Skip to main content
Fig. 2 | Genome Biology

Fig. 2

From: Vireo: Bayesian demultiplexing of pooled single-cell RNA-seq data without genotype reference

Fig. 2

Evaluation of Vireo using synthetic mixtures of up to 16 scRNA-seq datasets, including a comparison to models that require genotype data of the pooled samples (Vireo-GT and Demuxlet). ac Assessment of Vireo performance on one representative simulated dataset consisting of 8 pooled samples with 1000 cells per sample, 8% doublet rate, and 4000 UMIs per cell on average. a Vireo model evidence (variational lower bound) when varying the pool size assumed in the model. b Adjusted Rand index (ARI) between the most likely inferred and the true singlet assignment, when varying the assignment confidence. The recommended cutoff (prob_max >0.9 for Vireo and Vireo-GT, and PRB.SNG1 >0.67 for Demuxlet) are highlighted as dot. c Receiver operating characteristic (ROC) curve for detecting doublets, when varying the assignment confidence. The recommended cutoff (prob_doublet >0.9 for Vireo and Vireo-GT, and 0.67 for Demuxlet) are highlighted with dots. di Systematic assessment of Vireo and alternative methods on simulated data using a range of parameter choices, using five replicate runs. df Area under ARI for singlet assignment considering alternative simulation settings, either varying the number of input samples d, the number of total cells in the dataset (doublet rate varied proportional; 1.2 to 12%; Methods) e, as well as the number of UMIs per cell f. g-i Area under the ROC for doublet detection, considering the same simulation parameters as in panels df. Parameters not varied in either of these experiments were set to their default values (indicated by the star symbol). Small dots in each experiment denote the five replicate simulation experiments, and the big dot denotes the median performance across replicates

Back to article page