Hi,

I´m running a 16S experiment but have some strange results when doing quality control. Note that data load into DESeq has been filtered (adapter trimming , fastqc, removal of contaminants and phiX, chimeras..). I have observed wrong spearman correlations between tehcnical replicates so my assumption is that the experiment did not work well, at least would not be reporducible given the weak correlation between some technical replicates. There are more than 50 samples, some with tehcnical replicates , some others without.

I´m trying to use DESeq2 to do some quality control and would like to have your advice on the data.

Here below the dispersion graph.

Below the density of counts

Cooks distances

The data does not look good to me, any advice ??

I don't have much specific advice with respect to the use of DESeq2 with 16S data because this is not my area of expertise. The dispersion plot seems fine. Note that the "poscounts" type of size factor estimation seems to perform better than the default. See ?DESeq or ?estimateSizeFactors.

You might instead use Bray-Curtis beta diversity as a measure of dissimilarity, available in the phyloseq package. I could imagine differences in read depth, or a lot of low counts if there are only a couple dominant genera, resulting in low Spearman correlation even if the data are OK.