NOISeq

• For experiments with or without replication.
• Compares replicates within the same condition to estimate noise distribution of M (log-ratio ) and D (absolute value of the difference). A feature is considered to be differentially expressed if its corresponding M and D values are likely to be higher than noise values.
• Input can be raw or normalized counts.
• It has several normalization options: counts per million read, RPKM, Upper Quartile. It can also use pre-normalized data, for example, data normalized with DESeq.
• Output statistic is probability of being differentially expressed. So, the higher and the closer to 1 the better. (It is NOT a p-value).
• Sample code with replication:

    source("NOISeq.r")
    mydata <- readData(file = "All_Counts_no0_1pc_Header.txt", cond1 = c(2:3), cond2 = c(4:5), header = TRUE)
    myresults <- noiseq(mydata[[1]], mydata[[2]], repl = "bio", q = 0.9, nss = 0)
    write.table(cbind (myresults$Ms[myresults$deg],myresults$probab[myresults$deg] ), file="genesDE_FCUHRvbrain.txt", quote=F, sep="\t")
  
    #nss = 0 If the experiment didn’t include replicas the number of replicates to be simulated is provided by nss parameter
    #repl = "bio“ indicates that the experiment includes biological replicates
    #q = 0.9 indicates that the probability cut off for considering a gene differentially expressed is 0.9