M3D_TraditionalDE: Perform Traditional Differential Expression
In tallulandrews/M3D: Michaelis-Menten Modelling of Dropouts in single-cell RNASeq

Description Usage Arguments Details Value Examples

Performs traditional (i.e. compare defined groups) differential expression using a negative binomial model with MM zero-inflation. Functions tagged with "bg__" are not meant for direct usage and are not available in the Bioconductor release.

1 2	bg__get_mean2disp(expr_mat) bg__fitdispersion(expr_mat)

expr_mat

a numeric matrix of library-size normalized expression values, columns = samples, rows = genes.

THESE FUNCTIONS SHOULD NOT BE USED.

unfinished__m3dropTraditionalDE : Uses a log-likelihood ratio test to perform model selection between a model of constant mean expression vs a model of different mean expression across the biological groups. Probabilities of observing the data given the model are calculated using a zero-inflated negative binomial distribution. Global relationships between mean and dispersion (power-law) as well as mean and dropouts (Michaelis-Menten) for genes are fit independently for each batch. Dispersions are fixed for each batch and calculated using the fitted power-law using the global mean expression of each gene. Significance is evaluated using the chi-square distribution.

unfinished__m3dropTraditionalDEShiftDisp: Uses a log-likelihood ratio test to perform model selection between a model of constant mean expression vs a model of different mean expression across the biological groups. Probabilities of observing the data given the model are calculated using a zero-inflated negative binomial distribution. Global relationships between mean and dispersion (power-law) as well as mean and dropouts (Michaelis-Menten) for genes are fit independently for each batch. Dispersions are shifted from the global variance according to the mean expression for each biological group, using batch-specific power-laws. Significance is evaluated using the chi-square distribution.

broken__m3dropCTraditionalDE : Uses a log-likelihood ratio test to perform model selection between a model of constant mean expression vs a model of different mean expression across the biological groups. Probabilities of observing the data given the model are calculated using a zero-inflated negative binomial distribution. Global relationships between mean and dispersion (power-law) as well as mean and dropouts (Michaelis-Menten) for genes are fit to the full dataset. Significance is evaluated using the chi-square distribution.

bg__get_mean2disp fits a power-law relationship between the squared coefficient of variation and mean expression of each gene, which is used to calculate the expected dispersion parameter for the negative binomial distribution from a given mean expression value.

bg__fitdispersion estimates gene-specific dispersions from the mean and variance of gene expression values

r = mu^2/(var-mu)

. Then fits a power-law relationship between the estimated dispersion and mean exprssion.

bg__m3dropTraditionalDE : a table of observed mean expression levels for each biological group and each batch with raw p-values and FDR corrected p-values for each gene. bg__m3dropTraditionalDEShiftDisp : a table of observed mean expression levels for each biological group and each batch with raw p-values and FDR corrected p-values for each gene. broken__m3dropCTraditionalDE : a table of observed mean expression levels for each biological group with raw p-values and FDR corrected p-values for each gene. bg__get_mean2disp : a function which calculates the expected dispersion given the mean expression. bg__fitdispersion : exponent of the power-law relationship between dispersion and mean expression.

library(M3DExampleData)
#Normalized_data <- M3DropCleanData(Mmus_example_list$data,
#                  labels = Mmus_example_list$labels,
#                  is.counts=TRUE, min_detected_genes=2000)
#DE_output <- bg__m3dropTraditionalDE(Normalized_data$data[1:100,], Normalized_data$labels)
#DE_shifted_output <- bg__m3dropTraditionalDEShiftDisp(Normalized_data$data[1:100,], Normalized_data$labels)
#DE_output_batches <- bg__m3dropTraditionalDE(Normalized_data$data[1:100,], Normalized_data$labels, batches=Normalized_data$labels) # each biological condition was performed separately