摘要
目的对比分析Surv-MDR、Cox-MDR、Cox UM-MDR和KM-MDR四种用于识别生存分析中SNP位点交互作用的算法,为寻找影响患者生存时间的交互SNP位点时算法的选择提供理论指导。方法使用R 4.0.5完成四种算法以及模拟实验代码的编写,并使用ggplot2包绘制功效对比图;探索实际数据集上影响患者生存时间的交互SNP位点,并使用survminer包绘制相应的生存曲线。结果在不同最小等位基因频率、遗传度下四种方法的效能不同。Cox-MDR受协变量影响最大;Surv-MDR和KM-MDR功效相近且整体表现最好,且KM-MDR具有更好的稳定性;CoxUM-MDR功效强于Cox-MDR但受删失比例影响最大;在真实数据集上,Cox UM-MDR和Surv-MDR识别出了同一个有意义的一阶交互项,Cox-MDR和KM-MDR未能识别出有意义的交互项。结论当不存在协变量时,CoxUM-MDR、Surv-MDR和KM-MDR功效都明显高于Cox-MDR,不宜选用Cox-MDR;当存在协变量且删失比例不高时,Cox-MDR功效高于其他三种方法,宜选用Cox-MDR;但随着删失比例的增加,宜选用CoxUM-MDR或Surv-MDR。
Objective To compare and analyze Surv-MDR,Cox-MDR,Cox UM-MDR and KM-MDR four algorithms used to identify the interaction of SNP sites in survival analysis,in order to find the interactive SNP sites that affect the survival time of patients.Methods Use R4.0.5 to complete the compilation of the four algorithms and simulation experiment code,and use the ggplot2 package to draw the efficacy comparison chart;explore the interactive SNP sites that affect the survival time of the patient on the actual data set,and use the“survminer”package to draw the corresponding Survival curve.Results The four methods have different efficiencies under different minimum allele frequencies and heritability.Cox-MDR is most affected by covariates;Surv-MDR and KM-MDR have similar efficacy and have the best overall performance,and KM-MDR has better stability;Cox UM-MDR has more substantial efficacy than Cox-MDR but is more affected by censorship ratio.On the actual data set,Cox UM-MDR and Surv-MDR identified the same meaningful first-order interaction term,while Cox-MDR and KM-MDR failed to identify substantial interaction terms.Conclusion When there are no covariates,the efficacy of Cox UM-MDR,Surv-MDR and KM-MDR are significantly higher than that of Cox-MDR,so Cox-MDR should not be used;when there are covariates,and the censorship ratio is not high,The efficacy of Cox-MDR is higher than the other three methods,so Cox-MDR should be used;but as the censoring ratio increases,Cox UM-MDR or Surv-MDR should be used.
作者
田野
侯婧雯
王宗辉
郑睿
李若锦
刘艳
Tian Ye;Hou Jingwen;Wang Zonghui(School of Public Health,Harbin Medical University,150086,Harbin)
出处
《中国卫生统计》
CSCD
北大核心
2022年第5期663-668,共6页
Chinese Journal of Health Statistics
基金
黑龙江自然科学基金(LH2019H005)
国家自然科学基金(82173629)。