Feature Request: enriched ontologies in JSON format

[rossbar]

Just following up on the conversation from the all-hands meeting!

In an effort to expand our support for ontology-mapped cell types in deepcell-types, we're in the process of trying to expand the existing ontology-mapping to other tissue types etc.

Our general workflow is as follows:

• Scrape the JSON-LD files from asct+b site
• Use the CL lookup tool to find a "parent" for the cell types that makes the most sense in the context of defining a hierarchy for cell-type prediction/labeling.
• Add this info to the ontology-mapping (linked above)

One thing we discussed at the meeting was that the JSON-LD was probably not the appropriate resource for this, as it doesn't capture the "is-a" cell-parent relationships (by design). To be clear, we started with the JSON files only because they're the easiest to work with programmatically in terms of language support, human-readability etc.

So - in the longer term, it'd be great if the full "enriched"¹ ontology were also made available in JSON format. This is by no means a blocker nor a critical feature, just something that would make consuming ontology info slight easier for (at least one) downstream users!

Footnotes

1. apologies if I'm using the wrong terminology! ↩

[bherr2]

Could you give me an example of what you're doing? I think I could create a little API to do the lookup for you and return .csv but I'm missing some of the context of what step 1 is for. Thanks, Bruce

…

-- Bruce W. Herr II Technical Director CI for Network Science Center, http://cns.iu.edu<http://cns.iu.edu/> Please note that my email address is changing to ***@***.*** When the transition is complete, you will no longer be able to reach me at my old address ***@***.***). Please begin using my new address ***@***.*** now.

________________________________ From: Ross Barnowski ***@***.***> Sent: Monday, April 28, 2025 2:06 PM To: hubmapconsortium/hra-kg ***@***.***> Cc: Subscribed ***@***.***> Subject: [hubmapconsortium/hra-kg] Feature Request: enriched ontologies in JSON format (Issue #25) [https://avatars.githubusercontent.com/u/1268991?s=20&v=4]rossbar created an issue (hubmapconsortium/hra-kg#25)<#25> Just following up on the conversation from the all-hands meeting! In an effort to expand our support for ontology-mapped cell types in deepcell-types, we're in the process of trying to expand the existing ontology-mapping<https://github.com/hubmapconsortium/deepcelltypes-hubmap/blob/main/model/hubmap_ontology.yaml> to other tissue types etc. Our general workflow is as follows: * Scrape the JSON-LD files from asct+b site<https://humanatlas.io/asctb-tables> * Use the CL lookup tool<https://www.ebi.ac.uk/ols4/ontologies/cl> to find a "parent" for the cell types that makes the most sense in the context of defining a hierarchy for cell-type prediction/labeling. * Add this info to the ontology-mapping (linked above) One thing we discussed at the meeting was that the JSON-LD was probably not the appropriate resource for this, as it doesn't capture the "is-a" cell-parent relationships (by design). To be clear, we started with the JSON files only because they're the easiest to work with programmatically in terms of language support, human-readability etc. So - in the longer term, it'd be great if the full "enriched"1 ontology were also made available in JSON format. This is by no means a blocker nor a critical feature, just something that would make consuming ontology info slight easier for (at least one) downstream users! Footnotes 1. apologies if I'm using the wrong terminology! ↩ — Reply to this email directly, view it on GitHub<#25>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AIEE3I3KGY5GIVKI55CSKUL23ZUZ5AVCNFSM6AAAAAB4BBNEN2VHI2DSMVQWIX3LMV43ASLTON2WKOZTGAZDKNZTGI4DGNI>. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[rossbar]

Sure, we're doing something like:

import json                                                                     
import networkx as nx                                                           
                                                                                
# Where skin_graph.json is the downloaded JSON-LD for skin tissue               
# from the ASCT+B site                                                          
with open("skin_graph.json") as fh:                                             
    data = json.loads(fh.read())                                                
ctd = data["data"]["cell_types"]                                                
                                                                                
G = nx.DiGraph()                                                                
for ct in ctd:                                                                  
    G.add_node(ct["id"], name=ct["ccf_pref_label"])                             
for ct in ctd:                                                                  
    isa = ct.get("ccf_ct_isa", None)                                            
    if isa is not None:                                                         
        for parent in isa:                                                      
            G.add_edge(parent, ct["id"])

Which gives:

# Simple viz                                                                    
import matplotlib.pyplot as plt                                                 
pos = nx.nx_agraph.graphviz_layout(G, prog="twopi")                             
nx.draw(G, pos=pos, with_labels=False);                                         
nx.draw_networkx_labels(G, pos=pos, labels=dict(G.nodes(data="name")))          
plt.show()

@johardi already explained to us at the last meeting that the JSON-LD's are not expected to have the hierarchical information¹. In order to get the hierarchical info, it sounds like we'd need to replace the first chunk with parsing the RDF/XML files instead, which we can certainly do. For our use-case however, JSON is generally more convenient.

Footnotes

1. though curiously there is one instance of hierarchical relationship captured between Keratin (Onychocyte) and Keratin (Granular). ↩

[bherr2]

Ok, I can create a SPARQL query that can get you results. It sounds like given an ASCT+B table, you want all the cells in the ASCT+B plus their parents. I could return a table like this:

| table | cell_id | cell_label | parent_id | parent_label |

Would that get you what you need? Or what other information to do you need?