I used the genebank for taxonomic classification and some samples came out with a confidence level of 70%. Wouuld this be acceptable? Would you recommend trying Silva?
Is this the confidence level of the
classify-sklearn method in q2-feature-classifier?
If so, then yes that is acceptable. This confidence threshold is what the classifier uses to decide if classification at a specific rank is accurate enough to apply a label. So you can interpret that as “this sequence almost certainly belongs to genus X, and we are 70% confident it is this species”. This 70% threshold was optimized through a series of tests, you can read more about this method here:
And by the way you can always modify the confidence threshold with the
--p-confidence parameter in
classify-sklearn… but read the paper first to understand what that will do to your results!
This topic was automatically closed 31 days after the last reply. New replies are no longer allowed.