PCA and DeSeq_what is wrong?

[DL4Crawford]

Hi,
We have some unusual PCA patterns from DeSEQ
We have288 individuals from 3 population and two treatments.
When we do a PCA, each population/treatment is separated into two distinct groups (based on 5,000 genes for 288 individuals).

One is treatment “F" with three pop, where each of the three populations is divided into two groups in the first principle component.
The second is treatment “A", where each of the three population is again divided into two groups but this time by the 2nd principle component.

We are most concerned that these divisions within a population/treatment are an artifact of DeSeq normalization.

dlc

[Michael Love]

hi,

Id suggest you try the rlog transformation in DESeq2 (note an argument fast=TRUE to speed up computation for datasets with so many samples). We created a workflow here which explores PCA and multidimensional scaling using the rlog+Euclidean distance as well as the Poisson Distance: http://bioconductor.org/help/workflows/rnaseqGene/

Also it's a good idea to post your code and the output of: sessionInfo()