Extremely unbalanced data refers to datasets with independent or dependent variables showing severe imbalances in proportions, which might lead to deviation of classical test statistics from theoretical distribution and difficulties in controlling type Ⅰ error. The increased availability of genome-wide resources from large population cohorts has highlighted the growing demand for efficient and accurate statistical methods for the process of extremely unbalanced data to improve the development of genetic statistical methods. This paper introduces two widely used correction methods in current genome-wide association study for extremely unbalanced data, i.e. Firth correction and saddle point approximation, describes their effectiveness in controlling type Ⅰ errors confirmed by simulation experiments, finally, and summarizes the commonly used software for extremely unbalanced genomic data to provide theoretical reference and suggestion for its application for the statistical analysis on extremely unbalanced data in future.
极端不平衡数据定义为自变量或因变量指标的取值呈现严重比例失衡的数据，在此情境下，参数模型假设检验的经典统计量明显偏离大样本下的理论分布，导致第一类错误膨胀。超大型人群队列全基因组资源的日益共享使得高效准确处理极端不平衡数据的统计需求日益突出，也推动了遗传统计方法的发展。本文介绍当前全基因组关联研究中2种常用处理极端不平衡数据的校正方法：Firth校正方法和鞍点近似方法，并通过模拟实验展示其可有效控制第一类错误，最后，简单介绍极端不平衡基因组学数据常用分析软件。本文为研究者对极端不平衡数据的统计分析提供理论参考和应用推荐。.