algorithm2: SpatialDecon mixed cell deconvolution algorithm

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

Description

Runs the generic SpatialDecon decon workflow, including:

1. run deconvolution once

2. remove poorly-fit genes from first round of decon

3. re-run decon with cleaned-up gene set

4. compute p-values

Usage

algorithm2(
  Y,
  X,
  bg = 0,
  weights = NULL,
  resid_thresh = 3,
  lower_thresh = 0.5,
  align_genes = TRUE,
  maxit = 1000
)

Arguments

Y	p-length expression vector or p * N expression matrix - the actual (linear-scale) data
X	p * K Training matrix.
bg	Expected background counts. Provide a scalar to apply to all data points, or else a matrix/vector aligning with Y to provide more nuanced expected background.
weights	The same as the weights argument used by lm
resid_thresh	A scalar, sets a threshold on how extreme individual data points' values can be (in log2 units) before getting flagged as outliers and set to NA.
lower_thresh	A scalar. Before log2-scale residuals are calculated, both observed and fitted values get thresholded up to this value. Prevents log2-scale residuals from becoming extreme in points near zero.
align_genes	Logical. If TRUE, then Y, X, bg, and wts are row-aligned by shared genes.
maxit	Maximum number of iterations. Default 1000.

Value

a list:

• beta: matrix of cell abundance estimates, cells in rows and observations in columns

• sigmas: covariance matrices of each observation's beta estimates

• p: matrix of p-values for H0: beta == 0

• t: matrix of t-statistics for H0: beta == 0

• se: matrix of standard errors of beta values

• resids: a matrix of residuals from the model fit. (log2(pmax(y, lower_thresh)) - log2(pmax(xb, lower_thresh))).

SpatialDecon documentation built on Nov. 8, 2020, 6 p.m.