Yu Wenbao, Kwon Min-Seok, Park Taesung
Department of Statistic, Seoul National University, Seoul, South Korea.
Hum Hered. 2015;79(3-4):168-81. doi: 10.1159/000377723. Epub 2015 Jul 28.
To determine gene-gene interactions and missing heritability of complex diseases is a challenging topic in genome-wide association studies. The multifactor dimensionality reduction (MDR) method is one of the most commonly used methods for identifying gene-gene interactions with dichotomous phenotypes. For quantitative phenotypes, the generalized MDR or quantitative MDR (QMDR) methods have been proposed. These methods are known as univariate methods because they consider only one phenotype. To date, there are few methods for analyzing multiple phenotypes.
To address this problem, we propose a multivariate QMDR method (Multi-QMDR) for multivariate correlated phenotypes. We summarize the multivariate phenotypes into a univariate score by dimensional reduction analysis, and then classify the samples accordingly into high-risk and low-risk groups. We use different ways of summarizing mainly based on the principal components. Multi-QMDR is model-free and easy to implement.
Multi-QMDR is applied to lipid-related traits. The properties of Multi- QMDR were investigated through simulation studies. Empirical studies show that Multi-QMDR outperforms existing univariate and multivariate methods at identifying causal interactions.
The Multi-QMDR approach improves the performance of QMDR when multiple quantitative phenotypes are available.
在全基因组关联研究中，确定复杂疾病的基因-基因相互作用和缺失的遗传力是一个具有挑战性的课题。多因素降维（MDR）方法是识别二分表型基因-基因相互作用最常用的方法之一。对于定量表型，已经提出了广义MDR或定量MDR（QMDR）方法。这些方法被称为单变量方法，因为它们只考虑一种表型。迄今为止，分析多种表型的方法很少。
为了解决这个问题，我们提出了一种用于多变量相关表型的多变量QMDR方法（Multi-QMDR）。我们通过降维分析将多变量表型汇总为一个单变量分数，然后将样本相应地分为高风险组和低风险组。我们主要基于主成分使用不同的汇总方式。Multi-QMDR是无模型的且易于实现。
Multi-QMDR应用于脂质相关性状。通过模拟研究考察了Multi-QMDR的性质。实证研究表明，Multi-QMDR在识别因果相互作用方面优于现有的单变量和多变量方法。
当有多个定量表型时，Multi-QMDR方法提高了QMDR的性能。
