When building concept graphs, one must ensure that every graph for every concept is not empty. For example if one builds a graph with only Vessel cells, it is possible that some samples do not contain any cells of this type, therefore when building a concept graph under this concept the graph for this sample would be empty. The GCL workflow takes care of ensuring this through the sub-workflow filter_samples, by filtering samples that have less than N cells in any given concept (N is user-defined).
On the other hand when developing the ATHENA graph concept-building functionality I put in place a check that ensures that every label that you have in include_labels must be present in the filtering column base_graph_builder.py:65 (to avoid undefined behavior). However, this check is restrictive since one can potentially have cell types that are not present in a sample but it still constitutes a valid sample because it has enough cells from the other types.
Because of this, I removed this check. Actually, I just updated it. Now it checks that at least one of the labels in include_lables matches the labels present in filter_column.
When building concept graphs, one must ensure that every graph for every concept is not empty. For example if one builds a graph with only Vessel cells, it is possible that some samples do not contain any cells of this type, therefore when building a concept graph under this concept the graph for this sample would be empty. The GCL workflow takes care of ensuring this through the sub-workflow
filter_samples, by filtering samples that have less than N cells in any given concept (N is user-defined).On the other hand when developing the ATHENA graph concept-building functionality I put in place a check that ensures that every label that you have in
include_labelsmust be present in the filtering columnbase_graph_builder.py:65(to avoid undefined behavior). However, this check is restrictive since one can potentially have cell types that are not present in a sample but it still constitutes a valid sample because it has enough cells from the other types.Because of this, I removed this check. Actually, I just updated it. Now it checks that at least one of the labels in
include_lablesmatches the labels present infilter_column.