Novel statistical and bioinformatic methods to identify genetic factors involved in cognitive decline and rate of disease progression in pre-dementia stages of Alzheimer's disease

阿尔茨海默病(AD)等多因素疾病通常在临床诊断前几年就开始了。在临床前阶段调节疾病进展提供了推迟临床阶段开始的机会。因此，专注于确定疾病进展中涉及的因素和途径的研究预计将对多因素疾病的护理成本和预防政策产生重大影响。在大多数多因素疾病中，遗传因素是其归因风险的重要组成部分。因此，大多数调节疾病进展的病理生理途径很可能是由或包括遗传决定因素驱动的。不幸的是，关于疾病进展的基因研究目前还处于初级阶段。因此，这项提议的主要目标是开发创新和强大的统计方法，以分析遗传学在表型随时间推移中的作用。为此，我们将开发稳健的和计算上可行的线性混合模型(LMM)。现有的几种纵向数据遗传方法都是基于LMM的，因为这些统计模型具有几个优点，包括缺失数据的管理、重复测量的集成、固定和随机效应的结合。然而，它们在计算上很耗时，有时仅限于纵向表型的线性轨迹。为了解决这些问题，我们将使用计算速度更快的LMM作为条件LMM来开发改进的LMM。我们的目标是：a)开发纵向表型的方形轨迹的建模方法，b)开发一种将年龄特定的疾病发病风险视为随机效应的方法，以及c)开发一个模型来寻找驱动纵向表型的生物路径。为了在真实数据上测试这些模型，我们可以访问欧洲最大和全面的痴呆前期AD的纵向数据集，即轻度认知障碍(MCI)。对于所有的MCI病例，全基因组的基因数据已经在阿尔茨海默病联盟EADB中生成，包括9000个MCI样本。除了认知表型，MCI病例还有关于脑脊液的额外生物标记物数据和成像数据，这为我们的提案提供了一个独特的机会，可以将我们的研究扩展到疾病进展以外的假说。最后，我们将实现一种方法来生成甲基化调节的稳健遗传估计值。在此，甲基化被认为是遗传研究中确定的易感变异的功能相关性的分子介体。总之，我们的建议将为遗传学研究提供重要的工具来分析纵向表型。将这些方法应用于真实的MCI基因数据，将有助于识别调控AD疾病进展的新的遗传因素，以及它们驱动所观察到的遗传关联的潜在分子机制。

Multifactorial diseases, such as Alzheimer’s disease (AD), normally starts years before clinical diagnose is made. Modulating disease progression at preclinical stages offers the opportunity to delay the beginning of the clinical stage. Thus, research focused on identifying factors and pathways involved in disease progression is expected to have major impact on care cost and prevention policies of multifactorial diseases. In most multifactorial diseases, genetic factors account for an important part of their attributable risk. It is therefore likely that most of the pathophysiological pathways modulating disease progression will be driven by or include genetic determinants. Unfortunately, genetic research on disease progression is currently in its infancy. Consequently, the main objective of this proposal is to develop innovative and robust statistical methods to analyse the role of genetics on phenotypes progression over time. To this end, we will develop robust and computationally feasible linear mixed models (LMM). The few available genetic approaches on longitudinal data are based on LMM because these statistical models offer several advantages including management of missing data, integration of repeated measurements, combination of fixed and random effects. However, they are computationally time consuming and, sometime, limited only to linear trajectories of longitudinal phenotypes. To tackle these problems, we will develop improved LMMs using computationally faster LMMs, as the conditional LMM. We aim to: a) develop methods modelling square trajectories of longitudinal phenotypes, b) develop an approach considering age-specific risk on disease onset as a random effect, and c) develop a model to search for biological pathways driving longitudinal phenotypes. To test these models on real data, we have access to the European largest and comprehensive longitudinal dataset of pre-dementia AD, i.e. mild cognitive impairment (MCI). For all MCI cases, genome-wide genotype data has been generated within the Alzheimer’s disease consortium EADB, and comprises 9,000 samples of MCI. In addition to cognitive phenotypes, MCI cases have additional biomarker data on cerebrospinal fluid and imaging data providing our proposal with a unique opportunity to expand our research to hypotheses beyond disease progression. Finally, we will implement a method to generate robust genetic estimators of methylation regulation. Herein, methylation has been proposed as a molecular mediator for the functional relevance of susceptibility variants identified in genetic studies. In conclusion, our proposal will provide genetic research with important tools to analyse longitudinal phenotypes. Application of these methods to real MCI genetic data will lead to identification of novel genetic factors modulating disease progression in AD, as well as their potential molecular mechanism driving the observed genetic association.