If a Susceptible and a Resistant plant transcriptomes are sequenced in different batches and if the coverage obtained for one is say, twice that of the other, how correct is comparing the FPKM for particular genes since the no. of million reads for each sample differs greatly?
It could be possible, but, as has been said, the potential biases are many.
Why use FPKM if you are comparing sample to sample?
You could use normalized read counts, not normalized per lenght of transcript, but normalized taking into account the library size (edgeR, limma and DESeq2 all use some kind of RNA-Seq specific normalization with this purpose) if what you want is to compare one sample to the other. As far as the samples are coming from the same genome, with the same gene-length and you are measuring gene-level counts, that would reduce the potential biases and statistical complications.
If you are, on the other hand, interested in isoform level, then of course you will use some kind of FPKM on transcripts (cufflinks, rsem...) which include length in the normalisation. There it is harder to model everything and obtain robust DE results. However, what people could do in those cases is to:
1. Use as many replicates as possible
2. Introduce all potential biases as co-variates:
~ condition + 3bias + GCcontent + batch ... or alike
Jose
it is possible upon what questions you are asking... For genome-wide comparison including many genes, I would suggest you first identify some references genes that expression consistently (or with known pattern verified with RT_PCR) among samples.
It is possible in your case. However, if your read length is way too differ, i.e. 50bp vs 100bp, then I would use a second method, such as DESeq to further validate your comparison.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Meta-Analysis