callahantiff/PheKnowLator

What is PheKnowLator?

PheKnowLator (Phenotype Knowledge Translator) or pkt_kg is the first fully customizable knowledge graph (KG) construction framework enabling users to build complex KGs that are Semantic Web compliant and amenable to automatic Web Ontology Language (OWL) reasoning, generate contemporary property graphs, and are importable by today’s popular graph toolkits. Please see the project Wiki for additional information.

What Does This Repository Provide?

1. A Knowledge Graph Sharing Hub: Prebuilt KGs and associated metadata. Each KG is provided as triple edge lists, OWL API-formatted RDF/XML and NetworkX graph-pickled MultiDiGraphs. We also make text files available containing node and relation metadata.
2. A Knowledge Graph Building Framework: An automated Python 3 library designed for optimized construction of semantically-rich, large-scale biomedical KGs from complex heterogeneous data. The framework also includes Jupyter Notebooks to greatly simplify the generation of required input dependencies.

NOTE. A table listing and describing all output files generated for each build along with example output from each file can be found here.

How do I Learn More?

• Join and/or start a Discussion
• The Project Wiki for available knowledge graphs, pkt_kg data sources, and the knowledge graph construction process
• A Zenodo Community has been established to provide access to software releases, presentations, and preprints related to this project

Important Updates and Notifications

• October 2022:
  • Monthly builds are temporatily delayed as we make repairs and implement new functionality. We apologize for any inconvenience this may cause and thank you for your patience and understanding.
• July 2021:
  • Public SPARQL Endpoint: http://sparql.pheknowlator.com/
    • Current Build: July 2021
    • Build Type: OWL-NETS instance-based + inverse relations and no metadata
  • Access build data via a dedicated Google Cloud Storage bucket (see details under Releases)
  • New Jupyter Notebooks:
    • Applying OWL-NETS: OWLNETS_Example_Application.ipynb
    • Exploring pkt_kg knowledge graphs and other ontologies: RDF_Graph_Processing_Example.ipynb

Releases

All data and output for each release are free to download from our dedicated Google Cloud Storage Bucket (GCS). All data can be downloaded from the PheKnowLator GCS Bucket, which is organized by release and build. See full_pheknowlator_build_files.json for a list of all of the knowledge graph and data file URLs for all builds.

Current Release

• v2
  • Build Documentation
  • Data Access

Prior Releases

• v1.0.0
  • Build Documentation
  • Data Access

Getting Started

Install Library

This program requires Python version 3.6. To install the library from PyPI, run:

pip install pkt_kg

You can also clone the repository directly from GitHub by running:

git clone https://github.com/callahantiff/PheKnowLator.git

Note. Sometimes OWLTools, which comes with the cloned/forked repository (./pkt_kg/libs/owltools) loses "executable" permission. To avoid any potential issues, I recommend running the following in the terminal from the PheKnowLator directory:

chmod +x pkt_kg/libs/owltools

Set-Up Environment

The pkt_kg library requires a specific project directory structure.

• If you plan to run the code from a cloned version of this repository, then no additional steps are needed.
• If you are planning to utilize the library without cloning the library, please make sure that your project directory matches the following:

PheKnowLator/
    |
    |---- resources/
    |         |
    |     construction_approach/
    |         |
    |     edge_data/
    |         |
    |     knowledge_graphs/
    |         |
    |     node_data/
    |         |
    |     ontologies/
    |         |
    |     owl_decoding/
    |         |
    |     relations_data/

Dependencies

Several input documents must be created before the pkt_kg library can be utilized. Each of the input documents are listed below by knowledge graph build step:

DOWNLOAD DATA

This code requires three documents within the resources directory to run successfully. For more information on these documents, see Document Dependencies:

• resources/resource_info.txt
• resources/ontology_source_list.txt
• resources/edge_source_list.txt

For assistance in creating these documents, please run the following from the root directory:

python3 generates_dependency_documents.py

Prior to running this step, make sure that all mapping and filtering data referenced in resources/resource_info.txt have been created or downloaded for an existing build from the PheKnowLator GCS Bucket. To generate these data yourself, please see the Data_Preparation.ipynb Jupyter Notebook for detailed examples of the steps used to build the v2.0.0 knowledge graph.

Note. To ensure reproducibility, after downloading data, a metadata file is output for the ontologies (ontology_source_metadata.txt) and edge data sources (edge_source_metadata.txt).

CONSTRUCT KNOWLEDGE GRAPH

The KG Construction Wiki page provides a detailed description of the knowledge construction process (please see the knowledge graph README for more information). Please make sure the documents listed below are presented in the specified location prior to constructing a knowledge graph. Click on each document for additional information. Note, that cloning this library will include a version of these documents that points to the current build. If you use this version then there is no need to download anything prior to running the program.

• resources/construction_approach/subclass_construction_map.pkl
• resources/Master_Edge_List_Dict.json ➞ automatically created after edge list construction
• resources/node_data/node_metadata_dict.pkl ➞ if adding metadata for new edges to the knowledge graph
• resources/knowledge_graphs/PheKnowLator_MergedOntologies*.owl ➞ see ontology README for information
• resources/relations_data/RELATIONS_LABELS.txt
• resources/relations_data/INVERSE_RELATIONS.txt ➞ if including inverse relations

Running the pkt Library

pkt_kg can be run via the provided main.py script or using the main.ipynb Jupyter Notebook or using a Docker container.

Main Script or Jupyter Notebook

The program can be run locally using the main.py script or using the main.ipynb Jupyter Notebook. An example of the workflow used in both of these approaches is shown below.

import psutil
import ray
from pkt import downloads, edge_list, knowledge_graph

# initialize ray
ray.init()

# determine number of cpus available
available_cpus = psutil.cpu_count(logical=False)

# DOWNLOAD DATA
# ontology data
ont = pkt.OntData('resources/ontology_source_list.txt')
ont.downloads_data_from_url()
ont.writes_source_metadata_locally()

# edge data sources
edges = pkt.LinkedData('resources/edge_source_list.txt')
edges.downloads_data_from_url()
edges.writes_source_metadata_locally()

# CREATE MASTER EDGE LIST
combined_edges = dict(edges.data_files, **ont.data_files)

# initialize edge dictionary class
master_edges = pkt.CreatesEdgeList(data_files=combined_edges, source_file='./resources/resource_info.txt')
master_edges.runs_creates_knowledge_graph_edges(source_file'./resources/resource_info.txt',
                                                data_files=combined_edges,
                                                cpus=available_cpus)

# BUILD KNOWLEDGE GRAPH
# full build, subclass construction approach, with inverse relations and node metadata, and decode owl
kg = PartialBuild(kg_version='v2.0.0',
                  write_location='./resources/knowledge_graphs',
                  construction='subclass,
                  node_data='yes,
                  inverse_relations='yes',
                  cpus=available_cpus,
                  decode_owl='yes')

kg.construct_knowledge_graph()
ray.shutdown()

main.py

The example below provides the details needed to run pkt_kg using ./main.py.

python3 main.py -h
usage: main.py [-h] [-p CPUS] -g ONTS -e EDG -a APP -t RES -b KG -o OUT -n NDE -r REL -s OWL -m KGM

PheKnowLator: This program builds a biomedical knowledge graph using Open Biomedical Ontologies
and linked open data. The program takes the following arguments:

optional arguments:
-h, --help            show this help message and exit
-p CPUS, --cpus CPUS  # workers to use; defaults to use all available cores
-g ONTS, --onts ONTS  name/path to text file containing ontologies
-e EDG,  --edg EDG    name/path to text file containing edge sources
-a APP,  --app APP    construction approach to use (i.e. instance or subclass
-t RES,  --res RES    name/path to text file containing resource_info
-b KG,   --kg KG      the build, can be "partial", "full", or "post-closure"
-o OUT,  --out OUT    name/path to directory where to write knowledge graph
-r REL,  --rel REL    yes/no - adding inverse relations to knowledge graph
-s OWL,  --owl OWL    yes/no - removing OWL Semantics from knowledge graph

main.ipynb

The ./main.ipynb Jupyter notebook provides detailed instructions for how to run the pkt_kg algorithm and build a knowledge graph from scratch.

Docker Container

pkt_kg can be run using a Docker instance. In order to utilize the Dockerized version of the code, please make sure that you have downloaded the newest version of Docker. There are two ways to utilize Docker with this repository:

• Obtain Pre-Built Container from DockerHub
• Build the Container (see details below)

Obtaining a Container

Obtain Pre-Built Containiner: A pre-built containers can be obtained directly from DockerHub.

Build Container: To build the pkt_kg download a stable release of this repository (or fork/clone it repository). Once downloaded, you will have everything needed to build the container, including the ./Dockerfile and ./dockerignore. The code shown below builds the container. Make sure to replace [VERSION] with the current pkt_kg version before running the code.

cd /path/to/PheKnowLator (Note, this is the directory containing the Dockerfile file)
docker build -t pkt:[VERSION] .

Notes:

• Update PheKnowLator/resources/resource_info.txt, PheKnowLator/resources/edge_source_list.txt, and PheKnowLator/resources/ontology_source_list.txt
• Building the container "as-is" off of DockerHub will include a download of the data used in the latest releases. No need to update any scripts or pre-download any data.

Running a Container

The following code can be used to run pkt_kg from outside of the container (after obtaining a prebuilt container or after building the container locally). In:

docker run --name [DOCKER CONTAINER NAME] -it pkt:[VERSION] --app subclass --kg full --nde yes --rel yes --owl no --kgm yes

Notes:

• The example shown above builds a full version of the knowledge graph using the subclass construction approach with node metadata, inverse relations, and decoding of OWL classes. See the Running the pkt Library section for more information on the parameters that can be passed to pkt_kg
• The Docker container cannot write to an encrypted filesystem, however, so please make sure /local/path/to/PheKnowLator/resources/knowledge_graphs references a directory that is not encrypted

Finding Data Inside a Container

In order to enable persistent data, a volume is mounted within the Dockerfile. By default, Docker names volumes using a hash. In order to find the correctly mounted volume, you can run the following:

Command 1: Obtains the volume hash:

docker inspect --format='{{json .Mounts}}' [DOCKER CONTAINER NAME] | python -m json.tool

Command 2: View data written to the volume:

sudo ls /var/lib/docker/volumes/[VOLUME HASH]/_data

Get In Touch or Get Involved

Contribution

Please read CONTRIBUTING.md for details on our code of conduct, and the process for submitting pull requests to us.

Contact Us

We’d love to hear from you! To get in touch with us, please join or start a new Discussion, create an issue or send us an email 💌

Attribution

Licensing

This project is licensed under Apache License 2.0 - see the LICENSE.md file for details.

Citing this Work

ISMB Conference Pre-print:

Callahan TJ, Tripodi IJ, Hunter LE, Baumgartner WA. A Framework for Automated Construction of Heterogeneous Large-Scale Biomedical Knowledge Graphs. bioRxiv. 2020 Jan 1.

Zenodo

@misc{callahan_tj_2019_3401437,
  author       = {Callahan, TJ},
  title        = {PheKnowLator},
  year         = 2019,
  doi          = {10.5281/zenodo.3401437},
  url          = {https://doi.org/10.5281/zenodo.3401437}}