I have a data set of 51 samples over 4 different conditions, and I want to visualise the similarity between the groups. I have already identified a known blood contamination which affects 7 of the samples, and have added a column named "contamination", with the labels "yes" or "no".

However, when I include this term in the design matrix, it does not affect the appearance of the PCA plot. It looks the same as without the term, and the 7 samples are outliers in reference to the other samples of the same condition.

Code:

d.deseq <- DESeqDataSetFromMatrix(countData = raw_counts,
                                  colData = sample_data,
                                  design = ~ contamination + condition)

vsd <- vst(d.deseq, blind=FALSE)

pcaData <- plotPCA(vsd, intgroup=c("condition"), returnData = TRUE)
percentVar <- round(100 * attr(pcaData, "percentVar"))

p <- ggplot(pcaData, ...)

Thanks a lot for help with troubleshooting and/or other suggestion how to deal with the contamination.

A "frequently asked question":

https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html#why-after-vst-are-there-still-batches-in-the-pca-plot