Also, if you look at the DADA2 stats file you see that for many of the samples there is a significant drop in reads following the filtering and merging step. Not so many post the chimera removal step.
