Skip to main content
Fig. 5 | BMC Bioinformatics

Fig. 5

From: Censcyt: censored covariates in differential abundance analysis in cytometry

Fig. 5

Multiple cluster simulation results testing for the association of the binary covariate. True positive rate (TPR) vs. False discovery rate (FDR) curves for censoring rates of 30%, 50% and 70% (rows) and samples sizes of 50, 100, 200, 400 (columns). Dots represent values at different significance thresholds (0.01, 0.05, 0.1; dashed lines). Filled dots have a lower FDR than the corresponding threshold, while empty dots have a FDR above.The x-axis is square root transformed. cc: complete case analysis, km: Kaplan–Meier imputation, kme: Kaplan–Meier imputation with an exponential tail of the survival function, mrl: mean residual life imputation (conditional multiple imputation), pmm: predictive mean matching (treating censored values as missing), rs : risk set imputation. GLMM uses the (unobserved) ground truth of the survival time and can be considered to be the maximum possible performance of the other methods. ncGLMM: same as GLMM but uses only the binary covariate for fitting and testing

Back to article page