Multifactorial Design for DESEQ2
1
0
Entering edit mode
@b4de4377
Last seen 10 weeks ago
Netherlands

Hello community,

I'm interested in conducting a DEG analysis to obtain differentially expressed genes based on treatment per cell line. In other words, determine the effect of condition per cell line and extract DEGs based on treatment.

## Overview

Experimental Design: This consists of a two-factorial design with factors: cell line (A vs B) and treatment (control vs treatment).

Research Question: Does the treatment have a different given effect per cell line?

Goal: Compare the effect of treatment per cell line (A vs B)

## Experimental Design:

sample  cell_line condition
A_Ctr_1    A     control
A_Ctr_2    A     control
A_Ctr_3    A     control
A_Met_1    A     treatment
A_Met_2    A     treatment
A_Met_3    A     treatment
B_Ctr_1    B     control
B_Ctr_2    B     control
B_Ctr_3    B     control
B_Met_1    B     treatment
B_Met_2    B     treatment
B_Met_3    B     treatment


## DESeq2 Analysis

#Check multi-factorial design for experimental design
print(model.matrix(~cell_line + condition, expDesign))

# Constructing the DESeq2 object (using two design factor)
dds <- DESeqDataSetFromMatrix(countData = geneCountsMat,
colData = expDesign,
design = ~ cell_line + condition + cell_line:condition)

# Filter out lowly expressed genes, here the rowSums(counts(dds)) >= 10 filters out low-count genes
# i.e. keep rows that have at least 10 reads
dds <- dds[ rowSums(counts(dds)) >= 10, ]

#select the reference level for comparing cell lines (set the factor level)
#dds$cell_line <- relevel(dds$cell_line, ref = "A")

"Running DESeq"
# Estimate size factors and dispersion
dds <- DESeq(dds)

# see all comparisons (here there are two given we want to compare conditions and cel_lines)
resultsNames(dds)


## Questions:

1. Is the design here enough and how can I obtain genes per cell line, would this be done with contrasts in results?
2. Do I need to relevel the baseline per cell line in this case?
3. Should I instead use the interactions instead to obtain genes per cell line?
4. vst normalization also necessary here?
DESeq2 rna-seq • 206 views
1
Entering edit mode

There is a section in the vignette that covers interactions and (depending on use case) how one can do a full factorial design to make interpretation simpler. vst is not part of DE testing, it's for downstream tasks such as PCA and visualization, see vignette.

1
Entering edit mode
@mikelove
Last seen 2 days ago
United States

For questions about the experimental design, I recommend to consult with a local statistician. I have to reserve my time on the support site for software related questions.