Comparison and definition of clusters
To distinguish tumor cells, we used inferring copy number variation (CNV)10,11 to compare our PPB data with normal fibroblasts whose expression profiles were obtained from the open access repository of the Gene expression omnibus (Gse 128169)12. To further analyze tumor heterogeneity, the obtained clusters were identified by cluster-specific marker genes. Gene Ontology enrichment analysis (GO analysis) was performed with the ClusterProfiler package (https://bioconductor.org/packages/clusterProfiler) for functional comparison of similar clusters. Moreover, secondary sparse-nonnegative matrix factorization (NMF) analysis was applied to PPB to improve identification of both broad and rare cell clusters13.
Pseudotime trajectory analysis with monocle
Using Monocle 2 (http://cole-trapnell-lab.github.io/monocle-release) with differential genes of each clusters, we reshaped the process of cellular changes over time by constructing trajectories between tumor cells. After exclusion of non-tumor cells, the remaining cells were reduced dimensionality using the DDRTree method, sequenced in pseudotime, and finally visualized.