Support co annotation analysis #9
Conversation
pandasaurus_cxg/anndata_analyzer.py
Outdated
| lambda row: Predicate.CLUSTER_MATCHES.value | ||
| if row[text] in predicate_dict.get(row["cell_type"], []) | ||
| and len(predicate_dict.get(row["cell_type"], [])) == 1 | ||
| else ( |
There was a problem hiding this comment.
Bit opaque. I think this works, but looks like this only tests co-annotation with cell_type field. Should be looking at all fields.
There was a problem hiding this comment.
All fields as in tissue, diseases and organism etc. or just all other free text cell type field?
There was a problem hiding this comment.
All combinations of pairs free text and ontology cell type fields.
pandasaurus_cxg/anndata_analyzer.py
Outdated
| and len(predicate_dict.get(row["cell_type"], [])) == 1 | ||
| else ( | ||
| Predicate.SUBCLUSTER_OF.value | ||
| if row[text] in predicate_dict.get(row["cell_type"], []) |
There was a problem hiding this comment.
Based on the if/else logic - this looks like adds SUBCLUSTEROF if there are multiple rows with "cell_type" and text. But that could be a subcluster_of relationship in either direction or overlap.
There was a problem hiding this comment.
Let me try to explain my reasoning behind this;
Lets say we have the following structure in predicate_dict:
{
'endothelial cell': ['Descending Vasa Recta Endothelial Cell', 'Ascending Vasa Recta Endothelial Cell', 'Afferent / Efferent Arteriole Endothelial Cell', 'Peritubular Capilary Endothelial Cell ', 'Glomerular Capillary Endothelial Cell', 'Degenerative Peritubular Capilary Endothelial Cell', 'Cycling Endothelial Cell', 'Lymphatic Endothelial Cell', 'Degenerative Endothelial Cell'],
'podocyte': ['Podocyte', 'Degenerative Podocyte'],
'leukocyte': ['Natural Killer Cell / Natural Killer T Cell', 'M2 Macrophage', 'Neutrophil', 'Monocyte-derived Cell', 'T Cell', 'Plasma Cell', 'Cycling Mononuclear Phagocyte', 'Non-classical Monocyte', 'Classical Dendritic Cell', 'Mast Cell', 'B Cell', 'Plasmacytoid Dendritic Cell', 'Cycling Natural Killer Cell / Natural Killer T Cell']
}
I iterate through the df I have and lets say the first row[text] is 'Descending Vasa Recta Endothelial Cell' and it corresponds to 'endothelial cell' in the cell type field. I check if 'Descending Vasa Recta Endothelial Cell' is in ['Descending Vasa Recta Endothelial Cell', 'Ascending Vasa Recta Endothelial Cell', 'Afferent / Efferent Arteriole Endothelial Cell', 'Peritubular Capilary Endothelial Cell ', 'Glomerular Capillary Endothelial Cell', 'Degenerative Peritubular Capilary Endothelial Cell', 'Cycling Endothelial Cell', 'Lymphatic Endothelial Cell', 'Degenerative Endothelial Cell']. Since the length of the list is not 1 I infer 'Descending Vasa Recta Endothelial Cell' as subcluster_of 'endothelial cell'
Is there any way to determine the direction of this relationship with the tabular data?
I assumed that everything other than cluster_matches and subcluster_of should be cluster_overlaps, not sure for 100%.
pandasaurus_cxg/anndata_analyzer.py
Outdated
| if row[text] in predicate_dict.get(row["cell_type"], []) | ||
| else Predicate.CLUSTER_OVERLAPS.value | ||
| ), # All the other cases should be marked with 'cluster_overlaps', right? | ||
| axis=1, |
There was a problem hiding this comment.
I think this is wrong. I will add a cluster_overlaps example to the ticket.
|
Hi @dosumis, examples for each predicate group can be seen below;
|
fixes #7
I'm not clear about how to choose the predicates mentioned in the ticket;
I've added the table without the predicates.