Calculating Fold Change from regularized log2- transformed count data, generated by DESeq2 function rlog

Question

[tommaso.gastaldi]

Hello to everyone. I have regularized log2- transformed count data (from bulk RNAseq), generated by DESeq2 function rlog. I have technical triplicates of cells untreated and treated. I´ve been suggested to calculate the fold change between the triplicates for every gene to see easily a difference in the gene expression between untreated and treated conditions. Due to the triplicates are quite consistent between each other for all the genes, I decided to calculate the mean first and then make the ratios of the means (treated/untreated) to calculate the fold change. My questions are: 1) Does it make sense first to calculate a fold change from regularized log2- transformed count data generated by DESeq2 function rlog? 2) Is correct to calculate in this way a fold change? Thank you all so much for the replies and help!

Answer 1

[russhh]

You'd typically use DESeq2 to calculate to fold-changes.

[ATpoint]

Agreed. quite consistent is not a very informative term. Just run DESeq on your data with default settings and get the fold changes it produces. This is more meaningful than custom approaches as it corrects for high fold changes based on small counts. It will also tell you if the changes are significant with the given number of replicates and the observed dispersion between the replicates.

[tommaso.gastaldi]

I just had a look to the values from each triplicates from each condition (just an example below): untreated 8.635868965 8.717134233 8.644193702 treated 10.58008887 10.47967565 10.56590922 This what I think "quite consistent" as triplicates. Anyway, I got the DESeq2 results, so I'll use them. Thanks a lot for the help!

[tommaso.gastaldi]

Ok. My bioinformatician gave me also the DESeq2 files with the log2FCs. I proceed with them. Thanks for the reply!

[ATpoint]

Answer 2

[swbarnes2]

rlog transformed data is generally not used for calculating fold changes. It's for other things, like PCA calculations. Just find a tutorial and follow that. No on is going to understand why you want to do things your own way when experts who figured all this out don't do things like that.

Do you really only have technical replicates, and not biological ones? Like, the same library prep run on three lanes? That's not going to be very informative at all.

[tommaso.gastaldi]

I just tried to follow a suggestion from a colleague of mine. I wasn't even sure it was a good one. I just gave it the try ;). As stated in the other coments, I'll use the DESeq2 results that the bioinformatician now gave me. Thanks for the reply :)