STdiff.RdTests for differentially expressed genes using linear models with or without spatial covariance structures
STdiff(
x = NULL,
samples = NULL,
annot = NULL,
w = NULL,
k = NULL,
deepSplit = NULL,
topgenes = 5000,
pval_thr = 0.05,
pval_adj = "fdr",
test_type = "mm",
sp_topgenes = 0.2,
clusters = NULL,
pairwise = F,
verbose = 1L,
cores = NULL
)an STlist
an integer indicating the spatial samples to be included in the DE tests.
Numbers follow the order in names(x@counts). Sample names are also allowed.
If NULL, performs tests on all samples
a column name in x@spatial_meta containing the groups/clusters to
be tested. Required if k and w are empty.
the spatial weight used in STclust. Required if annot is empty.
the k value used in STclust, or dtc for dynamicTreeCut clusters. Required if annot is empty.
the deepSplit value if used in STclust. Required if k='dtc'.
an integer indicating the top variable genes to select from each sample based on variance (default=5000). If NULL, all genes are selected.
cut-off of adjusted p-values to define differentially expressed genes from
non-spatial linear models. A proportion of genes (sp_topgenes) under this cut-off
will be applied the spatial models. Default=0.05
Method to adjust p-values. Defaults to FDR. Other options as
available from p.adjust
Proportion of differentially expressed genes from non-spatial
linear models (and controlled by pval_thr) to use in differential gene expression
analysis with spatial linear models. If 0 (zero), no spatial models are fit. Default=0.2
cluster name(s) to test DE genes, as opposed to all clusters.
whether or not to carry tests on a pairwise manner. The default is
pairwise=F, meaning that DE genes are tested by comparing each cluster to the
rest of the pooled cell/spots.
output progress indicators. If verbose=0, no text is shown in console.
Other values are 1 and 2 indicating increasing level of verbosity. Default is
verbose=1
Number of cores to use in parallelization. If NULL, the number of
cores to use is detected automatically
a list with one data frame per sample with results of differential gene expression analysis
The method tests for differentially expressed genes between groups of spots/cells
(e.g., clusters) in a spatial transcriptomics sample. Specifically, the function
tests for genes with significantly higher or lower gene expression in one group of
spots/cells with respect to the rest of spots/cells in the sample. The method first
runs non-spatial linear models on the genes to detect differentially expressed genes.
Then spatial linear models with exponential covariance structure are fit on a
subset of genes detected as differentially expressed by the non-linear models (sp_topgenes).
If running on clusters detected via STclust, the user should specify the assignments
using the same parameters (w, k, deepSplit). Otherwise, the assignments are
specified by indicating one of the column names in x@spatial_meta. The function
uses spaMM::fitme and is computationally expensive even on HPC environments.
To run the STdiff using the non-spatial approach (faster), set sp_topgenes=0.