why is precision so low?

[gregcaporaso]

related to #62?

[gregcaporaso]

Likely contributors:

• all comparisons are at the genus level
• we haven't filtered singletons

[gregcaporaso]

After some investigation on this, it looks like there are a few issues:

1. We're counting assignment as Other as a false positive. We shouldn't, as it effectively means that the classifier didn't make an assignment at that level. That's not a precision hit, but rather should be a recall hit if the group is never identified.
2. After taking care of (1), most of the remaining false positions are low abundance (relative abundance of 1e-5 or lower). So we need to do some low abundance OTU filtering.
3. There are some annotation errors. The Broad mock communities list k__Bacteria;p__Thermi;c__Deinococci;o__Deinococcales, but the RDP classifier identifies k__Bacteria;p__[Thermi];c__Deinococci;o__Deinococcales (note the square brackets around Thermi).

[gregcaporaso]

Another error detected in Turnbaugh mock communities. These are listed as containing k__Bacteria;p__Firmicutes;c__Coriobacteriia;o__Coriobacteriales, but uclust assigns k__Bacteria;p__Actinobacteria;c__Coriobacteriia;o__Coriobacteriales.

Remaining false positives are low abundance (all <= 2e-4 relative abundance).

[gregcaporaso]

These issues have all been addressed.