Fig. 3

Melissa efficiently and accurately clusters cell sub-populations. a Clustering performance measured by ARI as we vary CpG coverage. Higher values correspond to better agreement between predicted and true cluster assignments. For each CpG coverage setting, a total of 10 random splits of the data to training and test sets was performed. Each colored circle corresponds to a different simulation. The plot shows also the LOESS curve for each method as we increase CpG coverage. b Clustering performance (ARI) for varying proportions of similar genomic regions between clusters. c Clustering performance (ARI) as we vary the total number of cells assayed. d Predicted number of clusters using two different prior settings: a broad and a strict prior as we vary cluster dissimilarity. Initial number of clusters was set to K=10. Melissa identifies the correct number of clusters in most parameter settings (K=4); notably when there is no dissimilarity across clusters (i.e., we have one global cell sub-population), Melissa prunes away all components and keeps only one cluster (K=1)