Fig. 4

2D visualisation of the coefficients for all evaluated methods for two components for two datasets. Shown are visualisations of the coefficients for the two datasets without subtypes (dataset E-MTAB-2565 in Subfigure a and dataset GSE112004 in Subfigure b). The respective dataset ID is given in the subtitle. The data points are coloured based on the experimental time points (colour encoded in legend). The parameters for t-SNE, respectively UMAP (perplexity=10,number_of_neighbours=10) are chosen to maximise the correlation among all values evaluated on average for both datasets. As the varied parameter for the other methods is the dimensionality, which is set to 2 here, for these methods no parameter search is conducted. Hence, for the show representations a parameter study is performed for t-SNE and UMAP only. For both datasets, for many methods, at least one component is representing well the dynamics of the data. It is striking that the low-dimensional representations from dynDLT represent the dynamics with low noise