Fig. 3From: Dictionary learning allows model-free pseudotime estimation of transcriptomic dataPercentage of correctly identified genes for the simulated datasets. The subfigures show evaluations for each method (method name in subtitle). The x-axis shows the number of genes with the simulated time pattern, |gsim|; The y-axis shows for each dataset the maximum percentage among the matrices with different method parameters. The dataset with one half of |gsim| increasingly ordered and the other half fluctuating is labelled “Incr_Fluct”, with “Incr_Fluct_1” being the half of increasing values and “Incr_Fluct_2” the half of fluctuating values. The high noise perturbation is labelled Noise+. DiL and ICA perform similarly well, whereas NMF does not reach as high percentages as the other methods. For PCA, high percentages are reached for most datasets, but they are mostly smaller than for DiL or ICA. Notably, for several datasets PCA-percentages decrease for increasing |gsim|, which presents an undesired behaviourBack to article page