Skip to contents

Prediction Using Pretrained NMF Model

Source: R/NMFpredict.R
NMFpredict.Rd

This function uses pretrained NMF models to recover cell states / spatial ecotypes. It takes a factorization matrix W representing a pretrained NMF model and a numeric gene expression matrix.

Usage

NMFpredict(
  W,
  testdat,
  scale = FALSE,
  ncell.per.run = 500,
  sum2one = TRUE,
  ncores = 1
)

Arguments

W

Matrix representing the factorization matrix W of a pretrained NMF model.

testdat

Numeric matrix containing the new data for which NMF scores are to be predicted.

scale

Logical indicating whether to scale the input data.

ncell.per.run

Integer specifying the maximum number of cells per NMF prediction run to avoid memory issues.

sum2one

Logical. If `TRUE`, normalizes the predicted scores so that they sum to 1 for each sample.

ncores

Integer specifying the number of CPU cores to use for parallel processing.

Value

A matrix representing the NMF prediction scores, with rows as samples/cells and columns as SE groups.

Examples


bulkdata <- fread("https://spatialecotyper.stanford.edu/inc/inc.public.vignettes.php?file=SKCM_RNASeqV2.geneExp.tsv",
                  sep = "\t", header = TRUE, data.table = FALSE)
rownames(bulkdata) = bulkdata[, 1]
bulkdata = as.matrix(bulkdata[, -1])
W <- readRDS(file.path(system.file("extdata", package = "SpatialEcoTyper"), "Bulk_SE_Recovery_W.rds"))

# Predict SE abundances in bulk tumors
preds <- NMFpredict(W = W, bulkdata, scale = TRUE)
head(preds[, 1:5])
#>                      NonSE        SE1        SE2         SE3        SE4
#> TCGA.3N.A9WB.06 0.10378493 0.08326065 0.25368726 0.094176154 0.07931696
#> TCGA.3N.A9WC.06 0.03110314 0.06302208 0.05111014 0.173416088 0.02882935
#> TCGA.3N.A9WD.06 0.46412765 0.15760965 0.07094936 0.007223942 0.05750158
#> TCGA.BF.A1PU.01 0.12913097 0.18470256 0.09710288 0.069662022 0.16008800
#> TCGA.BF.A1PV.01 0.10221472 0.05277487 0.05740510 0.159600536 0.12876903
#> TCGA.BF.A1PX.01 0.28552813 0.01531421 0.07115704 0.010845747 0.07130197