Pseudotime estimation: deconfounding single cell time series.

Area of Science:

• Single-cell genomics
• Computational biology
• Statistical modeling

Background:

• Single-cell time-series data analysis is confounded by measurement noise, cell-to-cell variability, and asynchronous cell progression.
• These confounding factors are amplified in single-cell assays due to the lack of population averaging.

Purpose of the Study:

• To develop a robust probabilistic model for analyzing repeated cross-sectional time-series single-cell data.
• To provide a method for estimating and correcting for sources of variation in such datasets.

Main Methods:

• A principled probabilistic model with a Bayesian inference scheme was developed.
• The method was validated on public microarray, nCounter, and RNA-seq datasets across three organisms.

Main Results:

• The model accurately recovered capture times in an Arabidopsis dataset.
• It precisely estimated cell cycle peak times in human prostate cancer cells.
• The method successfully identified precocious cells in mouse dendritic cell signaling studies and compared favorably to Monocle.

Conclusions:

• The developed probabilistic model effectively addresses confounding factors in single-cell time-series data.
• Uncertainty in the temporal dimension is a significant confounder that requires accounting for in analyses.
• The DeLorean package is available on CRAN for public use.