新见解

The motivation for the work is to better understand Alzheimer's disease through advances in disease progression modelling. Disease progression models aim to capture the temporal trajectories of biomarker changes that characterise a particular disease. These models provide crucial insight into disease processes and can be used to inform staging systems for patient stratification. This is particularly important for neurodegenerative diseases, such as Alzheimer's disease, which are extremely heterogenous and whose risk factors are poorly understood.Previous work has identified subgroups of patients with common patterns of disease progression. However, subgroup discovery is challenging in the presence of confounders, and existing modelling approaches are not well equipped to deal with outliers. Furthermore, existing approaches do not explicitly model the variation of disease trajectories within a subtype, which hinders their ability to make personalised disease predictions.This project combines disease progression modelling techniques developed by the POND group at UCL, such as the Subtype and Stage Inference model, SuStaIn (Young et al, Nature Communications 2018), with new approaches from the fields of statistical unsupervised learning and outlier detection, to improve the ability of these models to capture the complexity of the subtype landscape. We will focus initially on Alzheimer's disease data sets, as the necessary access to data sets and clinical expertise are readily available at UCL through projects such as EuroPOND and E-DADS.Research aims:Incorporate outlier detection methods into existing disease progression modelling approaches, to concurrently estimate subtypes and identify individuals consistent with the modelUsed advanced techniques from the field of statistical unsupervised learning to improve the models' ability to identify fine-grained subtypes and within-subtype variabilityApply these models in large clinical datasets to elucidate new insights into disease progression and risk factors that are associated with particular subtypes