RNA-seq data comparison across experiments

Question

[Afrodite]

Hello everybody,

just a simple question: How reliable statistically or biologically would be to compare RPKM expression values from different species, experiments, data sources?

And to be more specific: is it appropriate to compare let's say one gene's RPKM value from one experiment with the same gene's RPKM value found in the public/open RNA-seq databases? Do you think that the same normalization should be done across experiments?

Thank you in advance for the help!

[Arindam Ghosh]

Hello!

I am looking forward to do something similar. As it has been a long time since you have posted this question, I hope you have found a solution to this. Please share or recommend some paper regarding this.

Thank You

Answer 1

[Devon Ryan]

Even with a more reliable metric (i.e., not RPKM, which can be calculated in multiple ways), making comparisons across experiments (species are much harder and probably far from a solved problem) needs to be done with care. If you're going to compare across experiments, you absolutely need replicates within each experiment, ideally known constantly expressed genes (since you'd need these to normalize against, given that your experimental effect is completely confounded in the experimental batch effect), and to use something like SVA.

If possible, get the raw data from the public datasets and process it in an identical manner to your own. This won't get rid of the batch effect, but it will at least decrease it to something manageable.

BTW, this is not something I would recommend a beginner do without consulting someone locally. For the results to have utility, you really need to nicely correct for batch effects...and checking that this was done fully takes a bit of experience/trial-and-error.