DA-seq Step 4: STG feature selection

Use STG (stochastic gates) to select genes that separate each DA region from the rest of the cells. For a full description of the algorithm, see Y. Yamada, O. Lindenbaum, S. Negahban, and Y. Kluger. Feature selection using stochastic gates. arXiv preprint arXiv:1810.04247, 2018.

STGmarkerFinder(X, da.regions, da.regions.to.run = NULL, lambda = 1.5,
  n.runs = 5, return.model = T, python.use = "/usr/bin/python",
  GPU = "")

Arguments

X	matrix, normalized expression matrix of all cells in the dataset, genes are in rows, rownames must be gene names
da.regions	output from the function getDAregion()
da.regions.to.run	numeric (vector), which DA regions to run the marker finder, default is to run all regions
lambda	numeric, regularization parameter that weights the number of selected genes, a larger lambda leads to fewer genes, default 1.5
n.runs	integer, number of runs to run the model, default 5
return.model	a logical value to indicate whether to return the actual model of STG
python.use	character string, the Python to use, default "/usr/bin/python"
GPU	which GPU to use, default '', using CPU

Value

a list of results:

da.markers

a list of data.frame with markers for each DA region

accuracy

a numeric vector showing mean accuracy for each DA region

model

a list of model for each DA region, each model contains:

features

features used to train the model

selected.features

the selected features of the final run

pred

the linear prediction value for each cell from the model