Hi,

I really enjoyed your article! I was always hesitant to use estimators that rely on the doubleton / singleton ratio, knowing that the singleton frequency may be completely exaggerated. An earlier suggestion on how to deal with the inflation of spurious sequences comes from Robert Edgar, the author of USEARCH : he suggest to exclude global singletons before the clustering and subsequently assign the singleton sequences to the predefined clusters (OTUs). In my dataset this reduces the number of OTUs from 9000 (6000 of which are singletons) to 1000 (with no singletons at all, by design). Among those 8000 OTUs, many but certainly not all where sequencing errors. I guess my question is, should singletons be excluded or are they needed to estimate diversity?