NBumi_FSOther: Other Feature Selection Methods
In tallulandrews/M3D: Michaelis-Menten Modelling of Dropouts in single-cell RNASeq

Description Arguments Details Value Examples

Ineffective alternative feature selection methods based on the depth-adjusted negative binomial model. Functions tagged with "bg__" are not meant for direct usage and are not available in the Bioconductor release.

`fit`	output from NBumiFitModel or NBumiFitBasicModel.
`window_size`	window for calculating the moving median.

Calculates dropout probability for each observation using depth-adjusted negative binomial means and dispersions calculated from a fitted power-law relationship between mean and dispersion. Total dropouts per gene are modelled using the normal approximation of the sum of bernoulli variables. And significance is evaluated using a Z-test.

obsolete__nbumiFeatureSelectionDropouts Ranks genes by significance of increase in dropouts compared to expectation allowing for gene-specific dispersions. obsolete__nbumiFeatureSelectionHighVarDist2Med Ranks genes by the distance to median of log-transformed estimated dispersions.

Sorted vector of p-values/distances.

library(M3DExampleData)
#counts <- as.matrix(Mmus_example_list$data);
#counts <- counts[rowSums(counts) > 0,];
#fit <- NBumiFitModel(counts);
#Dropout_features <- names(obsolete__nbumiFeatureSelectionDropouts(fit)[1:2000]);
#dist2med_features <- names(obsolete__nbumiFeatureSelectionHighVarDist2Med(fit)[1:2000]);