ubkg-api 2.3.0

A REST API in front of the Unified Biomedical Knowlegde Graph

Unified Biomedical Knowledge Graph

ubkg-api

The components of the Unified Biomedical Knowledge Graph (UBKG) include:

• The source framework that extracts ontology information from the UMLS to create a set of CSV files (UMLS CSVs)
• The generation framework that appends to the UMLS CSVs assertion data from other ontologies to create a set of ontology CSVs.
• A neo4j ontology knowledge graph populated from the ontology CSVS.
• An API server that provides RESTful endpoints to query the ontology knowledge graph.

This repository contains the source for the API.

The ubkg-api is a Flask web application with Blueprint extensions that provides a REST API for parameterized queries against an instance of a UBKG neo4j instance.

The ubkg-api contains code that is common to all UBKG contexts, including code to handle connecting to a UBKG instance.

Other API implementations, such as the hs-ontology-api, extend the ubkg-api. Endpoints common to all UBKG contexts are deployed in the ubkg-api; other API implementations can both "pass through" calls to ubkg-api endpoints or execute endpoints specific to the implementation.

An API implementation extends the ubkg-api by calling a version of the ubkg-api that has been published as a PyPI.

SmartAPI documentation

The specification for the UBKG API can be found here.

Developer Information

Development/testing environment for ubkg-api

To enhance or fix the ubkg-api, you will need to establish an application development environment on the development machine. Components of the application development environment include:

1. An instance of a UBKG context--i.e., an instance of neo4j populated with UBKG content. Options include:
   • a local bare-metal instance of neo4j
   • a local Docker install of a UBKG distribution built from ubkg-neo4j
   • a cloud-based instance (development or production)
2. A local branch of ubkg-api
3. URLs that execute endpoints against the local instance of ubkg-api

Connecting to a UBKG instance

To connect the local branch of ubkg-api to a neo4j instance that hosts a UBKG context:

1. Copy the file named app.cfg.example in the src/ubkg-api/instance directory to a file named app.cfg.
2. Add to app.cfg the connection information for the neo4j instance. The .gitignore file at the root of this repo will force git to exclude the app.cfg file from commits.

Example UBKG neo4j connect string values for app.cfg

If you are working with a local Docker distribution based on from ubkg-neo4j, then

• SERVER='bolt://localhost:u', where u is the browser port associated with ui_port in the container.cfg file used by the build scripts.
• USERNAME=neo4j
• PASSWORD= the value of neo4j_password from the common container.cfg file.

Using API keys

Some deployments based on the UBKG API may require that calls to endpoints include an API key in the header. Because this involves integration with authorization architecture (e.g., API gateways and authorization specific to a network resource), the configuration of API keys is beyond the scope of the UBKG API.

The Data Distillery API is an example of an API that queries a UBKG instance, but requires an API key.

Starting your neo4j instance

If you are using a local instance of the UBKG, the instance should be running. In particular, if you installed a local Docker instance of UBKG, be sure that Docker Desktop is running. If the neo4j instance is not available, calls to API endpoints will result in a 500 error.

Connecting to the local instance of ubkg-api

For URLs that execute endpoints in your local instance, use the values indicated in the main.py script, in the section prefaced with the comment For local development/testing:

For example, if main.py indicates app.run(host='0.0.0.0', port="5002")

then your test endpoint URLs should start with http://127.0.0.1:5002/.

For example, if you test using PostMan, you can set a global variable corresponding to the first part of your test URLs:

Testing changes

To test changes to ubkg-api, you will need to start a local instance of the API. The following assumes that you have created a local branch of ubkg-api.

From the command line

1. Move to the root of your local branch.

2. Create a Python virtual environment. The following command creates a virtual environment named venv.

   python -m venv venv

3. Activate the virtual environment.

   source venv/bin/activate

4. Move to the /src directory and install dependencies, inclduing the ubkg-api package.

   pip install -r requirements.txt

5. Run app.py to start the local instance of the API.

   python app.py

In PyCharm

1. Create a new project based on a local clone of ubkg-api. PyCharm should establish a virtual environment. (Refer to instructions here.)
2. Use the Python Packages tab to install the packages listed in requirements.txt.
3. In the Terminal window, run app.py.
4. Note that you may need to enable execute permissions on the app.py script before you can run it locally--e.g., with a command like chmod 777 app.py

URL testing

Once you have connected your instance of hs-ontology-api to instances of both neo4j and ubkg-api, run the following tests:

1. Paste the root endpoint URL into a browser window--e.g.,http://127.0.0.1:5002/. You should see a window with the status message Hello! This is UBKG-API service :). The status message verifies that your local instance of hs-ontology-api is connected to an instance of ubkg-api.
2. Add to the root endpoint URL to execute a known endpoint--e.g., http://127.0.0.1:5002/concepts/C0678222/codes. You should see a response from the ubkg-api, depending on the endpoint.
3. If you are only testing endpoints in an extension api (e.g., hs-ontology-api) and are using the PyPi install of ubkg-api, calls to endpoints managed by the ubkg-api will fail with a 500 error. To test endpoints from both hs-ontology-api and ubkg-api, you will need a local instance of ubkg-api that connects to the same instance of neo4j to which the instance of hs-ontology-api connects.

Various methods of testing endpoint URLs are possible, including:

1. curl, from either the command line or a shell script
2. Requests in Postman
3. A Python script using Requests or pytest
4. Executing directly in the browser. This method is suitable for GET endpoints.

Adding new endpoints

Endpoint pattern

The ubkg-api is a Flask application with Blueprint extensions.

An endpoint in ubkg-api involves:

1. A Blueprint registered in app.py that defines route (URL) paths
2. A controller script in the common_routes path
3. A route decorator in the controller script that defines the endpoint's route
4. A controller routine invoked by the route decorator
5. A functional routine in the functional script (neo4j_logic.py), invoked by the controller routine
6. A Cypher query in the cypher path, executed by the functional script

In general, a controller script has multiple decorators, and thus multiple routes.

Endpoints follow a general pattern.

1. The controller routine:
   • validates request parameters
   • formats request parameters
   • executes the functional routine
   • processes the product of the functional routine:
     • handles the case of no data (HTTP 404)
     • optionally serializes neo4j Path objects as JSON
     • redirects output that exceeds configured size limits to a defined S3 bucket (HTTP 403) and returns the URL to the bucket
2. The functional routine:
   • reads the Cypher query from file
   • passes parameters to the Cypher query
   • executes the Cypher query
   • processes the response of the Cypher query:
     • handles the case of a server timeout (HTTP 408)
     • packages the response, extracting or wrapping as necessary
3. The Cypher query, in general, returns in either a stream in JSON or as a neo4j Path object.

Helpers and utilities

1. neo4j_connection_helper.py contains the Neo4jConnectionHelper class that manages connectivity to a UBKG context neo4j instance specified in the app.cfg.
2. The utils path contains utilities and classes that
   • validate and format request parameters for passing to Cypher queries.
   • build error strings for responses (e.g., 404)
   • serialize neo4j Path objects to JSON format
   • redirect output to a S3 bucket

Development tasks for a new endpoint

1. Using neo4j Browser, develop a parameterized Cypher query that returns in JSON format.
2. Store the Cypher query in a file in the cypher path.
3. If a new endpoint is required,
   • Build a new controller script.
   • Define and register a new Blueprint.
   • In the controller script, define a new decorator for the endpoint route.
4. Build a new functional routine in neo4j_logic.py for the endpoint.
5. In the controller script,
   • Define a new controller routine for the endpoint.
   • Validate request parameters.
   • Execute the functional routine.
   • Handle 404s.
   • Optionally serialize Path objects as JSON.
   • Redirect to S3 as needed.
6. In the functional routine,
   • Read the Cypher query.
   • Pass parameters to the query.
   • Execute the query.
   • Handle timeout errors.

Update tests scripts

Test scripts are in the tests path.

1. ubkg_unit_test.sh should test using curl for all known use cases, including:
   • invalid parameters
   • missing parameters
   • missing data
   • timeout errors
   • large output
2. ubkg_system_test.sh should test for a working gateway--i.e., that the route has been defined in the API server gateway.

Updating SmartAPI documentation

To add the specification for a new endpoint to the SmartAPI documentation for hs-ontology-api, update the files:

1. ubkg-api-spec.yaml for ubkg-api
2. ubkg-api-ubkgbox-spec.yaml for the UBKGBox deployment

The YAML files conform to Swagger OpenAPI format.

You will need to specify:

1. Paths that correspond to your endpoint routes.
2. Schemas that correspond to the responses from endpoints.

Building and publishing to PyPI

To build and publish the ubkg-api as a PyPI package,

python -m build
python3 -m twine upload dist/*

Requirements

Python 3.9 or newer

Optional JSON serialization

A number of routes in the concepts/paths path execute Cypher queries that return complex objects, including neo4j Path objects.

For example, the /paths/subgraph endpoint executes the following Cypher:

//For the set of paths,

// 1. Obtain an "edges" object with information on all relationships in all paths.
// 2. Obtain a "paths" object with path information on all paths.

UNWIND(relationships(path)) AS r
WITH path,collect({type:type(r),SAB:r.SAB,source:startNode(r).CUI,target:endNode(r).CUI}) AS path_r
WITH collect(path) as paths, apoc.coll.toSet(apoc.coll.flatten(COLLECT(path_r))) AS edges

// 3. Obtain a "nodes" object for all Concept nodes in all paths
UNWIND(paths) AS path
UNWIND(nodes(path)) AS n
// Obtain preferred terms for Concept nodes.
OPTIONAL MATCH (n)-[:PREF_TERM]->(t:Term)

WITH paths,edges,collect(DISTINCT{id:n.CUI,name:t.name}) AS nodes
RETURN {nodes:nodes, paths:paths, edges:edges} AS graph

The paths portion of the response is a neo4j Path object. Queries that attempt to return responses with Path objects fail with an error that they "are not JSON serializable".

Routes that return complex responses with Path objects include:

• /paths/expand
• /paths/trees
• /paths/subgraph
• /paths/subgraph/sequential

To convert neo4j Path objects into JSON-serializable objects, queries use the code in path_json_serializer.py. The script contains the class PathJSONSerializer, a custom serializer that can handle the returns from the Cypher queries behind the routes.

A controller script can serialize with the command

# Serialize the Path object resp as JSON.
    result = PathJSONSerializer(result).json

The custom serializer offers more control than the available APOC functions.

Optional Timeout Feature

The ubkg-api can be deployed in various environments.

When deployed behind a server gateway, such as AWS API Gateway, the gateway may impose constraints on timeout or response payload size. For example, AWS API Gateway has a timeout of 29 seconds and a maximum response payload of 10 MB.

The ubkg-api can be configured to enforce its own timeout that occurs before an actual gateway timeout, essentially acting as a proxy for the gateway. The ubkg-api's HTTP 504 timeout message will return detailed explanations for the timeout.

Code required

app.cfg

To enable custom management of timeout and payload size, specify values in the app.cfg file, as shown below.

# Maximum duration for the execution of timeboxed queries to prevent service timeout, in seconds
# The AWS API gateway timeout is 29 seconds.
TIMEOUT=28

If the ubkg-api is imported as a PyPI package into a child API (the default configuration), the timeout should be specified in the app.cfg of the child API. The configuration of the child API takes precedence over that of the ubkg-api.

Endpoint function code

To validate timeout, use a try/exception block in the code in the neo4j_logic.py in the utils folder.

Example:

from werkzeug.exceptions import GatewayTimeout
...
    #Set timeout for query based on value in app.cfg.
    query = neo4j.Query(text=querytxt, timeout=neo4j_instance.timeout)

    with neo4j_instance.driver.session() as session:
        try:
            recds: neo4j.Result = session.run(query)

            for record in recds:
                # process records

        except neo4j.exceptions.ClientError as e:
            # If the error is from a timeout, raise a HTTP 504.
            if e.code == 'Neo.ClientError.Transaction.TransactionTimedOutClientConfiguration':
                raise GatewayTimeout

    return <your records>

The ubkg-api returns a custom HTTP 504 response when a query exceeds the configured timeout.

Coding required

app.cfg

To enable timeout validation, specify the following key in the app.cfg file:

LARGE_RESPONSE_THRESHOLD = 9*(2**20) + 900*(2**10) #9.9Mb

Payload size validation with optional S3 redirection

APIs in environments employing an AWS API gateway have limits on the size of response payloads. The current default AWS API gateway limit on payloads is 10 MB.

The ubkg-api can be configured to check the size of response payloads. This feature can be used to prevent triggering of the actual gateway error. In addition, the ubkg-api's HTTP 403 error code message provides more detail than does the message from the gateway.

The ubkg-api can also work around a gateway payload limit by redirect large payloads to an AWS S3 bucket. The ubkg-api will return a URL that points to the file in the S3 bucket. The URL is "pre-signed": consumers can simply "get" the URL to download the file locally.

If S3 redirection is not configured, the ubkg-api will return a simple HTTP 403 response.

Coding required

app.cfg

To enable S3 redirection, specify values in the app.cfg file.

# Large response threshold, as determined by the length of the response (payload).
# Responses with payload sizes that exceed the threshold will be handled in one of the
# following ways:
# 1. If the threshold is 0, then the response will be passed without any additional processing.
# 2. If the threshold is nonzero and S3 redirection is not enabled, the API will return
#    a custom HTTP 403 response.
# 3. If the threshold is nonzero and S3 redirection is enabled, the API will stash the
#    response in a file in an S3 bucket and return a pre-signed URL pointing to the
#    stashed file.
# Setting the threshold to 9.9MB avoids triggering a HTTP 500 from an AWS API Gateway's hard
# 10 MB payload limit
LARGE_RESPONSE_THRESHOLD = 9*(2**20) + 900*(2**10) #9.9Mb

# OPTIONAL AWS credentials for S3 redirection. If there are no "AWS_*" keys, the
# API will return the default HTTP 403 exception.
# https://boto3.amazonaws.com/v1/documentation/api/latest/guide/credentials.html
AWS_ACCESS_KEY_ID = 'AWS_ACCESS_KEY_ID'
AWS_SECRET_ACCESS_KEY = 'AWS_SECRET_ACCESS_KEY'
AWS_S3_BUCKET_NAME = 'AWS_S3_BUCKET_NAME'
AWS_S3_OBJECT_PREFIX = 'AWS_S3_OBJECT_PREFIX'
AWS_OBJECT_URL_EXPIRATION_IN_SECS = 60*60 # 1 hour

Note that if the ubkg-api is being imported as a PyPI package (the default configuration), the S3 configuration should be specified in the app.cfg of the child api. Configuration in the child API takes precedence over that of the ubkg-api.

route logic

Add the following import to the controller:

# S3 redirect functions
from utils.s3_redirect import redirect_if_large

Send the result of the query to payload validation:

    result = <call to function in neo4j_logic.py>
    # Redirect to S3 if payload is large.
    return redirect_if_large(resp=result)