A function to calculate the importance score of ADT
importanceADT( sce, altExp_name = "ADT", exprs_value = "logcounts", method = c("randomForest", "PCA"), group = NULL, subsample = TRUE, times = 10, prop = 0.8, k_pca = 5, remove_first_PC = TRUE, ... )
sce | A singlecellexperiment object |
---|---|
altExp_name | A character indicates which expression matrix is used. by default is none (i.e. RNA). |
exprs_value | A character indicates which expression value in assayNames is used. |
method | A character indicates the method of ADT importance calculation, either randomForest or PCA |
group | A vector indicates the grouping of the data (for random forest) |
subsample | Whether perform subsampling (for random forest) |
times | A numeric indicates the times of subsampling is performed (for random forest) |
prop | A numeric indicates the proportion of cells are subsampled from the whole data (for random forest) |
k_pca | Number of principal component will be used to calculate the loading scores (for PCA) |
remove_first_PC | A logical input indicates whether the first component will be removed from calculation (for PCA). |
... | other arguments to `randomForest()` or `prcomp()` function |
A SingleCellExperiment object
For random forest, the importance scores are based on features importance. For PCA, it implements the method proposed in Levin et al (based on the loading of features).
Levine, J.H., Simonds, E.F., Bendall, S.C., Davis, K.L., El-ad, D.A., Tadmor, M.D., Litvin, O., Fienberg, H.G., Jager, A., Zunder, E.R. and Finck, R., 2015. Data-driven phenotypic dissection of AML reveals progenitor-like cells that correlate with prognosis. Cell, 162(1), pp.184-197.
data("sce_control_subset", package = "CiteFuse") sce_control_subset <- importanceADT(sce_control_subset, group = sce_control_subset$SNF_W_louvain, subsample = TRUE)