NBumi_FitModel: Fit Depth-Adjusted Negative Binomial Model

In M3Drop: Michaelis-Menten Modelling of Dropouts in single-cell RNASeq

Description

Fits means and dispersions for depth-adjusted or basic negative binomial models to a read/UMI count matrix.

Usage

	NBumiFitModel(counts)
	NBumiFitBasicModel(counts)
	

Arguments

counts

a numeric matrix of raw UMI counts, columns = samples, rows = genes.

Details

NBumiFitModel Fits a depth-adjusted negative binomial model for each expression observation. Each expression is modelled using a negative binomial distribution with mean equal to

t_i*t_j/T,

where t_i is the total counts for sample i, t_j is the total counts for gene j and T is the total counts. Dispersions (R 'siz'e parameter) are fit such that:

var_j(counts_ij-mu_ij) = sum(mu_ij+mu_ij^2/size_j).

Cases where genes exhibit poissonian behavior (size->infinity) are assigned a size of 10^10.

NBumiFitBasicModel Fits a basic negative binomial model for each expression observation. Each expression is modelled using a negative binomial distribution with mean equal to

t_j/n,

where t_j is the total counts for gene j and n is the number of cells. Dispersions (R 'size' parameter) are fit such that:

var_j(counts_ij) = sum(mu_j+mu_j^2/size_j).

Cases where genes exhibit poissonian behavior (size->infinity) are assigned a size of 10^10.

Value

A named list of:

mus

= a genes by samples matrix of means for the observation-specific negative binomial

sizes

= a vector of dispersions for each gene.

vals

= named list of summary statistics of the original count matrix:

tis

= total molecules per cell,

tjs

= total molecules per gene,

dis

= total dropouts per cell,

djs

= total dropouts per genes,

total

= total molecules in dataset,

nc

= number of cells,

ng

= number of genes

Examples

	library(M3DExampleData)
	counts <- NBumiConvertData(Mmus_example_list$data)
	fit <- NBumiFitModel(counts);

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