Statistical test for separation of groups in PCA and DESeq2 design with confounders
Hi,
I have a simple methodological question regarding RNA-seq analysis. For my project there are 3 confounders known.
Question 1: If two groups of samples appear separated in a PCA plot of bulk RNA-seq data, what statistical test is typically used to quantify whether this separation between groups is significant? Is PCA visualization alone acceptable?
Question 2: When performing DESeq2 analysis, if there are known confounders (e.g., subtype, region & alteration(fusion, mutation)), should these variables be included in the design formula, for example:
design = ~ confounder1 + confounder2 + confounder3
instead of using only:
design = ~ confounder1
Thanks for any guidance.
• 219 views
•
link
updated
by
shannonNorgenBiotek •
•
written
by
archanaverma433 •
•
written
by
archanaverma433 •
0 answers
No answers yet.
Log in to answer this question.
More posts like this
-
PCA plots in ATAC-seq replicates: Is it okay to use vst transformed data for visualization?
written by maplewj •
-
DESeq and SVA for batch effect
written by michelafrancesconi9 •
-
Exploring batch effects in RNA-seq data with plotPCA
written by andres_arroyo •
-
Can i use PCA to summarise RNA expression profiles before and after treatment?
written by antonhomilius •
-
Batch effects vs biological variables
written by l.uva •
-
Differential expression- batch effects
written by Ml6237 •
-
limma::removeBatchEffect and using DESeq2 for RNA-seq differential expression
written by 2405592M
-
How to see if adjusting batch effect in RNA-seq is working or not
written by iraia.munoa
-
Differential expression on Combat-adjusted FPKM
written by mare116 •
-
R LIMMA Longitudinal analysis adjusting for continuous variables
written by robert.maughan •
Hello,
For Question 1:
A PCA is generally more for exploratory data analysis, giving you an idea of possible groupings, or even outliers. It should NOT be used alone to show differences between groups. As for statistical tests, you can try the Mann-Whitney U test, which is a non-parametric which tests for significant differences between 2 independent groups, just ensure your data meets any other assumptions of the test.
Question 2:
Yes, if you have known confounders, you should include them in your design formula, as you showed in your first example (design = ~ confounder1 + confounder2 + confounder3). Including the confounders will takes into account the effects of these variables in your differential expression results, especially if these are biological factors that could affect gene expression.