Fig. 5

Piccolo enables identification of groups containing fewer cells as well as groups of cells with fewer differentially expressed genes. A Violin-box plots of the clustering metrics—ARI, AMI, Macro F1—obtained for simulated counts data generated for Scenario 1 based on parameters derived from 3 different real datasets (each row corresponds to one data set) and 3 different simulation tools (each column corresponds to one simulation tool). The colors correspond to the respective normalization methods used. For most simulated counts datasets, the highest values of the metrics were observed with Piccolo (red). B Heatmap showing the mean F1 scores per group aggregated over all the simulated counts data for Scenario 1 for each of the normalization methods. The tiles of the heatmap are colored according to the mean F1 scores with larger scores corresponding to brighter shades (yellow), and lower scores corresponding to darker shades (deep blue). Piccolo has the highest mean F1 scores for all 6 groups, with the improvement especially notable for Group 6 (group with fewest cells). C Same as panel B but for Scenario 2 (see Additional file 1: Fig. S22). Piccolo has the highest mean F1 scores for all 6 groups. Note, colors are scaled differently in panels B and C to bring out the differences between the mean F1 scores in the respective simulation scenarios