Hi @SamWilliamsUOB,
Welcome to the forum, and that's an excellent question! Agreed, that using older methods such as OTU clustering with a phred score of 2 would have been risky at best, however unlike previous methods where the phred score was simply used as a cutoff threshold, dada2 is inferring and correcting reads based not just on the quality scores but also the abundance of a feature, which are incorporated into its error-model. This allows it to infer highly accurate calls even in instances where a phred score is not great, assuming that other copies of that read exist with higher quality scores. In addition, keep in mind that with dada2 we are still excluding low quality nts by trimming low quality tails using --p-trim and --p-trunc and in these situations we do use more familiar q-score threshold such as those we saw with quality score based filtering methods prior to OTU clustering. For example, I generally trim all my sequences to instances where the median quality score seems to drop below 20-25.
As for why the default socre is 2 in dada2, this appears to have had some special consideration as per this discussion here.
All that said, you are of course encouraged to play around with these parameters to select what works best for you, you could try increasing the default score and see how it compares. I would certainly be interested in that comparison so if you do feel free to share your results here! My guess is you would see some minor differences, with higher qscore value giving you less reads and slightly less alpha diversity.
5 Likes