Hi
qiime sample-classifier classify-samples was run on table (dada2 pipeline, ASV counts) with 167 samples belonging to 4 groups. I would appreciate if you help me with following questions
Heatmap:
Is log10 frequency derived from abundance or relative abundance of ASVs?
Is there a table corresponding to heatmap? update: qiime sample-classifier heatmap generates a table of abundance of "important-features" without taxonomy. Is it possible get a table with log10 transformed values (as used in heatmap) with taxonomy?
Heatmap is labelled with ASV ids. Is it possible to change it to taxonomy? update: I was able to add taxonomy using qiime sample-classifier heatmap.
Important features:
4. feature_importance.qzv has a list of 69 features, with importance values ranging from 0.062 to 0. Why feature with 0 value included in the list?
5. Samples belong to 4 groups. Is it possible to determine which feature set is associated with each group?
Model Accuracy
6. Overall accuracy is 0.558824. Accuracy ratio is 1.1875. Are these acceptable numbers? Predictions of this model are 80% accurate with a 2nd set of samples (these samples were not involved in training).
It looks like this is generated from abundance. However I would note that they add a psuedocount of 1 before normalizing with log10.
There is no way to tranform feature tables in QIIME 2.
We report every feature's Importance, Some of them aren't important to the model hence the 0.
Looking at the heatmap should tell you what features are most abundant in your groups of interest. but as far as I know there is no automatic feature-set creation.
I can't really tell you how to interepret or use your data.I personally wouldn't consider the model being correct 55% of the time a effective model. I personally look for an accuray result over 1.5.
80% sounds better but I am not sure what the Accuracy ratio was for that run.