Differentially expression analysis

Question

[Sammy]

Heyya,

I'm performing a DE analysis on 5 groups, 1 control and 4 mutants.

The pipeline so far went: Raw data --> Trimmomatic --> Bowtie2 --> FeatureCounts --> Limma-Voom. I got the Limma-Voom report with the graphs and the 4 contrasts I set up: Control-Mutant1, Control-Mutant2, Control-Mutant3, Control-Mutant4. Everything is great, I even extracted the normalised count file from Limma-Voom.

However I want the differential gene expression tables with fold difference in expression. I can obtain that by processing the normalised counts and this is the blurry bit at the moment.

Question1: Do I need to calculate this separately for the 4 contrasts and basically get 4 tables at the end (each having the p-values and the log fold change)? I know this can be done in R with DESeq but can't I just extract it from Limma whilst doing the graphs?

Question2: Just to make sure I have the correct start here, all I need is the normalised counts, right?

Answer 1

[rodolfo.peacewalker]

Hi Sammy!

About your question 1, by using DESeq2 you might be able to indicate the contrast that you need. In this case, at the moment to construct your DESeqObject you should indicate in the design formula your groups to contrast:

dds <- DESeqDataSetFromMatrix(countData = s, #Put your raw counts table here
colData = meta, #Put your metadata table here
design = ~Mutants) #Indicate your design according how the counts depends on each group, in your case mutants vs control

In limma you specify which sample belongs to your groups (Control or Mutant) when you create the design matrix. Then, in your contrast matrix you have only to set the contrast by:

makeContrasts(
MutantsvsControl = Mutant - Control,
levels = colnames(design))

About your question 2, no you shouldn't use the normalised counts to perform a differential expression analysis. All you need is the raw counts estimates from Feature Counts to start the analysis.

Best regards!

Answer 2

[jared.andrews07]

You cannot/should not use normalised counts with DESeq2. There are some ways you could collapse your 4 results tables into a single one with additional columns for each comparison of interest via dplyr. Or, if appropriate, you could do an analysis comparing all mutants to your control group.

Alternatively, you can perform a meta-analysis comparing all 4 sets using GeneMeta or similar to identify commonalities and differences between them.

Answer 3

[Gordon Smyth]

The limma report from topTable includes log-fold-changes and p-values. That gives you want you want, no?