Other_FS: Other Feature Selection Methods

Description Usage Arguments Details Value Examples

Description

Other feature selection methods which only rank genes.

Usage

1
2
3
	irlbaPcaFS(expr_mat, pcs=c(2,3))
	giniFS(expr_mat, suppress.plot=TRUE)
	corFS(expr_mat, dir=c("both", "pos", "neg"), fdr=NULL)

Arguments

expr_mat

normalized but not log-transformed gene expression matrix, rows=genes, cols=cells.

pcs

which principle components to use to score genes.

suppress.plot

whether to plot the gene expression vs Gini score.

dir

direction of correlation to consider.

fdr

apply an fdr threshold for significant features.

Details

irlbaPcaFS Features are ranked by the sum of the magnitude of the loadings for the specified principle components. PCA is performed using the irlba package using sparse matricies for speed. giniFS Fits a loess curve between the maximum expression value and gini-index of each gene. Genes are ranked by p-value from a normal distribution fit to the residuals of the curve. As proposed by GiniClust [1]. corFS Calculates all gene-gene correlations then ranks genes by the magnitude of the most positive or negative correlation. "both" will rank by the average of the magnitudes of the most positive & negative correlation.

Value

Sorted vector of scores for each gene from best features to worst features.

Examples

1
2
3
4
5
library(M3DExampleData)
norm <- M3DropConvertData(Mmus_example_list$data[1:500,]);
Features_gini <- giniFS(norm, suppress.plot=FALSE);
Features_cor <- corFS(norm);
Features_pca <- irlbaPcaFS(norm);

M3Drop documentation built on Nov. 8, 2020, 5:06 p.m.