ples matrix. Both had the same count matrix. But the matrices

samples from two working data sets were different. However,

e normalisation factors had been used. Based on the normalised

ix Z1, the negative binomial conditional likelihood function was

stimate the dispersion value across all genes

Z2=estimateCommonDisp(Z1)

2 object inherited some entries of the Z1 object. In the Z2 object,

ounts data was inherited from previous data structures, such as

Z1. The normalised counts data was included in the Z2 object,

as saved in $pseudo.counts. The raw integer counts data had

malised. The value of $AveLogCPM stands for the average base

rithm of the counts-per-million for each row of counts.

the empirical Bayes approach was used to estimate the tag-wise

n for all genes. The estimation function was the weighted

al maximum likelihood. The code is shown below,

Z3=estimateTagwiseDisp(Z2)

dgeR model was constructed using exactTest based on the

t,

model=exactTest(Z3)

dgeR model has an entry named as $table, which was

d of three columns standing for the base two logarithm fold

he average base two logarithm of the counts-per-million and the

(model$table)

logFC logCPM PValue

000000003 0.293998840 5.087404 0.2525471

000000005 0.000000000 -3.498357 1.0000000

000000419 -0.051710730 4.565493 0.8372578

000000457 -0.002191778 3.879677 0.9967011

000000460 0.049206508 2.020877 0.8960975

000000938 0.586996884 -3.255919 1.0000000