Removing contaminants from samples

[msnyder424]

This program is great!

OK, I have identified which sequences are likely contaminants in my samples with the frequency, prevalence, and both methods. I now have a list of sequences and whether or not they are contaminants. I wrote a script to remove any contaminants from my sequence table output from dada2. But what if there is some chance that these sequences are not always present in a sample due to contamination? I'm not sure if in this case, they would be common enough in those samples to not be identified with the frequency method. They may still be identified with the prevalence method. But the contaminant sequences in negative controls have to originate from somewhere, right? Bottom line is I don't want to eliminate them from samples that they actually were in. I hope this is not a dumb question...

Thoughts?

Thanks, Matt

[benjjneb]

But what if there is some chance that these sequences are not always present in a sample due to contamination?

There is some non-zero possibility that the contaminants from your reagents or wherever and the true taxa in the samples overlap. Typically these will not be identified as contaminants by the decontam methods, and in fact decontam was developed in part to avoid removing legitimate taxa that show up in negative controls due to cross-contamination. However, to further reduce the likelihood of this sort of overlap, we recommend working at more resolved levels for contaminant removeal, i.e. ASVs instead of OTUs or species instead of genus, as overlap is less likely at more resolved levels.

Anecdotally, I have not encountered this being a problem with the decontam-classified contaminants in my experience, but it can't be ruled out in all cases. And if in doubt for certain taxa, manual inspection is a great way to convince yourself one way or the other!