When changing the order of samples (ensuring that colData and countData are in the same order), this influences the DE results. Very subtly (something like in the 9th digit) - but why does it happen?

Using the pasilla dataset and the code from the vignette :

options(digits = 15)

library(DESeq2)
library(pasilla)

pasCts <- system.file("extdata",
                      "pasilla_gene_counts.tsv",
                      package="pasilla", mustWork=TRUE)
pasAnno <- system.file("extdata",
                       "pasilla_sample_annotation.csv",
                       package="pasilla", mustWork=TRUE)
cts <- as.matrix(read.csv(pasCts,sep="\t",row.names="gene_id"))
coldata <- read.csv(pasAnno, row.names=1)
coldata <- coldata[,c("condition","type")]
coldata$condition <- factor(coldata$condition)
coldata$type <- factor(coldata$type)

rownames(coldata) <- sub("fb", "", rownames(coldata))
cts <- cts[, rownames(coldata)]

dds <- DESeqDataSetFromMatrix(countData = cts,
                              colData = coldata,
                              design = ~ condition)



dds <- DESeq(dds)
res <- results(dds)


print(head(res$log2FoldChange),digits=15)


[1] -1.02604541037965413 -0.00215142369260044
[3]  0.49673556850473838  1.88276170249200669
[5]  0.24002523000310516  0.10479911223675623

When I randomly shuffle coldata and ensure that the counts is in the same order as count data, the logfold changes (very slightly). Why?

coldata2 <- coldata[sample(1:nrow(coldata),replace=F),]
setequal(rownames(coldata2), colnames(cts))

cts2 <- cts[,rownames(coldata2)]


dds2 <- DESeqDataSetFromMatrix(countData = cts2,
                              colData = coldata2,
                              design = ~ condition)



dds2 <- DESeq(dds2)
res2 <- results(dds2)


print(head(res2$log2FoldChange),digits=15)

[1] -1.02604541037965524 -0.00215142640531776
[3]  0.49673554526085623  1.88276152384409690
[5]  0.24002523019401523  0.10479911227720901

The calculations are not identical. IRLS is not a closed form solution, and we iterate until a convergence criterion.