How to use limma for a categorial outcome variable? Where outcome is the cohort status and predictor is gene expression.

Question

[halo22]

Hello All,

I am very new to limma and trying to build a logistic regression like model using limma where my goal is to understand associations between the cohort status and gene expression while accounting for variables like age and sex. My gene expression is quantified using RSEM with TPM counts. I have normalized my data using TMM normalization. The following is my code snippet for limma.

design <- model.matrix(~0+Cohort+sex+age, data=PhenoType)   #Cohort:1/0
v <- voomWithQualityWeights(myNormalized_data, design=design, normalization="none", plot=TRUE)
fit1 <- lmFit(v,design)
fit2 <- eBayes(fit1)

Is this design correct? Again, I am not interested in group comparison but predicting cohort status using expression values.

Answer 1

[h.mon]

Do not use TPM values, use raw counts.

Differential expression analysis starting from TPM data

[Kevin Blighe]

Following from h.mon's point, can you please be very clear about your data processing steps?

You can use the RSEM estimated counts with limma, edgeR, or DESeq2. Take a look at these answers on Bioconductor:

• https://support.bioconductor.org/p/91054/#91098
• https://support.bioconductor.org/p/91054/#91073