reverseDecon: Reverse deconvolution

In Nanostring-Biostats/SpatialDecon: Deconvolution of mixed cells from spatial and/or bulk gene expression data

Reverse deconvolution

Description

Performs "reverse deconvolution", modelling each gene expression's ~ cell scores. Returns a matrix of "fitted" expression values, a matrix of residuals, a matrix of reverse decon coefficients for genes * cells.

Usage

reverseDecon(norm, beta, epsilon = NULL)

Arguments

norm	Matrix of normalized data, with genes in rows and observations in columns
beta	Matrix of cell abundance estimates, with cells in rows and observations in columns. Columns are aligned to "norm".
epsilon	All y and yhat values are thresholded up to this point when performing decon. Essentially says, "ignore variability in counts below this threshold."

Value

A list:

• coefs, a matrix of coefficients for genes * cells, where element i,j is interpreted as "every unit increase in cell score j is expected to increase expression of gene i by _".

• yhat, a matrix of fitted values, in the same dimension as norm

• resids, a matrix of log2-scale residuals from the reverse decon fit, in the same dimension as norm

• cors, a vector giving each gene's correlation between fitted and observed expression

• resid.sd, a vector of each gene's residual SD, a metric of how much variability genes have independend of cell mixing.

Examples

data(mini_geomx_dataset)
data(safeTME)
# estimate background:
mini_geomx_dataset$bg <- derive_GeoMx_background(
  norm = mini_geomx_dataset$normalized,
  probepool = rep(1, nrow(mini_geomx_dataset$normalized)),
  negnames = "NegProbe"
)
# run basic decon:
res0 <- spatialdecon(
  norm = mini_geomx_dataset$normalized,
  bg = mini_geomx_dataset$bg,
  X = safeTME
)
# run reverse decon:
rdecon <- reverseDecon(
  norm = mini_geomx_dataset$norm,
  beta = res0$beta
)

Nanostring-Biostats/SpatialDecon documentation built on Jan. 26, 2024, 8:20 p.m.