Generate graphs using cell sets as unifying concept #24
Conversation
| def visualize_rdf_graph(self): | ||
| nx_graph = rdflib_to_networkx_multidigraph(self.graph) | ||
| # Plot Networkx instance of RDF Graph | ||
| pos = nx.spring_layout(nx_graph, scale=2, k=2) | ||
| edge_labels = nx.get_edge_attributes(nx_graph, "r") | ||
| nx.draw_networkx_edge_labels(nx_graph, pos, edge_labels=edge_labels) | ||
| nx.draw(nx_graph, with_labels=True) | ||
| plt.show() |
There was a problem hiding this comment.
This is just a placeholder; I have used OBASK to visualize graphs and examine them for validation purposes.
There was a problem hiding this comment.
This is a place where oaklib could really help. Worth talking to @anitacaron about how she uses it for visualising validation.
There was a problem hiding this comment.
I had a conversation with Anita yesterday in the office. From what I gathered, it seems that visualising the relations and neighbours of a set of terms is necessary. However, I am unsure if this is something we actually need.
There was a problem hiding this comment.
I didn't know the context. I can have a closer look into oaklib.
There was a problem hiding this comment.
There's the OboGraph Interface
| cl_namespace = Namespace("http://purl.obolibrary.org/obo/CL_") | ||
| for curie, label in self.cell_type_dict.items(): | ||
| resource = cl_namespace[curie.split(":")[-1]] | ||
| self.graph.add((resource, RDFS.label, Literal(label))) | ||
| for s, _, _ in self.graph.triples((None, self.ns["cell_type"], Literal(label))): | ||
| self.graph.add((s, self.ns["consists_of"], resource)) | ||
| # add subClassOf between terms in CL enrichment | ||
| for _, row in self.enriched_df.iterrows(): | ||
| for s, _, _ in self.graph.triples((None, RDFS.label, Literal(row["s_label"]))): | ||
| for o, _, _ in self.graph.triples((None, RDFS.label, Literal(row["o_label"]))): | ||
| self.graph.add((s, RDFS.subClassOf, o)) |
There was a problem hiding this comment.
I'm not entirely certain if any methods, other than simple_enrichment, contribute additional information to the graph. This is because those methods may involve CL terms from a subset and context that are not utilized in annotations.
We should talk about this.
There was a problem hiding this comment.
By definition all enrichment methods link to terms related to those used in annotation.
Looking at this again, I think it is clear that I have underspecified. I think the challenge is how we deal with flattening. In the pipelines Anita has worked on we use ROBOT or Souffle to strip redundancy from the flattened graph. If we're sticking with pure python we will need something similar to the the Souffle redundancy stripping algo here - which will require some thought.
For this PR I'd suggest an MVP of building a graph based on co-annotation first. We can then move folding in the enrichment graph to a second ticket/PR.
ubyndr
force-pushed
the
13-generate-graphs-using-cell-sets-as-unifying-concept
branch
from
a221a81 to
bb04f1e
Compare
* Merged from main * Updated anndata_analyzer.py * Removed state and state.l2 from free-text annotations * Refactored visualize_rdf_graph method * Refactored save_rdf_graph, visualize_rdf_graph method and added transitive_reduction method * Format changes in co_annotation_report * Added state and state.l2 to free-text annotations
This reverts commit 625b237.
ubyndr
deleted the
13-generate-graphs-using-cell-sets-as-unifying-concept
branch
Resolves #26
TODO:
Currently, we are using enriched_df to add cell type terms to the graph. However, we have noticed that if a cell type does not have any subClassOf relations with other cell types, those terms are missing from the graph. To address this issue, it would be better to utilize the co_annotation report for adding the cell type terms.I use the obs attribute in the anndata object to generate a cell type dictionary. This dictionary consists of cell type IDs and labels, which I then utilize to add the cell type terms to the graph. Then, we can use enriched_df specifically to incorporate the subClassOf relations between those terms.The root cause of the missing cell terms in the neo4j UI is attributed to the way I currently add the cell terms. To address this issue, I will be making updates to the enrich_rdf_graph enrich_rdf_graph method