Skip to content

NCATS Translator ARA TRAPI wrapper for the Monarch Initiative system

License

Notifications You must be signed in to change notification settings

monarch-initiative/monarch-trapi-kp

Repository files navigation

Monarch SemSimian Multi-CURIE Query TRAPI Knowledge Provider ("MMCQ")

This project is a NCATS Translator API ("TRAPI") wrapper for the Monarch Initiative information system - or rather specifically, the Semantic Similarity ("SemSimian") component of the system - making it behave like a TRAPI Knowledge Provider ("KP") responding to Multi-Curie "similarity" queries against its embedded "SemSimian" algorithm.

The initial release of the wrapper application supports the following use case:

Given a set of (Human Phenotype Ontology term 'Human Phenotype Ontology ("HPO") term identified) phenotypes, what Monarch Disease Ontology ("MONDO") indexed diseases do they match?

The goal is to find a good, probably creative answer that satisfies as many of the N inputs as possible, but may not satisfy all of them.

Installation

Install dependencies within a suitable virtual environment

The Python virtual environment and dependencies of MMCQ are managed using Poetry. Assuming that you have Poetry and a suitable version of Python (i.e. ">=3.9,<3.12") installed, then:

poetry shell
poetry install

Jupyter Notebook

The MultiCurieQueries.ipynb Jupyter Notebook may require a minimum Python version of 3.11 to run and does require installation of the optional poetry dependencies. Modifying the above command as follows:

poetry install -E jupyter

Please note that the MMCQ server itself must be running (see below) for the Jupyter Notebook itself to work!

Configure MMCQ settings

Copy the .env-template file, saved as .env in repository root dir, then customize accordingly, for example:

 # SemSimian backend API for 'development' environment 
 # (see the '.env-template' file for possible alternat
 # parameters for a 'production' deployment)
 SEMSIMIAN_MODE="monarch"
 SEMSIMIAN_SCHEME="http://"
 SEMSIMIAN_HOST="api-v3.monarchinitiative.org"
 SEMSIMIAN_PORT=""  # default is HTTP port '80'
 SEMSIMIAN_SEARCH="/v3/api/semsim/search"
 
 # Front End Web Service
 WEB_HOST=0.0.0.0
 WEB_PORT=8080
 
 # TRAPI Service Endpoint
 # Use a real host name or IP here during deployment,
 # e.g. something like:
 # MMCQ_SERVICE_ADDRESS=trapi-mcq.monarchinitiative.org
 MMCQ_SERVICE_ADDRESS="localhost"
 MMCQ_TITLE="Monarch SemSimian MCQ"
 MMCQ_VERSION='1.4.0'

Troubleshooting

You may occasionally see the following mysterious error:

Running uvicorn APP with --host  --port 8080
INFO:     Will watch for changes in these directories: ['/code/monarch-trapi-kp']
ERROR:    [Errno -2] Name or service not known

especially in Docker container runs. If you look closely here, you'll see that although the --host parameter is given to uvicorn, in fact, the parameter value is empty!

First, for reliable 'source' reading of the .env file, enclose all environmental variable values in "double quotes".

Secondly, if you are developing under Microsoft Windows (even if using a cygwin or equivalent bash shell), whenever you change the contents of your .env file, ensure that your .env file has 'unix' style \n end-of-line characters (i.e. no Windoze \r carriage returns!) by running a *nix command line tool like 'dos2unix' to force all end-of-line indications to be *nix compatible.

Running the System

Run the Server from the CLI

Run the following script to start up the server from the command line terminal (under a Unix-type operating system):

    ./main.sh

Running the Server within a Docker Container

Or build an image and run it. From the root directory, type:

  docker build --tag mmcq-test .
 docker run  --name mmcq -p 8080:8080  mmcq-test

View logs using:

  docker logs -f mmcq

A quicker way to deployment is to install Docker Compose then use the provided docker-compose.yaml file:

  docker-compose build
  
  # -d runs the container in the background
  docker-compose up -d
  docker-compose logs -f
  
  # shut the server down when finished...
  docker-compose down

Note that use of Docker Compose allows local Docker container execution of the SemSimian Server image as the 'semsim' service. In such a case, the SEMSIMIAN_MODE environment variable (i.e. in ..env) should be set to value 'server'.

Note that the SemSimian server may be run by Compose, then accessed either by an independently running MMCQ application or the MMCQ can itself be run as a Compose service. However, the .env environment variable file should have the correct SEMSIMIAN parameters set for this (the .env_template file documents these various optional sets of parameters). The key difference is that an external MMCQ application sees the SemSimian Server running on localhost, but the the Compose managed MMCQ expects to see the SemSimian Server service on host 'semsim'. This is reflected in the environment variables.

Viewing the System

TRAPI API

When run the system locally from the CLI or using Docker (but not within any named host), an OpenAPI web form exposing the TRAPI API is available at http://localhost:8080/1.5/docs. The web form has a sample JSON query input that should work (the source file for this is here).

All of the standard TRAPI 1.5 endpoints directly accessible, as expected. These mainly consist of the /meta_knowledge_graph returning the Biolink Model compliant dictionary of edge templates and the /query endpoint for posting queries to the system.

Note that for the /query endpoint, the TRAPI query graph body can have the (optional) extra non-TRAPI standard JSON object key limit which instructs the system about the maximum number of results should be returned (Default: return the top 5 results). The current maximum allowable SemSimian value for this value appears to be 50. Higher values will trigger a 422 HTTP return code error.

'Common' API

An additional set of endpoints - so-called 'COMMON' API endpoints - is available at http://localhost:8080/common/docs. Aside from accessing available release metadata about the system (via the /common/metadata path), this set of endpoints also provides a few non-TRAPI general purpose endpoints to retrieve specific data results more conveniently than TRAPI, such as retrieving a node record by CURIE.

Only the /metadata endpoint is implemented at this moment.

AWS deployment

  • create an AWS EC2 instance
  • ssh to the instance
  • install docker / start the service
  • build the docker image
  • run the docker image in the background