Markov Chain Construction and Probability Calculation

This function preprocesses data, constructs a Markov chain, and calculates transition probabilities based on pseudotime information.

Usage

markovProbability(
  milo,
  diffusionmap,
  terminal_state = NULL,
  root_cell,
  knn = 30L,
  diffusiontime = NULL,
  pseudotime_key = "pseudotime",
  scale_components = TRUE,
  num_waypoints = 500,
  n_eigs = NULL,
  verbose = TRUE,
  use_RANN = TRUE
)

Arguments

milo

A Milo or SingleCellExperiment object. This object should have pseudotime stored in colData, which will be used to calculate probabilities. If pseudotime is available in milo, it takes precedence over the value provided through the diffusiontime parameter.

diffusionmap

A DiffusionMap object corresponding to the milo object. Used for Markov chain construction.

terminal_state

Integer. The index of the terminal state in the Markov chain.

root_cell

Integer. The index of the root state in the Markov chain.

knn

Integer. The number of nearest neighbors for graph construction. Default is 30L.

diffusiontime

Numeric vector. If pseudotime is not stored in milo, this parameter can be used to provide pseudotime values to the function.

pseudotime_key

Character. The name of the column in colData that contains the inferred pseudotime.

scale_components

Logical. If TRUE, the components will be scaled before constructing the Markov chain. Default is FALSE.

num_waypoints

Integer. The number of waypoints to sample when constructing the Markov chain. Default is 500L.

n_eigs

integer, default is NULL. Number of eigen vectors to use.

If is not specified, the number of eigen vectors will be determined using the eigen gap.

verbose

Logical. If TRUE, print progress. Default is TRUE.

use_RANN

parameter to make user choose whether to use RANN to construct Markov chain, or keep using bluster

Value

milo or SinglCellExperiment object with pseudotime, probabilities in its colData

Examples

data(sce_vdj)
# downsample to first 2000 cells
sce_vdj <- sce_vdj[, 1:2000]
sce_vdj <- setupVdjPseudobulk(sce_vdj,
    already.productive = FALSE,
    allowed_chain_status = c("Single pair", "Extra pair")
)
#> Checking productivity from productive_abT_VDJ, productive_abT_VJ ...
#> 1461 of cells filtered
#> checking allowed chains...
#> 3 of cells filtered
#> VDJ data extraction begin:
#> extract_cols not specified, automatically generate colnames for extraction.
#> Extract main TCR from v_call_abT_VJ, j_call_abT_VJ, v_call_abT_VDJ, d_call_abT_VDJ, j_call_abT_VDJ ...
#> Complete.
#> Filtering cells from v_call_abT_VJ_main, j_call_abT_VJ_main, v_call_abT_VDJ_main, j_call_abT_VDJ_main ...
#> 11 of cells filtered
#> 525 of cells remain.
# Build Milo Object
set.seed(100)
milo_object <- miloR::Milo(sce_vdj)
milo_object <- miloR::buildGraph(milo_object,
    k = 50, d = 20,
    reduced.dim = "X_scvi"
)
#> Constructing kNN graph with k:50
milo_object <- miloR::makeNhoods(milo_object,
    reduced_dims = "X_scvi",
    d = 20
)
#> Checking valid object
#> Running refined sampling with reduced_dim

# Construct Pseudobulked VDJ Feature Space
pb.milo <- vdjPseudobulk(milo_object, col_to_take = "anno_lvl_2_final_clean")
pb.milo <- scater::runPCA(pb.milo, assay.type = "Feature_space")
#> Warning: more singular values/vectors requested than available
#> Warning: You're computing too large a percentage of total singular values, use a standard svd instead.

# Define root and branch tips
pca <- t(as.matrix(SingleCellExperiment::reducedDim(pb.milo, type = "PCA")))
branch.tips <- c(which.min(pca[, 2]), which.max(pca[, 2]))
names(branch.tips) <- c("CD8+T", "CD4+T")
root <- which.min(pca[, 1])

# Construct Diffusion Map
dm <- destiny::DiffusionMap(t(pca), n_pcs = 10, n_eigs = 5)
#> 'as(<dsCMatrix>, "dgTMatrix")' is deprecated.
#> Use 'as(as(., "generalMatrix"), "TsparseMatrix")' instead.
#> See help("Deprecated") and help("Matrix-deprecated").
dif.pse <- destiny::DPT(dm, tips = c(root, branch.tips), w_width = 0.1)

# Markov Chain Construction
pb.milo <- markovProbability(
    milo = pb.milo,
    diffusionmap = dm,
    diffusiontime = dif.pse[[paste0("DPT", root)]],
    terminal_state = branch.tips,
    root_cell = root,
    pseudotime_key = "pseudotime"
)
#> Sampling and flocking waypoints...
#> Markov chain construction...
#> Computing fundamental matrix and absorption probabilities...
#> Project probabilites from waypoints to each pseudobulk...
#> Complete.