Entering edit mode
9.2 years ago
melati
▴
10
Hi,
I have RNA-seq data with 3 isolates (Isolate 1, isolate 2, isolate 3), for each isolate there are 2 samples(Media A, Media B) and each sample have 3 biological replicates (R1, R2, R3). So, is there any way to combine these biological replicates to be representative as one using Cuffdiff? Please let me know your opinions. Thank you.
Why do you want to use cuffdiff? You can create a design matrix and use
edgeRorDESeqfor more robust analysis?What if they want to use Cuffdiff instead? I'm thinking the same way as them, is there any possibility to combine the biological replicate? Otherwise, is it practical enough if simply pick any biological replicate when making the differential expression analysis?
I second Goutham Atla suggestion to use a method that explicitly takes into account replicates within design variables (i.e. linear models as implemented in edgeR/DEseq). Picking one representative replicate or combining replicates will result in less data being used and/or ignoring the variation between replicates. (I haven't used cuffdiff for a long time, so I don't know if a better handling of replicates is possible)
Isolates as in 3 strains of different bacteria/virus ?
The isolates came from same pathogen with different level of pathogenecity which is isolate 1 with most virulent, isolate 2 is medium and isolate 3 is less virulent.
Now, what you want to compare ? The effect of media with in each isolate ? or the DE genes between isolates ?
I want to campare DE genes between isolates. Somehow it makes me confuse. Let say.. can I compare isolate1 (replicate1) with isolate2 (replicate1) or.. maybe isolate1 (replicate2) with isolate2 (replicate3) etc?
If you want to compare DE genes between isolates, better to create a design matrix and use edgeR/DESeq. I don't think so Cuffdiff can handle this experimental design. I don't think you are "interested" to use cuffdiff, may be you feel its easy to use or you don't have proper annotations for organism of your interest.
The best method is to compare all of the replicates from isolate1 with all of them from isolate2. Do not merge the samples or otherwise just use only 1 of them! If you want to include the samples for both media types as well then you can't use cuffdiff, it's simply not flexible enough. Use edgeR, DESeq2, or limma/voom instead.