We first performed comprehensive benchmarks on synthetic data across a wide range of scenarios (Methods—“Simulation”) that focused on scCODA’s primary application: the behavior of a single binary covariate that models the effect of a perturbation of interest in the respective scRNA-seq experiment. To detect statistically credible changes in cell-type compositions, we calculate the model inclusion probability for each covariate determined by the spike-and-slab prior (Methods—“Model description”). By using a direct posterior probability approach, scCODA automatically determines a cutoff on the posterior inclusion probability for credible effects that controls for the false discovery rate (FDR, Methods—“Spike-and-slab threshold determination”).
