Independent of the abundance of a cell type, scCODA detected the effects only if a relatively large share of the samples was responsive to the condition. For abundant cell types (base count \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{0} = 100\, {{{\mbox{or}}}}\,1000$$\end{document}μ0=100or1000), a response share of about 40% was enough to achieve reliable detection, while for very rare cell types (base count \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mu }_{0}=1$$\end{document}μ0=1), more than half of the samples needed to show a response. If the share of responding samples was 70% or higher, scCODA reliably detected the effects (Supplementary Fig. 9).
