We check the scree plot to choose ten dimension as the intrinsic dimensions to reconstruct the developmental trajectory for the Paul dataset (cells used in Figure 1 of the original study9 (link)). Five branch points and six terminal lineages (monocytes, neutrophils or eosinophil, basophils, dendritic cells, megakaryocytes, and erythrocytes) are revealed. We ordered the cells using genes Paul et al. used to cluster their data rather than the genes from dpFeature, for the sake of consistency with their clusetering analysis. Similarly, we reconstruct Olsson datasets in four dimensions. The major bifurcation between the granulocyte and monocyte branch (GMP) as well as the intricate branch between GMP and megakaryocyte/erythrocyte (Ery/Meg) are revealed. Top 1, 000 genes from dpFeature based on WT cells are used in both of the WT and full datasets. The distribution (related to confusion matrix) of percentages of cells in each cluster from the original papers over each segment (state in Monocle 2) of the principal graph are calculated and visualized in the heatmap.
We applied BEAM analysis to identify genes significantly bifurcating between Ery/Meg and GMP branch on the Olsson wildtype dataset. We then calculate the instant log ratios (ILRs) of gene expression between Ery/Meg and GMP branch and find genes have mean ILR larger than 0.5. The ILRs are defined as:
So
is calculated as the log ratio of fitted value at interpolated pseudotime point
for the Ery/Meg lineage and that for the GMP lineage. Those genes are used to calculate the lineage score (simply calculated as average expression of those genes in each cell, same as stemness score below) for both of the Olsson and the Paul dataset which is used to color the cells in a tree plot transformed from the high dimensional principal graph (seeSupplementary Notes ). The same genes are used to create the multi-way heatmap for both of the Paul and Olsson dataset (see plot multiple_branches_heatmap function). Critical functional genes from this procedure are identified. Car1, Car2 (important erythroid functional genes for reversible hydration of carbon dioxide) as well as Elane, Prtn3 (important proteases hydrolyze proteins within specialized neutrophil lysosomes as well as proteins of the extracellular matrix) are randomly chosen as example for creating multi-lineage kinetic curves in both of the Olsson and Paul dataset (see plot_multiple_branches_pseudotime function).
In addition, pseudotime dependent genes for the Ery/Meg and GMP branch are identified in the Olsson wildtype dataset. All genes that always have lower expression from both lineages than the average in the progenitor cells are selected. Those genes are used to calculate the stemness score for both of the Olsson and the Paul dataset which is used to color the cells in the tree plot.
We applied BEAM analysis to identify genes significantly bifurcating between Ery/Meg and GMP branch on the Olsson wildtype dataset. We then calculate the instant log ratios (ILRs) of gene expression between Ery/Meg and GMP branch and find genes have mean ILR larger than 0.5. The ILRs are defined as:
So
is calculated as the log ratio of fitted value at interpolated pseudotime point
for the Ery/Meg lineage and that for the GMP lineage. Those genes are used to calculate the lineage score (simply calculated as average expression of those genes in each cell, same as stemness score below) for both of the Olsson and the Paul dataset which is used to color the cells in a tree plot transformed from the high dimensional principal graph (see
In addition, pseudotime dependent genes for the Ery/Meg and GMP branch are identified in the Olsson wildtype dataset. All genes that always have lower expression from both lineages than the average in the progenitor cells are selected. Those genes are used to calculate the stemness score for both of the Olsson and the Paul dataset which is used to color the cells in the tree plot.
