How can I select highly variable genes from the RNASeq gene dataset?

Question

[naseerkhan861]

How can I select only the genes with a certain variance threshold, I know techniques like Z statistic and T statistics and then calculation p-value and then correct for FWER or FDR? but is there some modern easy to use techniques that is also available in R with some solid references that I can deploy to my data? I have RNASeq FPKM dataset to be specific.

Regards

[hafiz.talhamalik]

what's wrong with the typical Z and T statistics with FDR ???

[naseerkhan861]

See the great and well explained answer from the person below.

Answer 1

[Michael]

There is no need to cite simple statistics such as variance or Z statistics, however, you might consider median absolute deviation (MAD) as a robust alternative to variance, afaik this is commonly used as a filtering step in network analysis. If you do the filtering in R or another software, you can add a sentence like "all statistical analyses were done in R (R Core Team (2018))". You might also use other more advanced differential expression statistics like limma, DESeq2, etc. If you are using a particular package, you cite this package.

In general, there is no single best or authoritative way of filtering prior to downstream analyses.

In particular, I do not know what your intended downstream analysis is, so there is not much more I can recommend at this stage.

R Core Team (2018). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://www.R-project.org/.

Caveat: do not use FPKM, use CPM or TPM instead (this has been discussed here many times).

[naseerkhan861]

My objective is to do coexpression analysis after I filter some genes with low variance, is that now more clearer to you now what I want to do?

[Michael]

For co-expression analysis, you might use MAD with a certain threshold (e.g. MAD > 2) but remember that such thresholds are necessarily arbitrary.