HMKG 0.0.1

HMKG, a Knowledge Graph on Human Metabolome

HMKG

This package aims to download, preprocess, and analyze the Human Metabolome Database (HMDB) for metabolomics knowledge graph construction and knowledge graph embedding (KGE) tasks.

Package Structure

The package is structured as follows:

• HMKG.HMDB: contains scripts to download, preprocess and convert HMDB data into triples and subgraphs.
• HMKG.KGE: contains scripts to construct triples and split data for KGE models.
• HMKG.search: contains scripts for searching the knowledge graph in both forward and backward directions.
• HMKG.stats: contains scripts to generate summary statistics, visualize the knowledge graph, and provide utility functions.

Usage

To use this project, follow these steps:

1. Run HMKG.HMDB.download_HMDB() function from HMDB/download.py to download the HMDB dataset and save it to a local directory. This function takes two arguments: the URL to download the dataset and the file path to save it to.
2. Run HMKG.HMDB.convert_xml_to_json() function to convert the dataset from XML to JSON format. This function takes two arguments: the input file path (the downloaded HMDB dataset) and the output file path to save the converted data to.
3. Run HMKG.HMDB.create_triples() function to create triples from the converted dataset. This function takes one argument: the path to the converted HMDB dataset in JSON format.
4. Run HMKG.KGE.split_data() function to split the triples into training and testing datasets for KGE models. This function takes one argument: the path to the triple data.
5. Instantiate a KGE model from HMKG.KGE.KGEmbedding and run the KGE_model_pipeline() method to train the model. This script requires you to have already run HMKG.KGE.split_data() to create the training and testing datasets.
6. Use the scripts in HMKG.search to search the knowledge graph for entities and relationships.
7. Use the scripts in HMKG.stats/ to generate summary statistics and visualize the knowledge graph.

Dependencies

This project requires the following dependencies:

• Python 3
• Pandas
• NumPy
• Scikit-learn
• PyTorch
• NetworkX

Example Usage

from HMKG.HMDB import download_HMDB, create_triples
from HMKG.search import search_backward
from HMKG.stats import summary
from HMKG.KGE import split_data, KGEmbedding

output_path = download_HMDB(url="https://hmdb.ca/system/downloads/current/saliva_metabolites.zip", file_path="data/saliva_metabolites.zip")

output_path = convert_xml_to_json(output_path, "data/saliva_metabolites.json")

triples = create_triples("data/saliva_metabolites.json")

split_data(triple_path=triples)

kge = KGEmbedding(model_name="TransE")

kge.construct_triples()
kge.save_id_mapping()
kge.KGE_model_pipeline()