Hi Friends,
I know that differential testing typically requires removing low-prevalence taxa.
Do I also need to remove ASVs (inferred by DADA2) that are detected in only a single sample before performing alpha and beta diversity analyses?
Thank you!
-Brandon
Hello!
In general, I remove all the ASVs that were counted in total (in all samples) less than 10 times and that were found in less than 3 of the samples, but you can adjust it based on the questions you are trying to answer, experimental design, and number of samples per group. It reduces “noise“ and speeds up the analyses.
Best,
Thank you for the suggestions. Are there any benchmarks on this? I’m wondering whether removing rare ASVs would influence the alpha and beta diversity calculations.
We did a large benchmark of this years ago showing impacts of filtering on both alpha and beta diversity (only abundance filtering and quality control parameters, don’t think we did prevalence filtering):
https://www.nature.com/articles/nmeth.2276
This and the paper that @colinbrislawn shared above are both using OTUs, not ASVs, so precise filtering thresholds will vary (as dada2 is already filtering out rare and erroneous reads during denoising), just keep that in mind for interpretation. But the impacts on alpha and beta diversity will generalize. It also largely depends on the diversity metrics that you use… unweighted metrics will be much more sensitive to rare features.