floodPseudotime: Calculate pseudotime by 'flooding'

In farrellja/URD: URD

Description

This calculates pseudotime by performing a probabilistic breadth-first search of the k-nearest neighbor graph that was used to generate the diffusion map calculated on the data. The results of this function should be passed to floodPseudotimeProcess to convert them into pseudotime.

Usage

floodPseudotime(
  object,
  root.cells,
  n = 20,
  minimum.cells.flooded = 2,
  tm.flood = NULL,
  verbose = F,
  verbose.freq = 10
)

Arguments

object	An URD object
root.cells	(Character vector) Names of cells that will be considered the root and assigned pseudotime 0.
n	(Numeric) Number of simulations to perform and average
minimum.cells.flooded	(Numeric) Stop pseudotime calculation when fewer cells are newly visited in a given iteration, and assign remaining unvisited cells pseudotime NA. This is designed to prevent a few poor data points from throwing off the entire pseudotime calculation.
tm.flood	(Sparse matrix or matrix) A sparse matrix of normalized transition probabilities (as returned by floodBuildTM. If unprovided (NULL), this will be calculated automatically. While it is very RAM intensive, this function runs about ten times faster if the matrix is provided as a standard (rather than sparse) matrix.)
verbose	(Logical) Report on progress?
verbose.freq	(Numeric) Give a report every this many iterations.

Details

A group of 'root' cells are pre-initialized as visited, and an iterative search of the graph is performed. The probability that a cell is newly visited is the cumulative transition probability from all cells that have previously been visited in the graph. The process continues until either the entire graph is visited, or fewer than minimum.cells.flooded new cells are visited in a given iteration.

On a computing cluster, this step can be parallelized by running a smaller number of simulations on many computers and combining the resulting data.frames with cbind.

Value

(data.frame) Rows are cells, columns are flood simulations, values are the iteration that newly visited each cell. This can be processed into pseudotime and stored in an URD object with floodPseudotimeProcess.

Examples

# Define the root cells as cells in HIGH stage
root.cells <- rownames(object@meta)[object@meta$STAGE == "ZFHIGH"]
# Do the simulation
flood.result <- floodPseudotime(object, root.cells = root.cells, n = 10, minimum.cells.flooded = 2, verbose = T)
# Convert to pseudotime in the URD object
object <- floodPseudotimeProcess(object, flood.result, floods.name = "pseudotime", max.frac.NA = 0.4, pseudotime.fun = mean, stability.div = 5)

# Use gobs of RAM to make it run faster
flood.result <- floodPseudotime(object, root.cells = root.cells, n = 10, minimum.cells.flooded = 2, verbose = T, tm.flood=as.matrix(floodBuildTM(object)))

farrellja/URD documentation built on June 17, 2020, 4:48 a.m.