TDR vs TAR Bokulich et al 2018

taxonomy
feature-classifier

(Rocio) #1

Hi all,
We are facing some problems when dealing with the calculation of TDR (taxon detection rate) and TAR (taxon accuracy rate) definitions:

I am posting here the definitions taken from Bokulich et al., 2018:

TAR = TP/(TP + FP) or the fraction of observed taxa that were expected at level L.
TDR = TP/(TP + FN) or the fraction of expected taxa that are observed at level L.

For our experimental conditions we calculated these parameters both manually and using QIIME2. The problem we found is that we obtained the same but oposite results with the 2 strategies:

Our data:

Nº of detected species: 77
Nº of expected species: 20
FP=69
TP=8
FN=12

Then:
TAR=TP/(TP+FP)=8/(8+69)=0.10
TDR=TP/(TP+FN)= 8/(8+12)=0.40

Data reported by QIIME2 with the same experiment:
TAR=0.4
TDR=0.1
Just the opposite!

Just now we cannot find where the problem is. Could you please give us some input? Are we applying the defitions in a correct way.

Thanks in advance!

Regards,

Rocio Gonzalez-Soltero


Incorrect results produced by q2-quality-control evaluate-composition
(Nicholas Bokulich) #2

(Nicholas Bokulich) #3

Thanks for finding this @Rocio! Looks like you discovered a bug in q2-quality-control.

I have raised an issue here — will fix asap but this will not make it in until the next release (August). We will post back here when this is fixed in the release version.

Thank you for reporting!


(Nicholas Bokulich) #4

(Rocio) #5

Thank you for the reply! We will wait for the next release!
Regards,
Rocio


(Nicholas Bokulich) #6

Thanks again @Rocio for finding and describing this bug! Note that we have issued an updated release — see this announcement

No need to wait for the next release — just re-download the latest release and the bug is fixed.

Thank you!