Fig. 3: Differential expression of gene signatures between anti-PD-1 naive vs. anti-PD-1 treated ER+ breast cancers.

a Schematic showing the comparison groups of anti-PD-1 naive vs. anti-PD-1 treated ER+ breast cancers. b UpsetPlot depicting the overlap between differentially expressed gene signatures identified by BEANIE, MWU-BH, GLM, pseudobulk and their LOOCV variants, across tumor states (c) Barplot depicting examples of biologically relevant gene signatures identified as differentially expressed by BEANIE for each of the tumor states. d Sample Dropout Plot depicting the samples resulting in non-robustness of various statistically significant gene signatures across tumor states using BEANIE. The heatmap highlights the samples with a Fold Rejection Ratio (FRR) less than threshold (0.9), and the horizontal bar plot depicts the number of statistically significant but non-robust gene signatures (‘dropout’ signatures) when the specific sample is excluded. e Histogram illustrating the effect of sample exclusion on the robustness of evaluated gene signatures during differential expression testing with BEANIE. During the testing of differential expression of HALLMARK_INTERFERON_ALPHA_RESPONSE, when BRCA_20 is present, the distribution of subsamples is less than the 5th percentile of the background distribution (i.e., statistically significant). However, when BRCA_20 is excluded, the distribution shifts such that the median is not in the top 5th percentile of the background distribution. Therefore, the empirical p value is greater than 0.05 for the fold when BRCA_20 is excluded.