A small suggestion on the heatmaps generated by gneiss dendrogram heatmaps. When processing subsets of my data (filtering with feature-table filter-samples) and provided with my complete metadata file, the generated heatmaps contain large areas of white space:
I suggest that it would be beneficial if the generated heatmaps automatically only contained space for the samples within the FeatureTable[Composition] as opposed to for all samples within the metadata file.
Hi @five. Thanks for reporting this – this would be certainly advantageous to have resolved. Could you specify the input files / command used to generate this plot?
I think I found a corresponding open issue on the q2-gneiss issue tracker. It sounds like if the sample metadata is a superset of the samples in the table, those extra metadata samples can be filtered out by q2-gneiss – this is generally how we’re handling metadata supersets in QIIME 2 (QIIME 1 often behaved this way too). @gregcaporaso is working on adding an API on the QIIME 2 Metadata object to help with this so it should be an easy fix.
We’ll follow up here when it’s fixed in an upcoming release!
Hello @mortonjt, as @jairideout described on the issue tracker, the metadata for my first plot (upper in original post), is a superset of the samples used in the analysis.
The commands used to generate the two plots followed the form of:
With white space
qiime gneiss dendrogram-heatmap \
--i-table Data/Table.qza \
--i-tree Data/correlated_hierarchy.qza \
--m-metadata-file Data/metadata.tsv \
--m-metadata-category Treatment \
--p-method log \
--o-visualization Data/treatment_heatmap_log.qzv
Without white space
qiime gneiss dendrogram-heatmap \
--i-table Data/Table.qza \
--i-tree Data/correlated_hierarchy.qza \
--m-metadata-file Data/alt_metadata.txt \
--m-metadata-category Treatment \
--p-method log \
--o-visualization Data/altmeta_treatment_heatmap_log.qzv
The only difference between generating the plots was the provision of a different metadata file, which I reduced (outside of Qiime2) to only the relevant samples. The other inputs (table, correlated hierarchy) were the identical.
