Skip to main content

Table 1 Melissa robustly imputes CpG methylation states on sub-sampled ENCODE scRRBS and scWGBS synthetic data. Entries with italics denote the model with the highest performance in terms of AUC

From: Melissa: Bayesian clustering and imputation of single-cell methylomes

  Pseudo scRRBS Pseudo scWGBS
Model AUC 20% cov AUC 50% cov AUC 20% cov AUC 50% cov
Melissa 0.96 (7.3×10−4) 0.96 (6.8×10−4) 0.96 (6.3×10−4) 0.96 (6.6×10−4)
DeepCpG 0.94 (1.5×10−3) 0.94 (1.5×10−3) 0.96 (1.4×10−3) 0.96 (1.4×10−3)
BPRMeth 0.88 (2.2×10−3) 0.91 (2.5×10−3) 0.90 (1.9×10−3) 0.92 (1.5×10−3)
RF 0.79 (3.2×10−3) 0.87 (2.0×10−3) 0.83 (2.2×10−3) 0.89 (2.1×10−3)
Melissa rate 0.88 (1.8×10−3) 0.88 (1.3×10−3) 0.70 (2.2×10−3) 0.71 (2.5×10−3)
Rate 0.82 (2.6×10−3) 0.84 (2.5×10−3) 0.76 (4.2×10−3) 0.77 (3.0×10−3)
  1. Imputation performance in terms of AUC as we vary the proportion of covered CpGs used for training. Higher values correspond to better imputation performance. For each CpG coverage setting, a total of 10 random splits of the data to training and test sets was performed; shown are the mean AUC value together with two standard deviations of the estimate in parenthesis. Note that DeepCpG was trained once on two chromosomes; hence, the values do not change as we vary the CpG coverage