Analysis of myeloid progenitor data

We followed the procedure described in http://cole-trapnell-lab.github.io/monocle-release/Paul_dataset_analysis_final.html to apply Monocle 2 to analyze the data generated by Paul et al. and generated a processed data set that was used to compare CellRouter to Monocle 2, DPT, Wishbone, and Waterfall. Briefly, we downloaded scripts to reproduce the analysis in Paul et al. from http://compgenomics.weizmann.ac.il/tanay/?page_id=649. UMI counts were downloaded from GEO accession number {"type":"entrez-geo","attrs":{"text":"GSE72857","term_id":"72857"}}GSE72857. A final data set of 3004 informative genes and 2699 cells (lymphoid cells were excluded) was used to compare these five algorithms. We applied Monocle 2 to this data set as described in the Monocle 2 tutorial and DPT as described in https://github.com/theislab/scAnalysisTutorial/blob/master/MARSseq_analysis_tutorial.ipynb. To run Wishbone, we used the top five principal components as input to perform the t-SNE analysis required by Wishbone and selected the diffusion components 1 and 3 as input. We used 150 waypoints. To run CellRouter in this data set, we selected k = 10 to build the kNN graph. Root cells were properly selected for each software (a cell in the common myeloid progenitors population), except for Waterfall, that does not have an option to define the starting point of the trajectory. In CellRouter, we selected the starting subpopulation 20 and a cell within this population is properly selected (as described in the section “Selecting sources and targets”).