SKAT与惩罚回归模型两阶段策略在基因组关联研究中的应用

发布时间：2019-03-31 19:28

【摘要】：目的本研究提出两阶段分析策略,将SKAT与惩罚回归模型联合应用,为遗传关联研究提供方法学选择的依据和指导。方法本研究使用遗传分析工作组18的数据,分别采用SKAT,LASSO,EN,cMCP,Gel以及两阶段统计分析策略(SKAT+EN,SKAT+LASSO,EN+SKAT,LASSO+SKAT)进行关联性分析。结果在基因水平,SKAT法的平均灵敏度与约登指数最高。除SKAT法外,其余统计策略的关联基因选出率均与对结局方差解释的比例和基因中包含SNPs的数目存在关联。在SNPs水平,EN法与EN+SKAT的灵敏度与约登指数最高。不同的统计策略均能把对结局效应贡献最大的真关联基因MAP4与SNPs筛选出来。结论SKAT和EN联合分析策略能够在数百万SNPs中筛选主要的疾病关联基因与SNPs,并在基因水平上统计推断,有着较高灵敏度,同时还能控制严重的假阳性错误,为遗传关联研究提供了一种较为高效的统计分析策略。
[Abstract]:Aim in this study, a two-stage analysis strategy was proposed, which combined SKAT with penalty regression model to provide the basis and guidance for the selection of methodology for genetic association research. Methods SKAT,LASSO,EN,cMCP,Gel and two-stage statistical analysis strategy (SKAT EN,SKAT LASSO,EN SKAT,LASSO SKAT) were used for correlation analysis using genetic analysis group 18 data. Results at the gene level, the average sensitivity and Joden index of SKAT method were the highest. Except for the SKAT method, the selection rate of the associated genes in other statistical strategies was related to the proportion of the variance explanation of the outcome and the number of SNPs contained in the genes. At the SNPs level, en and EN SKAT had the highest sensitivity and Joden index. MAP4 and SNPs, which contribute the most to the outcome effect, can be screened out by different statistical strategies. Conclusion SKAT and EN combined analysis strategy can screen major disease-associated genes and SNPs, in millions of SNPs and infer them statistically at the gene level. It has high sensitivity and can control serious false positive errors at the same time. It provides a more efficient statistical analysis strategy for genetic association research.
【作者单位】：广东药科大学公共卫生学院流行病与卫生统计学系;
【基金】：国家自然科学基金(81302493) 广东省科技厅社会发展基金(2014A020212307) 广东省自然科学基金(2016A030313809)
【分类号】：R394

【相似文献】