Handler for BioThings API providing graph type data

[kevinxin90]

Example Graph representation:

{
    "subject": {
        "id": "MONDO:000123",
        "type": "Disease"
    },
    "object": {
        "id": "NCBIGene:1017",
        "type": "Gene",
        "taxid": "9606"
    },
    "association": {
        "predicate": "negatively_regulates",
        "publications": ["PMID:123", "PMID:124"]
    }
}

Above output could be represented in another way by switching the subject & object and reverse the predicate, e.g.

{
    "subject": {
        "id": "NCBIGene:1017",
        "type": "Gene",
        "taxid": "9606"
    },
    "object": {
        "id": "MONDO:000123",
        "type": "Disease"
    },
    "association": {
        "predicate": "negatively_regulated_by",
        "publications": ["PMID:123", "PMID:124"]
    }
}

So if the user provides the following query

biothings.ncats.io/api1/query? \
    subject.id:"MONDO:000123" AND \
    object.id:"NCBIGene:1017" AND \
    association.predicate:"negatively_regulates"

It should be translated into two queries

1. same as user query

biothings.ncats.io/api1/query? \
    subject.id:"MONDO:000123" AND \
    object.id:"NCBIGene:1017" AND \
    association.predicate:"negatively_regulates"

2. reverse it

biothings.ncats.io/api1/query? \
    object.id:"MONDO:000123" AND \
    subbject.id:"NCBIGene:1017" AND \
    association.predicate:"negatively_regulated_by"

And the response from the 2nd query should also be reversed and merge with the first query.

In summary:

1. translate user query into two queries (one original, one reverse query)
2. For the reverse query,

• all fields starting with object (e.g. object.id) should be replaced with subject
• all fields starting with subject (e.g. subject.id) should be replaced with object
• reverse the value association.predicate based on a mapping file

3. For the reverse query result

• change root key object into subject
• change root key subject into object
• reverse the value of association.predicate based on a mapping file

4. merge the results from two queries.

We have an API set up providing graph type data for testing: https://biothings.ncats.io/biggim

[kevinxin90]

A good test case is:
Currently https://biothings.ncats.io/biggim/query?q=subject.NCBIGene:6494%20AND%20object.NCBIGene:1956%20AND%20association.context.disease.id:%22MONDO:0006046%22 returns 1 hit.

But https://biothings.ncats.io/biggim/query?q=object.NCBIGene:6494%20AND%20subject.NCBIGene:1956%20AND%20association.context.disease.id:%22MONDO:0006046%22 doesn't.

If the feature is implemented correctly, the second query should be able to return 1 hit which is a reverse version of the 1 query.

[kevinxin90]

The predicate mapping file is available here: https://github.com/biothings/pending.api/blob/master/predicate_mapping.json. But feel free to suggest other places which is more appropriate to put it. The key of the dictionary is the predicate, and the value is the reverse of the predicate.

[namespacestd0]

@kevinxin90 I wanna go over this initial implementation with you before I refine the details, please check if this is the intended behavior:

For a query to http://localhost:8000/biggim/query/graph:

{
    "subject": {
        "NCBIGene": 1956
    },
    "object": {
        "NCBIGene": 6494
    },
    "association": {
        "context": {
            "disease":{
                "id": "MONDO:0006046"
            }
        }
    }
}

It is translated to:

  {
      "query": {
        "bool": {
          "should": [
            {
              "bool": {
                "must": [
                  {
                    "multi_match": {
                      "fields": "subject.NCBIGene",
                      "lenient": true,
                      "operator": "and",
                      "query": 1956
                    }
                  },
                  {
                    "multi_match": {
                      "fields": "object.NCBIGene",
                      "lenient": true,
                      "operator": "and",
                      "query": 6494
                    }
                  },
                  {
                    "multi_match": {
                      "fields": "association.context.disease.id",
                      "lenient": true,
                      "operator": "and",
                      "query": "MONDO:0006046"
                    }
                  }
                ]
              }
            },
            {
              "bool": {
                "must": [
                  {
                    "multi_match": {
                      "fields": "association.context.disease.id",
                      "lenient": true,
                      "operator": "and",
                      "query": "MONDO:0006046"
                    }
                  },
                  {
                    "multi_match": {
                      "fields": "object.NCBIGene",
                      "lenient": true,
                      "operator": "and",
                      "query": 1956
                    }
                  },
                  {
                    "multi_match": {
                      "fields": "subject.NCBIGene",
                      "lenient": true,
                      "operator": "and",
                      "query": 6494
                    }
                  }
                ]
              }
            }
          ]
        }
      }
    }

and got the result:

{
    "took": 2,
    "total": 1,
    "max_score": 5.329261,
    "hits": [
        {
            "_id": "HGNC_10885-HGNC_3236--MONDO_0006046",
            "_score": 5.329261,
            "association": {
                "edge_label": "coexpressed_with",
                "relation_id": "RO:0002610",
                "relation_name": "correlated_with",
                "correlation": 0.1451,
                "pvalue": 0.054954087385762455,
                "context": {
                    "disease": {
                        "id": "MONDO:0006046",
                        "UBERON": "MONDO:0006046",
                        "name": "ovarian serous cystadenocarcinoma"
                    }
                }
            },
            "object": {
                "id": "HGNC:3236",
                "HGNC": 3236,
                "NCBIGene": 1956,
                "SYMBOL": "EGFR",
                "type": "gene"
            },
            "subject": {
                "id": "HGNC:10885",
                "HGNC": 10885,
                "NCBIGene": 6494,
                "SYMBOL": "SIPA1",
                "type": "gene"
            }
        }
    ]
}

If this is the expected behavior, I'll add the result transformation logic and other refinements.

[namespacestd0]

@newgene do you have an opinion on the additional input format support, should we take extra time to implement dot notation support, mixing url params & url encoded POST body, etc?

[namespacestd0]

Implemented in

pending.api/web/handlers/graph.py

Line 19 in 7530fec

class GraphQueryHandler(ESRequestHandler):