Video talks on 16S data analysis posted.

URMAP ultra-fast read mapper posted (paper).

~20% of taxonomy annotations in SILVA and Greengenes are wrong (paper).

Taxonomy prediction is <50% accurate for 16S V4 sequences (paper).

97% OTU threshold is wrong for species, should be 99% for full-length 16S, 100% V4 (paper).

Matthews Correlation Coefficient (MCC) metric for OTU clustering

Westcott and Schloss define the Matthews Correlation Coefficient (MCC) for OTUs as follows.

The variables are (pairs = pairs of sequences from the input data):

TP = number of pairs in the same cluster which have >=97% identity
TN = number of pairs in different clusters which have <97% identity
FP = number of pairs in the same cluster which have >97% identity
FN = number of pairs in different clusters which have >=97% identity

In general, it is not possible to construct error-free OTUs as defined by MCC, and in some simple cases MCC is undefined and fails to identify the best clusters.