getNBGaussianLikelihood: Calculate the log likelihood of observed read counts

In Carldeboer/MAUDE: Mean Alterations Using Discrete Expression

Calculate the log likelihood of observed read counts

Description

Uses a normal distribution (N(mu,sigma)) to estimate how many reads are expected per bin under nullModel, and calculates the log likelihood under a negative binomial model. This function is usually not used directly.

Usage

getNBGaussianLikelihood(x, mu, k, sigma = 1, nullModel, libFract)

Arguments

x	a vector of guide counts per bin
mu	the mean for the normal expression distribution
k	the vector of total counts per bin
sigma	for the normal expression distribution (defaults to 1)
nullModel	the bin bounds for the null model (for no change in expression)
libFract	the fraction of the unsorted library this guide comprises (e.g. from unsorted cells, or sequencing the vector)

Value

the log likelihood

Examples

#usually not used directly
#make a bin sorting model with 6 10% bins
curSortBins = makeBinModel(data.frame(Bin = c("A","B","C","D","E","F"), fraction = rep(0.1,6)))
readsForGuideX =c(10,20,30,100,200,100); #the reads for this guide
getNBGaussianLikelihood(x=readsForGuideX, mu=1, k=rep(1E6,6), sigma=1, nullModel=curSortBins, 
  libFract = 50/1E6)
getNBGaussianLikelihood(x=readsForGuideX, mu=-1, k=rep(1E6,6), sigma=1, nullModel=curSortBins, 
  libFract = 50/1E6)
#mu=1 is far more likely (closer to 0) than mu=-1 for this distribution of reads

Carldeboer/MAUDE documentation built on March 27, 2022, 8:50 p.m.