Reasoner Validator

This package provides software methods to Translator components (e.g. Knowledge Providers and Autonomous Relay Agents) using any version of the Translator Reasoner API (TRAPI) and the Biolink Model.

See the full documentation and the contributor guidelines.

Using the Package

Python Dependency

The Reasoner Validator now requires Python 3.9 or later (some library dependencies now force this).

Installing the Module

The module may be installed directly from pypi.org using (Python 3) pip or pip3, namely:

pip install reasoner-validator

If you want to install it with the extra dependencies for using it as a web service, you can use:

pip install "reasoner-validator[web]"

Installing and working with the module locally from source

As of release 3.1.6, this project uses the poetry dependency management tool to orchestrate its installation and dependencies.

After installing poetry and cloning the project, the poetry installation may be run:

git clone https://github.com/NCATSTranslator/reasoner-validator.git
cd reasoner-validator
poetry install

To develop this package, install with all extras dependencies using:

poetry install --all-extras

Running Validation against an ARS UUID Result(*) or using a Local TRAPI Request Query

A local script trapi_validator.py is available to run TRAPI Response validation against either a PK (UUID) indexed query result of the Biomedical Knowledge Translator "Autonomous Relay System" (ARS), a local JSON Response text file or a locally triggered ad hoc query Request against an directly specified TRAPI endpoint.

Note that it is best run within a poetry shell created by poetry install.

For script usage, type:

./trapi_validator.py --help

(*) Thank you Eric Deutsch for the prototype code for this script

Running tests

To run the test locally install with the dev dependencies group if not already done:

poetry install --extras dev

Run the tests with coverage report:

poetry run pytest --cov

Note that poetry automatically uses any existing virtual environment, but you can otherwise also enter the one that is created by poetry by default:

poetry shell
# run your commands, e.g. the web service module
exit  # exit the poetry shell

The use of the Poetry shell command allows for running of the tests without the poetry run prefix. We will continue in this manner.

% poetry shell
(reasoner-validator-py3.9) % pytest --cov

Run the tests with detailed coverage report in a HTML page:

pytest --cov --cov-report html

Serve the report on http://localhost:3000:

python -m http.server 3000 --directory ./htmlcov

Building the Documentation Locally

All paths here are relative to the root project directory.

First install the documentation-specific dependencies.

poetry install --extras docs  # or poetry install --all-extras

The validation codes MarkDown file should first be regenerated if needed (i.e. if it was revised):

cd reasoner_validator
python ./validation_codes.py

Then build the documentation locally:

cd ../docs
make html

The resulting index.html and related pages describing the programmatic API are now available for viewing within the docs subfolder _build/html.

Validation Run as a Web Service

The Reasoner Validator is available wrapped as a simple web service. The service may be run directly or as a Docker container.

API

The web service has a single POST endpoint /validate taking a simple JSON request body, as follows:

{
  "trapi_version": "1.4.1",
  "biolink_version": "3.5.0",
  "target_provenance": {
    "ara_source": "infores:aragorn",
    "kp_source": "infores:panther",
    "kp_source_type": "primary"
  },
  "strict_validation": true,
  "response": "{<some full JSON object of a TRAPI query Response...>}"
}

The request body consists of JSON data structure with two top level tag:

• An optional trapi_version tag can be given a value of the TRAPI version against which the message will be validated, expressed as a SemVer string (defaults to 'latest' if omitted; partial SemVer strings are resolved to their 'latest' minor and patch releases). This value may also be a GitHub branch name (e.g. 'master').
• An optional biolink_version tag can be given a value of the Biolink Model version against which the message knowledge graph semantic contents will be validated, expressed as a SemVer string (defaults to 'latest' Biolink Model Toolkit supported version, if omitted).
• An optional target_provenance with an object dictionary (example shown) specifying the ARA and KP infores-specified knowledge sources expected to be recovered in the TRAPI query results (specified by infores CURIE) and the expected KP provenance source type, i.e. 'primary' implies that the KP is tagged as a 'biolink:primary_knowledge_source'. Optional in that the root "target_provenance" or any of the subsidiary tags may be omitted (default to None)
• An optional strict_validation flag (default: None or 'false'). If 'true' then follow strict validation rules, such as treating as 'error' states the use of category, predicate and attribute_type_id that are of type abstract or mixin as errors.
• A mandatory message tag should have as its value the complete JSON TRAPI Response to be validated (See the example below)

Running the Web Service Directly

First install the web-specific dependencies.

poetry install --extras web  # or poetry install --all-extras

The service may be run directly as a Python module. The web services module may be directly run, as follows.

python -m api.main

Go to http://localhost/docs to see the service documentation and to use the simple UI to input TRAPI messages for validation.

Typical Output

As an example of the kind of output to expect, if one posts the following TRAPI Response JSON data structure to the /validate endpoint:

{
  "trapi_version": "1.4.1",
  "biolink_version": "3.5.0",
  "response": {
      "message": {
        "query_graph": {
            "nodes": {
                "type-2 diabetes": {"ids": ["MONDO:0005148"]},
                "drug": {"categories": ["biolink:Drug"]}
            },
            "edges": {
                "treats": {"subject": "drug", "predicates": ["biolink:treats"], "object": "type-2 diabetes"}
            }
        },
        "knowledge_graph": {
            "nodes": {
                "MONDO:0005148": {"name": "type-2 diabetes"},
                "CHEBI:6801": {"name": "metformin", "categories": ["biolink:Drug"]}
            },
            "edges": {
                "df87ff82": {"subject": "CHEBI:6801", "predicate": "biolink:treats", "object": "MONDO:0005148"}
            }
        },
        "results": [
            {
                "node_bindings": {
                    "type-2 diabetes": [{"id": "MONDO:0005148"}],
                    "drug": [{"id": "CHEBI:6801"}]
                },
                "edge_bindings": {
                    "treats": [{"id": "df87ff82"}]
                }
            }
        ]
      },
      "workflow": [{"id": "annotate"}]
  }
}

one should typically get a response body something like the following JSON validation result back:

{
  "trapi_version": "1.4.1",
  "biolink_version": "3.2.1",
  "messages": {
    "_comment": "some categories of messages may be absent, hence, empty dictionaries",
    "critical": {},
    "errors": {
      "error.knowledge_graph.node.category.missing": {
          "_comment": "source scope of the validation error ('global' or some knowledge source path string)",
          "global": {
              "_comment": "this message template does not have any additional parameters other than identifier hence it just has the unique identifier  value as a dictionary key, with associated value None",
               "MONDO:0005148": null            
          }
        }
    },
    "warnings": {
      "_comment": "validation code",
      "warning.knowledge_graph.node.unmapped_prefix": {
          "global": {
              "_comment": "template identifier field value",
              "CHEBI:6801": [  
                  {
                    "_comment": "additional message template field values, if applicable",
                    "categories": "['biolink:Drug']"  
                  }
              ]       
          }

        }
    },
    "information": {},
  }
}

To minimize redundancy in validation messages, messages are uniquely indexed in dictionaries at two levels:

1. the (codes.yaml recorded) dot-delimited validation code path string
2. for messages with templated parameters, by a mandatory 'identifier' field (which is expected to exist as a field in a template if such template has one or more parameterized fields)

OpenTelemetry and Jaeger

The web service may be monitored for OpenTelemetry by setting an environment variable TELEMETRY_ENDPOINT to a suitable trace collecting endpoint in an application like Jaeger (see also the Translator SRI Jaeger-Demo).

Note: the current system Docker (Compose) design only supports OpenTemplate tracing using the internal Jaeger container and may require further refinements to enable use of an external telemetry collector.

Running the Web Service within Docker

The Reasoner Validator web service may be run inside a docker container, using Docker Compose.

First, from the root project directory, build the local docker container

docker-compose build

Then, run the service:

docker-compose up -d

Once again, go to http://localhost/docs to see the service documentation.

To stop the service:

docker-compose down

Of course, the above docker-compose commands may be customized by the user to suit their needs. Note that the docker implementation assumes the use of uvicorn

Change Log

Summary of earlier releases and current Change Log is here.

Code Limitations (implied Future Work?)

• The release of the reasoner-validator after v2.2.0 will not likely be able to (reliably, if at all) validate TRAPI JSON data models prior to 1.3.0
• Biolink Model release <= 2.4.8 versus 3.0.0 validation: the reasoner-validator uses the Biolink Model Toolkit. As it happens, the toolkit is not backwards compatible with at least one Biolink Model structural change from release 2.#.# to 3.#.#: the tagging of 'canonical' predicates. That is, the 0.8.10++ toolkit reports canonical <= 2.4.8 model predicates as 'non-canonical'.
• This release of the Reasoner Validator Web Service will detect TRAPI 1.0.* releases but doesn't strive to be completely backwards compatible with them (considering that TRAPI 1.0.* is totally irrelevant now).
• The web service validation doesn't do deep validation of the Results part of a TRAPI Message
• The validation is only run on the first 1000 nodes and 100 edges of graphs, to keep the validation time tractable (this risks not having complete coverage of the graph)
• Biolink Model toolkit is not (yet) cached so changing the model version during use will result in some latency in results
• The validator service doesn't (yet) deeply validate non-core node and edge slot contents of Message Knowledge Graphs
• The validator service doesn't (yet) attempt validation of Query Graph nodes and edges 'constraints' (e.g. biolink:Association etc. domain and range constraints)
• Query Graph node 'ids' are not validated except when an associated 'categories' parameter is provided for the given node. In general, Query Graph Validation could be elaborated.
• The system should leverage the Reasoner Pydantic Models

Core Contributors

• Kudos to Patrick Wang, who created the original implementation of the Reasoner-Validator project while with CoVar (an entrepreneurial team contributing to the Biomedical Data Translator).
• Thanks to Kenneth Morton (CoVar) for his reviews of the latest code.
• The project is currently being extended and maintained by Richard Bruskiewich (Delphinai Corporation, on the SRI team contributing to Translator)