Supplementary figure S5.

Leave-out 2-cell-type rejection evaluation. “Leave-out 2-cell-type” experiment was performed on the same simulation dataset pair used in inter-dataset prediction. In this experiment, signatures of any combination of two cell types were removed in the reference matrix while keeping the query intact. The evaluation was repeated ten times for all ten different combinations. A. The x-axis lists methods with rejection options (e.g. allowing “unknown” labels), and the y-axis is the classification metric after excluding the leave-out groups. B. Boxplots showing the accuracy of methods in (A), when assigning “unlabeled” class to leave-out groups in the query.