Next, we applied scCODA to a number of scRNA-seq data examples1,3,4,6,19 (Fig. 3, Supplementary Figs. 5–9, and Supplementary Data 1). To confirm scCODA’s applicability on real data with known ground truth, we first considered a recent study of age-related changes in peripheral blood mononuclear cells (PBMCs)3, where cellular characteristics of supercentenarians (n = 7) were compared against the ones of younger controls (n = 5; Fig. 3a). The original study used a Wilcoxon rank-sum test and reported a significant decrease of B cells in supercentenarians, which is known from literature20. Moreover, the result was validated by FACS measurements. scCODA also identified B-cell populations as the sole affected cell type using CD16 + monocytes as a reference at an FDR level of 0.2. This suggests that scRNA-seq data indeed comprise enough information to study compositional changes, and that scCODA can correctly identify the experimentally validated age-related decrease of B cells even in low-sample regimes.
