Fig. 2: Sensitivity and specificity analysis of methods for differential expression of gene signatures.

a Schematic for generating the simulated dataset and introducing controlled perturbations to model different levels of inter- and intra-patient variation (see Methods). b Boxplot depicting the specificity of BEANIE in identifying the differential expression of gene signatures (n = 50) in the simulated dataset (n = 1000 trials) benchmarked against six conventional methods (MWU-BH, MWU-BH-LOOCV, GLM, GLM-LOOCV, pseudobulk and pseudobulk-LOOCV). The box represents the interquartile range (IQR), the line within the box indicates the median, whiskers represent the smallest and largest data points within 1.5 times the IQR and outliers are represented as individual points. c–e Lineplots showing the sensitivity of BEANIE and conventional methods (MWU-BH, MWU-BH-LOOCV, GLM, GLM-LOOCV, pseudobulk and pseudobulk-LOOCV) in identifying the differential expression of gene signatures (n = 50) in the simulated dataset (n = 1000 trials) as a function of varying fraction of perturbed samples and magnitude of perturbation (1–3 std). Solid lines represent the mean sensitivity, and the shaded regions around each line indicate the 95% confidence intervals derived from the variability across 1000 trials for each perturbation condition. Perturbation of less than 50% samples in the group are shaded in gray.