Search
Question: Up and down regulation in LIMMA for microarray dataset
0
17 months ago by
danielle.newby0 wrote:

Hi,

I need some clarification on my results as i have been getting myself very confused!

I have an an microarray experiment where i have controls and cases (dosed with a drug). I have used the following in the design matrix for limma:

design <- model.matrix( ~CLASS) )

This creates a design matrix:

intercept CLASS$Drug 1 0 1 0 1 1 1 1 1 1 The zero's are my control group and the 1s are those samples dosed with the drug. Therefore the control group is my reference group. From my output from limma for each gene i get a logFC. If i have a positive logFC for a gene does this mean that my gene is upregulated in my Drug group compared with my control group for that gene? I have been reading lots of posts but i have now gotten very confused so any help will be appreciated! Thanks for any help, Danielle ADD COMMENTlink modified 17 months ago by Aaron Lun21k • written 17 months ago by danielle.newby0 In general, it's useful to "inspect the raw data". For example, take one of your most significant DEGs and make a box plot of expression values (raw counts, TPM, voom-normalized) in both conditions; repeat with some non-significant genes for comparison. Among other things, this will show you if the expression is higher with the drug or without. This is normal part of my own debugging/QA process to make sure I've got the design matrix correct and stuff - if you've done it wrong, your top DEGs won't look very different from non-significant DEGs. ADD REPLYlink written 17 months ago by glocke010 0 17 months ago by Aaron Lun21k Cambridge, United Kingdom Aaron Lun21k wrote: Yes. If you have a positive log-fold change, the gene should be upregulated in the drug-treated group compared to the control. In your case, the log-fold change is the CLASS$Drug coefficient; so if you have positive values for this coefficient, the fitted value for the drug-treated group (Intercept + CLASS\$Drug) must be greater than those of the control group (Intercept only).