Different clustering methods

Have you considered adding more options for clustering into otus? For over 100 000 unique sequences (sequenced with Illumina), Mothur generates over 4600 otus. Mostly these are singletons. This recent article shows that singletons plus doubletons take over 50% of the total number of tags (doi:10.1371/journal.pone.0030230). Many articles have blamed MSA and suggests pairwise alignment. And I tend to agree with them if you have millions of short sequences (not a problem with 454, but common for Illumina).

Some programs I have tested so far in the order of accuracy (in my opinion):
Two-Stage clustering http://hwzhoulab.smu.edu.cn/paperdata/
Crunchclust http://code.google.com/p/crunchclust/
CROP http://code.google.com/p/crop-tingchenlab/ (maybe even the most accurate, but not efficient with large datasets)

And there are even more (e.g. ESPRIT, CD-HIT)

These other methods are generally poor substitutes (see our AEM 2011 paper on clustering). Generally, the problem is the sequencing error associated with the platforms artificially increasing the number of unique reads.