Methods for Epidemiology Studies

进行了详细的模拟研究，以评估从全基因组关联研究中建立多基因风险预测模型的替代策略。本研究对SNP选择的最佳策略、模型中SNP系数的估计以及连锁不平衡中SNP间相关性的处理进行了一些重要的观察。已经进行了调查，以测试在极端疾病风险下模型的充分性，在这种情况下，数据往往很少，但模型偏离更有可能具有重要的临床意义。该方法在一项大型乳腺癌病例对照研究中的应用表明，在logistic尺度上，常见snp以累加方式作用于该疾病的风险。基于对家系内独立染色体重采样的新方法，开发了检测受影响家系中观察到的突变模式的统计显著性的方法。将该方法应用于外显子组测序数据，为最近发现黑色素瘤的主要基因提供了统计支持。考虑到复杂相互作用的可能性，已经开发了用于检测一个区域内具有多个snp的疾病遗传关联的方法。开发了一种创新的计算方法，用于快速评估基因-基因相互作用的树结构模型，并结合基于排列的重采样方法来测试关联的统计显著性。采用随机效应或脆弱性模型来评估未观察到的危险因素对宫颈癌风险的异质性。该模型拟合了一项来自近百万妇女的队列研究的数据，以深入了解在大型HMO注册的妇女的风险分布。提出了一种新的流行病学研究加权逻辑分析方法，该方法涉及重新分层和人口扩张的增加抽样。通过仿真研究和人乳头瘤病毒血清学研究对该方法的性能进行了评价。开发了一种基于置换的重采样方法，利用代谢组学数据通过中间生物标志物检验暴露（例如吸烟）对疾病（例如肺癌）风险的影响的中介假设。一项研究调查了自适应LASSO算法的理论性质，该算法通常用于存在高维协变量的模型拟合，因为它能够控制底层变量选择过程中的错误发现率。

A detailed simulation study was conducted to evaluate alternate strategies for building polygenic risk prediction models from genome-wide association studies. The study made a number of important observations regarding optimal strategies for SNP selection, estimation of their coefficients in the model and handling of correlation among SNPs in linkage disequilibrium. Investigations have been conducted for testing the adequacy of models at extremes of disease-risk where data are often sparse but model departure is more likely with important clinical implications. Application of the method to analysis of a large case-control study of breast cancer indicate common SNPs act in additive fashion on the risk of the disease in a logistic scale. Methods have been developed for testing statistical significance of observed pattern of mutations in affected pedigrees based on novel method of resampling of independent chromosomes within pedigree. Application of the method to exome sequencing data has provided the statistical support behind the finding of a major gene for melanoma recently. Methods have been developed for testing genetic association of diseases with multiple SNPs within a region taking into account for possibility of complex interactions. An innovative computational approach was developed for rapid evaluation of tree-structure models for gene-gene interactions combined with permutation-based resampling methods for testing statistical significance of associations. Method has been developed to use random effect or frailty model to assess heterogeneity of risks of cervical cancer due to unobserved risk-factors. Model was fitted to data available from a cohort study of almost million women to obtain insight into risk distribution of women enrolled in a large HMO. New method was developed for weighting logistic analysis of epidemiologic studies that involves augmentation sampling with re-stratification and population expansion. Performance of the method was evaluated using simulation study and study of human papiloma virus serology. A permutation-based resampling method was developed for using metabolomic data for testing the hypothesis of mediation of the effect of an exposure (e.g smoking) on the risk of a disease (e.g lung cancer) through intermediate biomarkers. A study was conducted to investigate the theoretical properties of the adaptive LASSO algorithm, which is popularly used for model fitting in the presence of high-dimensional covariates, for its ability to control false discovery rate in the underlying variable selection procedure.