PFOCR result-augmentation: adjust logic

[AlexanderPico]

Per discussion notes and Slack thread with Gus, we have consensus on the following plans:

1. Algorithm - As a general policy/logic, we will identify PFOCR pathway figures that contain all supported-category nodes in the node-bindings (2 in the case of MVP1/2; 3 in the case of Pathfinder). And, obviously, we skip case where there are fewer than 2 supported-category nodes. After passing this minimal requirement, we then include any additional supported-category nodes from the result's aux_graph in the calculation of an enrichment score. This way all results are relevant to the full context of query and are scored to reflect the inclusion of aux_graph nodes as well.

• NOTE: Per this issue, we are expanding the "supported-category nodes" to include Chemicals and Diseases in addition to Genes.

2. Output location - The PFOCR hits can continue to go into message/results/analyses. No change.

3. Structure - Per this issue, we want to include two URLs per hit, like so...

"pfocr": [
    {"matchedCuries": [
         "NCBIGene:8445",
         "NCBIGene:1859",
         "NCBIGene:2932",
         "NCBIGene:2735"
       ],
      "score": 0.2352941176470588,
      "pmc": "PMC2743241",
      "figureUrl": "https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2743241/bin/nihms-104435-f0001.jpg",
      "pfocrUrl": "https://pfocr.wikipathways.org/figures/PMC5463358__fnagi-09-00176-g003.html"
    },

[colleenXu]

@AlexanderPico

My understanding is that only point 1 has work that needs doing. And Points 2 + 3 are already done (@tokebe to confirm?)

Work to do:

• implementing new general logic/policy: "all that can match" must match, min 2
• adding chem and disease support. includes mesh/chebi (chem only)/doid (disease only) support

[colleenXu]

@AlexanderPico

And I found your "point 1 algorithm wording" a little confusing. Here's my understanding in case it differs from yours...pending @tokebe clarifying things for both of us.

• Pathfinder results can have 2 or 3 "supported-category" Nodes in node-bindings; both cases should trigger PFOCR result-augmentation.
• I think from the POV of the "result": a result must have >= 2 "supported-category" Nodes in its node-bindings to trigger PFOCR result-augmentation. Then, all of those nodes (node-binding, supported-category) must be present in a PFOCR pathway figure for that figure to be kept as a match.
  • MVP1/2 results only have 2 Nodes/node-bindings total so both need to be a "supported-category" for trigger, need to be in figure for match
  • Pathfinder results have 3 Nodes/node-bindings total. If only 2 are supported-category, then both need to be in figure. If all 3 are supported-category, then all 3 need to be in figure.
• additional nodes from aux-graphs are included in the matchingCuries section and accounted for in the enrichment scoring.

[tokebe]

RE @colleenXu's comments:

• Yes, only point 1 still needs work, and does still need work to implement "All that can match, must match, minimum 2". Currently, only the minimum is respected.
• We'll use the other issue to track adding more node type support.
• Thanks for the clarification, I see no discrepancy between @AlexanderPico's explanation and yours -- yours just gets a little more into implementation procedure and examples.

[colleenXu]

@tokebe @AlexanderPico

I've adjusted the titles of this issue and https://github.com/biothings/biothings_explorer/issues/847 to make it clearer what work is being tracked in each.

I think this is correct but feel free to change it if it's not correct.