This repository provides the process of the construction of the Ketamine Pathway (KetPath) knowledge graphs wlong with all the datasets and query protocols used in the work. KetPath is a biomedical pathway knowledge graph designed for exploring the pathway relations of the antidepressant ketamine, and ketamine's effects on brain neurotransmitters and gut microbiota. This project is done by IBIVU and L&R group, Verije Universteit Amsterdam.
Ting Liu ([email protected]), K. Anton Feenstra ([email protected]), Zhisheng Huang ([email protected]) and Jaap Heringa ([email protected])
The datasets used to construct the KetPath knowledge graph are as follows:
- KetFact (28 KB) - created by manual fact curation
- KetCept (14 MB) - generated by the XMedlan tool
- KetRela (0.9 MB) - derived by the BioKetBERT model
- MiKG (0.1 MB) - created by manual fact curation, results in the MiKG paper
- PPKG (0.3 MB) - created by combining manual fact curation and the XMedlan tool, results in the PPKG paper
- WikiPathway (18 MB) - only includes biological pathways of homo sapiens
- Gene Ontology (7 MB) - contains gene and gene product attributes of homo sapiens
- KEGG Pathways (0.7 MB) - includes data profiles of pathways related to neurotransmitters
- DrugBank (180 MB) - includes basic information on drug targets, pharmacology and metabolism
- UMLS (31 MB) - consists of named biomedical vocabularies and defined relationships between terms
- SNOMED CT (48 MB) - consists of named biomedical vocabularies and defined relationships between terms
- BioBERT-Base v1.1 (+ PubMed 1M) - based on BERT-base-Cased, see the BioBERT repository page
Retrieve PubMed IDs (PMIDs) for relevant articles by using the following query: https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pubmed&term="Ketamine+Depression"&RetMax=10000. Save the retrieved list of PIMDs locally in text format under the pmid2html-->data-->pmidlist file directory.
Right-click and edit the run-PubmedDownload.bat file located in the pmid2html directory. Updated the name of the PMID list file as needed, save changes, and then double-click the run-PubmedDownload.bat file to initiate the download of articles in HTML format.
(1) Manually collect hand-drawn images and free-text descriptions of the ketamine pathway from relevant publications.
(2) Manually extract entities and relations of interest from those images and texts.
(3) Define entities and their relations using KetFact as a domain name.
(4) Map entities to the KEGG and UMLS databases by matching their URIs with the owl:sameAs statement. The generated dataset is stored in the KetFact.ttl file.
Use XMedlan, a Xerox Concept Identifier (CI-er) tool, for named entity recognition and semantic annotation of biomedical text with terminologies UMLS and SNOMED CT.
The dataset generated by XMedlan is stored in the KetCept.zip file.
1. Downloading and starting the GraphDB
Download GraphDB free edition from https://www.ontotext.com/products/graphdb/graphdb-free/. GraphDB is a Java application and requires Java 8. Make sure you have an up-to-date Java 8 installation. Please download Java from https://java.com/en/download/help/download_options.xml. Run the GraphDB script in the bin directory. To access the Workbench, open http://localhost:7200 in a browser. Please get more information on GraphDB for a quick start guide.
2. Creating repositories in GraphDB
By default, the GraphDB will initialize a location but not create any repositories.
Create a new repository from the menu Setup -> Repositories page in the GraphDB Workbench.
To create a new repository in the location, click Create new repository.
Fill in KetPath at the Repository ID*, then click creat.
3. Importing databases in GraphDB
Import all databases from the menu Import -> RDF page -> User data. Click on Upload RDF files and select files to upload.
Click on Import, and fill the Target graphs with a URL as Named graphs.
If the database is too large to upload from the User data page, you can store the database in the local graphdb-import file and then import the database from the menu Import -> RDF page -> Server files.
All databases are available at KetPath Release Page.
4. Accessing and reasoning data in GraphDB
The data in GraphDB can be accessed and reasoned by SPARQL query language. Start query from the menu SPARQL.
Please find query templates in the sparql-codes. Users can perform their own queries by changing the parameters in the templates.
Use the SPARQL query endpoint of UniProt to retrieve its data without downloading the data stack.
1. Preparing training and testing datasets for relation extraction
Run the query protocol for case 1 from the 'sparql-codes.txt' file in GraphDB to retrieve sentences where 'ketamine' and 'neurotransmitter' co-occurrence. The target-named entities need to be tagged with the pre-defined tags such as @entity$ for further relation extraction tasks. For the training set, a label needs to be added indicating whether the two entities in a sentence are related: 1 for a relation, 0 otherwise.
2. Fine-tuning the BioBERT model and performing relation extraction
Fine-tune the BioBERT model and dubber the resulting model BioKetBERT. Perform relation extraction tasks withBioKetBERT according the process provided by Lee et al. in the BioBERT repository. More details can be find in their paper BioBERT: a pre-trained biomedical language representation model for biomedical text mining for more details.
The performance of the relation extraction task needs to be evaluated using five-fold cross-validation, where the dataset containing 2,143 sentences is randomly partitioned into five stratified subsets of equal size: for each round, one data block (428 sentences) is used for training and the remaining (1,715 sentences) four are for testing.
The relational dataset generated by the BioKetBERT model is stored in the KetRela.ttl file, which also need to be added to the KetPath repository in GraphDB.
Please contact Ting Liu ([email protected]; [email protected]) if you have any questions about the construction process, the datasets and query protocols.
@article{liu2020ketpath, author = {Liu, Ting and Feenstra, K Anton and Huang, Zhisheng and Heringa, Jaap}, title = "{Mining literature, microbiome and pathway data to explore the relations of ketamine with neurotransmitters and gut microbiota using a knowledge-graph}", journal = {}, year = {2023}, volume = {}, number = {}, pages = {}, publisher = {}, doi = {}, }