log vs square-root transformation

Question

[igor]

Traditionally, people log-transform expression data. I recently saw someone square-root transform their data. Is there a reason that method is not more widely used? One of the clear benefits is that it will work with 0s, so there is no need to add a pseudocount to the entire matrix, which will disproportionally impact the low values.

[ATpoint]

I guess log transformation more efficiently takes care of large differences in read counts.

Adding a tiny pseudocount like 0.1 probably does not really harm any downstream analysis, because low counts will still be low and lack power for differential analysis and high counts will not effectively be changed.

[jomo018]

Log transform symmetrizes under-expression and over-expression representations.

[Santosh Anand]

see https://stats.stackexchange.com/a/107690