Benchmarking DESeq / EdgeR / LIMMA vs Lasso vs RF

Question

[fr]

Hi!

I'm wondering if anyone is aware of any benchmarking between approaches traditionally used in RNAseq or microarray analysis (e.g. DESeq2 / EdgeR / LIMMA) against others such as Random Forest or Lasso regression to identify genes / transcripts that are differentially expressed between conditions?

I'm aware that there may be different assumptions about data distributions, but still I'm curious about this. I tried searching for this but cannot find anything...

Thanks in advance

[Carlo Yague]

To my knowledge, this has never been tested.

Dedicated tools to analyze expression data are designed to accommodate specific difficulties, such as the high number of dimension and the usually (very) low number of replicates. I expect regression of any kind to perform poorly in comparison, because of those difficulties (try to fit a line when you have only 2 or 3 replicates). Not sure how a random forest would be useful in that context.