Low abundance ASV filtering and relative abundance

I am trying to accomplish a task similar to this post - I want to generate (export for use in R) relative frequency feature tables with low abundance ASVs removed: (Removing Singletons and Filtering Out Low Relative Abundance). My question is that qiime feature-table filter-features only takes a feature (not relative feature) table. If you remove low abundant features based on counts first (as suggested), and then convert the table to relative abundance you distort the original abundance proportions of the sample. I can of course calculate this elsewhere by dividing by the total sample reads before filtering was done, but is there a way to do this in qiime that I’m missing?

Welcome to the forum @ahibb!

Yes this is something that will be fixed in a future release of QIIME 2, to allow filtering relative abundance tables.

Unless if you have samples with low read counts, or highly imbalanced counts across samples, then this effect should be negligible.

One possibility (if your samples are not highly imbalanced) would be to rarefy your feature tables at an even depth, filter based on counts (which will be even across samples), and then convert to relative frequency.

The other possibility: you are exporting to R anyway, and it would be trivial in R to filter a relative frequency table based on some threshold.

Do any of these offer a palatable solution for you?

Hi, thanks for the reply. I have tried the rarefy and filter ASVs less than 0.1% approach using the count data, but some of my samples have a high number of reads assigned to low abundance features (hence the distortion when converting to relative abundance after count filtering). So I will do the relative abundance filtering in R, but glad to hear it’s on the list for a future release!

1 Like

This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.