CellRank for directed single-cell fate mapping

Abstract

Computational trajectory inference enables the reconstruction of cell-state dynamics from single-cell RNA sequencing experiments. However, trajectory inference requires that the direction of a biological process is known, largely limiting its application to differentiating systems in normal development. Here, we present CellRank (https://cellrank.org) for mapping the fate of single cells in diverse scenarios, including perturbations such as regeneration or disease, for which direction is unknown. Our approach combines the robustness of trajectory inference with directional information from RNA velocity, derived from ratios of spliced to unspliced reads. CellRank takes into account both the gradual and stochastic nature of cellular fate decisions, as well as uncertainty in RNA velocity vectors. On data from pancreas development, we show that it automatically detects initial, intermediate and terminal populations, predicts fate potentials and visualizes continuous gene expression trends along individual lineages. CellRank also predicts a novel dedifferentiation trajectory during regeneration after lung injury, which we follow up experimentally by confirming the existence of previously unknown intermediate cell states.

Optimal transport analysis reveals trajectories in steady-state systems

“…Estimation of dynamics in the case where the underlying drift v does not arise from a potential gradient requires additional information to be available, such as potentially noisy or partial estimates of the velocity of cells [16, 1, 17]. Since at its core our method is based on solving a convex optimisation problem, additional information such as velocity estimates can be incorporated into our estimation procedure in a straightforward manner by modifying the cost matrix C .…”

Spatiotemporal heterogeneity of glioblastoma is dictated by microenvironmental interference

“…After performing the dynamical model, we estimated macro states which represent initial, terminal states as well as transient intermediate states using the CellRank package (v1.1.0, https://github.com/-theislab/cellrank) 26,45 . We constructed a transition matrix using the connectivity kernel which was analyzed by Generalized Perron Cluster Cluster Analysis (GPCCA) 46 after computing a Schur triangulation.…”

“…Single cell analysis was performed by the Seuratv4.0 package and SPATA 1.0 package. We used the Seurat wrapper for scVelo 45 to performe pseudotime analysis and Cell Rank 26 for cell fate estimation. After preprocessing of the data through Seurat, we imported the data into SPATA.…”

“…To comprehensively interrogate dynamic adaptations, we annotated the RNA-velocity of all tumor cells (InferCNV-analysis: gain Chr7 and loss Chr 10), Figure 2c-d , and realigned all cells according to their lineage origin, presented by the first 3 principal components, Figure 2e . Macro states including initial and terminal states were estimated by Markov chains based on annotated RNA velocity and transcriptomic similarity (CellRank 26 ). Our data indicated that reactive states likely arose from former phylogenetic lineage-differentiated origins, Figure 2f .…”

Crosstalk between lymphoid and myeloid cells orchestrates glioblastoma immunity through Interleukin 10 signaling

“…In order to determine dynamic gene expression changes, we extracted spliced and unspliced genes from the bam output created by cell ranger using the velocyto.py tool 40 . The resulting *.loom files were merged and transformed into .h5ad format for further processing by scVelo 41 and CellRank 42 . The pipeline is integrated into the SPATA toolbox 43 .…”