Inference of differentiation time for single cell transcriptomes using cell population reference data

Abstract

Single-cell RNA sequencing (scRNA-seq) is a powerful method for dissecting intercellular heterogeneity during development. Conventional trajectory analysis provides only a pseudotime of development, and often discards cell-cycle events as confounding factors. Here using matched cell population RNA-seq (cpRNA-seq) as a reference, we developed an “iCpSc” package for integrative analysis of cpRNA-seq and scRNA-seq data. By generating a computational model for reference “biological differentiation time” using cell population data and applying it to single-cell data, we unbiasedly associated cell-cycle checkpoints to the internal molecular timer of single cells. Through inferring a network flow from cpRNA-seq to scRNA-seq data, we predicted a role of M phase in controlling the speed of neural differentiation of mouse embryonic stem cells, and validated it through gene knockout (KO) experiments. By linking temporally matched cpRNA-seq and scRNA-seq data, our approach provides an effective and unbiased approach for identifying developmental trajectory and timing-related regulatory events.

Single-cell RNA-sequencing reveals the existence of naive and primed pluripotency in pre-implantation rhesus monkey embryos

“…S1C). To test whether these samples were sufficient to capture all cell fates at each developmental stage, we compared the single-cell RNA-sequencing average profile to that of single-embryo RNA-sequencing at each matching stage (Wang et al 2017) using our iCpSc package (Sun et al 2017). The detection of major cell types at stages from EM through to HB apparently reached saturation level at our current coverage depth (Supplemental Fig.…”

Actual time-series of single-cell transcriptomics reveals circadian clocks as key regulators of cell differentiation in Arabidopsis

“…Actual time-series with interval scale contains temporal information, whereas ordinal scale-based pseudo time-series lacks temporal information. Thus, time-series analysis of scRNA-seq remains a di cult problem 25 and few studies have addressed circadian rhythms in single-cell transcriptomes. To overcome this limitation and to reconstruct actual time-series from single-cell transcriptome datasets, we developed PeakMatch method.…”

Regulatory network characterization in development: challenges and opportunities

“…However, these pseudotime methodologies encounter difficulties in accurately reconstructing branching trajectories in the event that more than one path derives from a single point or from multiple origins, as often happens in in vivo development (as shown in recent studies 29 , 30 ). The main assumption and an intrinsic limitation of the pseudotime reconstruction is that the gene expression similarity reveals the lineage relationship, which sometimes is not real, as there are discontinuous cell states, such as asymmetrical cell division 31 , not to mention that many transcriptome similarities, such as common cell cycle or metabolic states 32 , 33 , are irrelevant to lineage relationships and that factors other than the transcriptome, such as metabolism regulation and splicing regulation, are also vital for lineage differentiation 34 – 37 . Moreover, the performance and robustness of these pseudotime methods are difficult to benchmark because of large diversity in the outputted data structures and the lack of authentic experimental replicates.…”