Inverse logFC on DESeq2 Rstudio
1
0
Entering edit mode
Luiz • 0
@luiz-13209
Last seen 5.5 years ago
Brazil

Hi

I'm trying to work with DESeq2, i already worked with edgeR and DESeq1 with relatively sucess.

They give me the same results as logFC (+ or -) but, when i work with DESeq2, i have different results. Every gene who are positive for edgeR and DESeq1 are negative for DESeq2 and vice-versa.

DESeq2 v. 1.14.1

Rstudio v.1.0.136

I'll post my pipeline, thanks for your help!

bckCountTable <- read.table("DF.txt", header=TRUE, row.names=1)
head(bckCountTable)
plot(bckCountTable)
samples <- data.frame(row.names=c("CTL1", "CTL2", "CTL3", "HT1","HT2","HT3"), condition=as.factor(c(rep("R",3),rep("IR",3))))
samples
bckCDS <- DESeqDataSetFromMatrix(countData = bckCountTable, colData=samples, design=~condition)
bckCDS <- bckCDS[ rowSums(counts(bckCDS)) > 1, ]
bckCDS_1 <- DESeq(bckCDS)
res <- results(bckCDS_1, alpha=0.05, betaPrior = F,  contrast=c("condition", "CTL", "HT"), pAdjustMethod = "BH")
rld <- rlogTransformation(bckCDS)
print(plotPCA(rld, intgroup=c('condition')))
head(assay(rld))
hist(assay(rld))
plotDispEsts(bckCDS_1)
bckCDS_1 <- estimateSizeFactors(bckCDS_1)
head(res)
table(res$padj<0.05)
resdata <- merge(as.data.frame(res), as.data.frame(counts(bckCDS_1, normalized=TRUE)), by="row.names", sort=FALSE)
names(resdata)[1] <- "Gene"
head(resdata)
X = as.data.frame(res)
plotSparsity(X, normalized=T)
boxplot(bckCountTable)
summary(res)
plotMA(res, ylim=c(-15,15))
topGene <- rownames(res)[which.min(res$pvalue)]
with(res[topGene, ], {
    points(baseMean, log2FoldChange, col="dodgerblue", cex=2, lwd=2)
    text(baseMean, log2FoldChange, topGene, pos=2, col="dodgerblue")
})
hist(X)
mat = assay(rld)[head(order(res$padj),30),]
mat = mat - rowMeans(mat)
pheatmap(mat)
simple <- ggplot(X, aes(x = log2FoldChange, y = stat)) + geom_point(size = 3, alpha = 0.7, na.rm = T) + theme_bw(base_size = 16) + geom_text_repel(aes(log2FoldChange, stat, label = rownames(bck_res)), colour="black", segment.colour="grey")
simple

DESeq2 • 2.7k views
ADD COMMENT
0
Entering edit mode

In your contrast=c("condition", "CTL", "HT") you're saying that you're looking at fold changes relative to "HT" - the second element in the character vector is the numerator, and the third is the level that will be denominator (or 'be subtracted' on the log scale).  Guessing that "CTL" means control, and that normally we want fold-changes relative to control, it looks like have got the order of those two levels the unintended way round, so just setting contrast=c("condition", "HT", "CTL")  should get you back to the more intuitive interpretation that you've already got in edgeR and DESeq1?

ADD REPLY
0
Entering edit mode

Thank you Gavin!!

My summary before was

out of 20088 with nonzero total read count
adjusted p-value < 0.05
LFC > 0 (up)     : 17, 0.085% 
LFC < 0 (down)   : 12, 0.06% 
outliers [1]     : 48, 0.24% 
low counts [2]   : 7006, 35% 
(mean count < 9)
[1] see 'cooksCutoff' argument of ?results
[2] see 'independentFiltering' argument of ?results

NOW is

out of 20088 with nonzero total read count

adjusted p-value < 0.05
LFC > 0 (up)     : 12, 0.06% 
LFC < 0 (down)   : 17, 0.085% 
outliers [1]     : 48, 0.24% 
low counts [2]   : 7006, 35% 
(mean count < 9)
[1] see 'cooksCutoff' argument of ?results
[2] see 'independentFiltering' argument of ?results

 

The truth is in minimal details! haha

Really thank for your time.

 

ADD REPLY
0
Entering edit mode
@mikelove
Last seen 1 day ago
United States
See the vignette section on setting factor levels.
ADD COMMENT

Login before adding your answer.

Traffic: 636 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6