Data from: Using the generalized index of dissimilarity to detect gene-gene interactions in multi-class phenotypes

To find genetic association between complex diseases and phenotypic traits, one important procedure is conducting a joint analysis. Multifactor dimensionality reduction (MDR) is an efficient method of examining the interactions between genes in genetic association studies. It commonly assumes a dichotomous classification of the binary phenotypes. Its usual approach to determining the genomic association is to construct a confusion matrix to estimate a classification error, where a binary risk status is determined and assigned to each genotypic multifactor class. While multi-class phenotypes are commonly observed, the current MDR approach does not handle these phenotypes appropriately because the thresholds for the risk statuses may not be clear. In this study, we suggest a new method for estimating gene-gene interactions for multi-class phenotypes. Our approach adopts the index of dissimilarity (IDS) as an evaluation measure. This is analytically equivalent to the common association measure of balanced accuracy (BA) for the binary traits, while it is not required to determine the risk status for the estimation. Moreover, it is easily expandable to the generalized index of dissimilarity (GIDS), which has an explicit form that can handle any number of categories. The performance of the proposed method was compared with those of other approaches via simulation studies in which fifteen genetic models were generated with three class outcomes. A consistently better performance was observed using the proposed method. The effect of a varying number of categories was examined. The proposed method was also illustrated using real genome-wide association studies (GWAS) data from the Korean Association Resource (KARE) project.为探究复杂疾病与表型性状间的遗传关联，核心流程之一是开展联合分析。多因子降维法（Multifactor dimensionality reduction, MDR）是遗传关联研究中检测基因间互作的高效手段，其常规前提是将表型进行二分类划分。该方法判定基因组关联的常规思路是构建混淆矩阵以估计分类误差，过程中需为每一类多因子基因型类别指定二分类风险状态。但实际研究中多分类表型十分常见，现有MDR方法却无法妥善处理这类表型，原因在于不同风险状态间的阈值往往难以明确界定。本研究针对多分类表型，提出一种全新的基因-基因互作估测方法。该方法采用非相似性指数（index of dissimilarity, IDS）作为评估指标，其在分析层面等价于二分类性状中常用的平衡准确率（balanced accuracy, BA）关联测度，且无需预先指定风险状态即可完成估测。此外，该方法可便捷拓展至广义非相似性指数（generalized index of dissimilarity, GIDS），该指数具备明确的数学表达式，可适配任意数量的表型类别。本研究通过模拟实验对比了所提方法与其他同类方法的性能：实验共生成15种包含三类结局的遗传模型，结果表明所提方法的性能始终优于其他对比方法。研究同时分析了表型类别数量变化对方法性能的影响，并利用韩国关联资源（Korean Association Resource, KARE）项目的真实全基因组关联研究（genome-wide association studies, GWAS）数据对所提方法进行了实例验证。