Open JPReceveur opened 2 months ago
Error I am seeing from the gEAR logs is that it is actually missing the "louvain" clustering column. Looking at the dataset in the Python REPL, I see several columns (seurat_clusters, cluster_label, cell_type) that you could just copy to make the louvain adata.obs column
I though the primary analysis worked off of more than just 'louvain'. Do you need 'louvain' for it to work correctly? I was under the assumption primary analysis could work off of other columns as well. e.g. a primary analysis can get assigned based off cell_type in the make primary analysis script https://github.com/IGS/gEAR/blob/main/bin/add_primary_analyses_to_datasets.py
After clicking "Primary Analysis", the get_stored_analysis.cgi
script says that louvain was not calculated ("false), so this lines up with the add_primary_analysis_to_datasets.py
script. The h5ad_find_marker_genes.cgi
script does require the "louvain" column to be present
You are right that the "cell_type" cluster should have been detected, so I'm not sure why the h5ad and the pipeline.json files did not get updated to have the louvain analysis. Would it make sense just to rerun it?
There is a profile where the user has used a different coordinate system in their paper and gEAR dataset (SWNE instead of tsne /UMAP). In the single cell workbench, it has a primary analysis available but using it results in an error. I'm guessing its because neither tsne or umap coordinates are being found. Shaun could you take a look to see if the marker gene step can work with SWNE coordinates?
Dataset ID: 2b71927b-8dc2-66ff-bc2a-103532196c07 Profile: https://umgear.org/index.html?multigene_plots=0&layout_id=human-utricle-sc-atlas&gene_symbol_exact_match=1&gene_symbol=sox2
All the datasets in the profile should be public