Fig. 3

AMULET detects multiplets with high precision and recall. a, b Recall for detecting a heterotypic and b homotypic artificial multiplets. AMULET consistently detected both heterotypic and homotypic multiplets with similar recall, while ArchR was only effective for predicting heterotypic multiplets for data with high heterogeneity. c Unsupervised clustering of multiplexed PBMC data identified 8 clusters corresponding to B cells, T cells, cytotoxic T, NK cells, monocytes, and dendritic cells. Multiplet detection via Vireo identified 244 multiplets among donors. d Multiplets identified by AMULET (red) and both AMULET and Vireo (black). 59% of AMULET multiplets overlapped with Vireo multiplets, showing high precision for detecting known multiplets among donors. e Multiplets identified by ArchR (red) and both ArchR and Vireo (black). Only 28% of multiplets detected by ArchR were also multiplets detected by Vireo. f Comparison of precision and recall for AMULET (FDR 10%, 5%, and 1%) and ArchR (default parameters). AMULET has higher precision than ArchR (0.57–0.61 vs. 0.28) for all FDR cutoffs with similar recall (0.17–0.20 vs. 0.20)