Hossain Ahmed, Beyene Joseph
Department of Clinical Epidemiology and Biostatistics, McMaster University, 1280 Main Street West, Hamilton, Ontario L8S4K1, Canada.
BMC Proc. 2014 Jun 17;8(Suppl 1 Genetic Analysis Workshop 18Vanessa Olmo):S80. doi: 10.1186/1753-6561-8-S1-S80. eCollection 2014.
This article compares baseline, average, and longitudinal data analysis methods for identifying genetic variants in genome-wide association study using the Genetic Analysis Workshop 18 data. We apply methods that include (a) linear mixed models with baseline measures, (b) random intercept linear mixed models with mean measures outcome, and (c) random intercept linear mixed models with longitudinal measurements. In the linear mixed models, covariates are included as fixed effects, whereas relatedness among individuals is incorporated as the variance-covariance structure of the random effect for the individuals. The overall strategy of applying linear mixed models decorrelate the data is based on Aulchenko et al.'s GRAMMAR. By analyzing systolic and diastolic blood pressure, which are used separately as outcomes, we compare the 3 methods in identifying a known genetic variant that is associated with blood pressure from chromosome 3 and simulated phenotype data. We also analyze the real phenotype data to illustrate the methods. We conclude that the linear mixed model with longitudinal measurements of diastolic blood pressure is the most accurate at identifying the known single-nucleotide polymorphism among the methods, but linear mixed models with baseline measures perform best with systolic blood pressure as the outcome.
本文使用遗传分析研讨会18的数据，比较了全基因组关联研究中用于识别基因变异的基线、平均和纵向数据分析方法。我们应用的方法包括：（a）带有基线测量的线性混合模型；（b）带有平均测量结果的随机截距线性混合模型；（c）带有纵向测量的随机截距线性混合模型。在线性混合模型中，协变量作为固定效应纳入，而个体之间的亲缘关系作为个体随机效应的方差协方差结构纳入。应用线性混合模型去相关数据的总体策略基于奥尔琴科等人的GRAMMAR。通过分别将收缩压和舒张压作为结果进行分析，我们比较这三种方法在识别与3号染色体血压相关的已知基因变异以及模拟表型数据方面的情况。我们还分析了实际表型数据以说明这些方法。我们得出结论，在这些方法中，带有舒张压纵向测量的线性混合模型在识别已知单核苷酸多态性方面最为准确，但以收缩压为结果时，带有基线测量的线性混合模型表现最佳。
