Handling ITS reads with variable lengths: Should ASVs be normalized to avoid redundancy?

[CarolTMartins]

Hello all,

I am working with ITS amplicon data, which naturally contains fragments of variable lengths. I noticed that in my dataset, some ASVs have identical sequences to others but are shorter versions of the same sequence. This raised a question about how DADA2 handles ITS length variability.

Given this, I would like to ask:

Should ITS sequences be length-standardized (trimmed or padded) to avoid generating redundant ASVs that are biologically the same but differ only in length?
Or does DADA2 internally handle such cases so that shorter fragments do not artificially inflate ASV diversity?

In our case, the samples consist almost exclusively of Saccharomyces cerevisiae yeast, with extremely low diversity. Our goal is to detect very small differences (ideally down to the strain level). We are concerned that the presence of ITS fragments of different lengths might affect our ability to resolve fine-scale variation, especially given the biology of our samples.

Any guidance on best practices for ITS processing with DADA2 in such low-diversity, strain-focused datasets would be greatly appreciated.

Thank you very much!

[benjjneb]

Internally the DADA2 algorithm will separate treat sequences differing only by length as distinct "unique sequences". These can be joined into the same ASV, and usually will be when occurring in the same sample and if the length difference is small. But if the length difference is larger or the length-variants occur in different samples, they can end up as different ASVs in the output. Thus it is recommended not to create non-biological variation in your data processing. For example, do not truncate based on a sliding window quality score.

There is a function implemented in the DADA2 R package to deal with the issue of length variation in an ASV table: collapseNoMismatch. This will collapse together all ASVs that have no mismatches between the overlapping portions of the sequences. That may work for your application.