Skip to main content

Validation tools for Reasoner API

Reason this release was yanked:

Superseded by 3.4.9

Project description

Reasoner Validator

Pyversions pypi License: MIT

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.

pip install reasoner-validator

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

Running Validation against an ARS UUID Result(*)

A local script ars_uuid_result_test_runner.py is available to run TRAPI Response validation against a UUID indexed query result of the Biomedical Knowledge Translator "Autonomous Relay System" (ARS).

For usage, type:

./ars_uuid_result_test_runner.py --help

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

Building the Documentation Locally

All paths here are relative to the root project directory.

First install the documentation-specific dependencies.

poetry install --extras docs

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 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": "1.3.0",
  "biolink_version": "3.2.1",
  "sources": {
    "ara_source": "infores:aragorn",
    "kp_source": "infores:panther",
    "kp_source_type": "primary"
  },
  "strict_validation": true,
  "message": {...}
}

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).
  • 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 sources with an object dictionary (example shown) specifying the ARA and KP sources involved in the TRAPI call (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 "sources" 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 TRAPI Message JSON data structure to be validated (see example below).

Running the Web Service Directly

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

python -m api.main

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

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 JSON message data to the /validate endpoint:

{
  "trapi_version": "1.3.0",
  "biolink_version": "3.2.1",
  "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": "1.3.0",
  "biolink_version": "3.2.1",
  "messages": {
    # some categories of messages may be absent, hence, empty dictionaries
    "information": {},
    "warnings": {
      # validation code
      "warning.knowledge_graph.node.unmapped_prefix": {
          # template identifier field value
          "CHEBI:6801": [  
              {
                # additional message template field values, if applicable
                "categories": "['biolink:Drug']"  
              }
          ]
          
        }
      
    },
    "errors": {
      "error.knowledge_graph.node.category.missing": {
          # 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": None
        }
    }
  }
}

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

  1. the (codes.yaml recorded) dot-delimited error 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)

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?)

  • 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)

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

reasoner_validator-3.4.7.tar.gz (63.9 kB view hashes)

Uploaded Source

Built Distribution

reasoner_validator-3.4.7-py3-none-any.whl (38.2 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page