Rarefaction and low frequency reads

[Fra (Francesco)]

Hello everybody,

I have a quick question, would you remove low frequency reads before or after the rarefaction step?

Thanks!

[Mehrbod_Estaki]

Hi @Fra,

I assume you mean rarefying as in sub-sampling to equal depth and not referring to creating rarefaction plots. Rarefying should -in my opinion- be done after removing low frequency taxa since the whole purpose is to reach an even depth across all samples. If you were to remove some taxa after rarefying then you would be back to potentially having uneven depth. Though in most cases removing rare taxa following rarefying probably isn’t going to re-introduce that much variance in sampling depth so it wouldn’t be the end of the world.
For rarefaction plots I personally like to do this on the original non-rarefied tables, but only because my aim is to simply analyse my sampling efforts not run statistical tests.

Just a note that if you are using log-ratio transformed methods such as those in gneiss and ANCOM, you don’t need to rarefy your data, though it still is a good idea to remove low abundant features.

[Fra (Francesco)]

Thanks @Mehrbod_Estaki!