A Knowledge Graph Platform, is by definition helps you build and operate knowledge graphs from disparate structured and unstructured datasets.
First and foremost the Platform should be useful to its users: Life Scientists, Data Scientists, Data Analysts, and R&D leaders. Knowledge Graph Platforms (KGPs) have a range of functionalities, but these seven features are must-haves for any research organization that is aiming to improve their discovery process and accelerate their R&D.
Integration with Public Datasets
One thing that Data engineers would immediately realize when trying to build Knowledge graphs from scratch is that there is a ton of data already available out there. So it is only reasonable to assume that any Knowledge Graph Platform worth its salt must support integrating with said public data sources already and make it easy to be able to do it. Here are some of the Public Data sources you should expect your Knowledge Graph Platform to support out of the box:
One thing that Data engineers would immediately realize when trying to build Knowledge graphs from scratch is that there is a ton of data already available out there. So it is only reasonable to assume that any Knowledge Graph Platform worth its salt must support integrating with said public data sources already and make it easy to be able to do it. Here are some of the Public Data sources you should expect your Knowledge Graph Platform to support out of the box:
- NCBI Gene
- Ensembl
- UniProt
- InterPro
- Protein Data Bank (PDB)
- CIViC DB
- PubChem
- RxNorm
- NDF-RT
- LIPID MAPS
- MassBank
- PDB Ligand
- CompTox Chemicals Dashboard
- Reactome Pathways Database
- WikiPathways
- IUPHAR Pharmacology Guide
- Human Disease Ontology
- Monarch Disease Ontology
- MEDLINE Abstracts
- NIH Grants
- ClinicalTrials.gov
- Cancer Care Treatment Outcome Ontology (CCTO)
Flexible Querying
Once you have integrated the datasets you require, the first big use case for knowledge graphs would be to query them in different ways. These include the following
- Simple keyword based search Being able to quickly lookup entities in the graph using a keyword based search is vital.
- Semantic Search: One step above keyword search is able to lookup specific concepts not just keywords for e.g. Lookup the Disease 'Diabetes' should return all related keywords to the disease without having to specify them individually.
- Multi-Hop Querying: For any realistic use-case that is involved in drug discover, its is more important to be able to join multiple datasets together to answer the query effectively for e.g. The following query:
- Clinical Trials since January 2020 about compounds that bind EGFR
- Involves joining ClinicalTrials database with the Compounds database and the Proteins Database
- Being able to easily specify a query as complicated as this visually is critical
Visualizations
Flexible querying is the first stage of accessing the Knowledge Graph. The other side of this coin for Scientists is to be able to visualize the results of the queries in different ways to be able to uncover hidden insights. Here are some of the different forms of visualizations you should expect from your knowledge graph platform to be useful to various stakeholders:
Graph Exploration and Browsing
Custom Insights and Visualizations
Open and Extensible Pipelines
Having your knowledge graph built with the public and private datasets and quickly accessible is important. However often times there will be reasons to extend the knowledge graph pipeline to customize it for your specific use-case. This is where an open and extensible pipeline is crucial for this purpose. Here is an example of one such extensible and open pipeline implemented in the Wisecube Knowledge Graph Platform:
Advanced NLP and Analytics
State of the art NLP
Building Knowledge Graphs involves parsing both structured and unstructured data. For unstructured data like Pubmed articles it is imperative you have the very best NLP engines to be do it effectively. Here is an example of some of the capabilities you should expect in the Natural Language Processing Engine in your Knowledge Graph Platform.
Advanced Graph Analytics
The benefit of building a knowledge graph that combines public and private datasets is that even if you do not have a lot of labeled data in domains like Drug discovery, you can still leverage techniques like Link-Prediction to apply transfer learning to still come up with very reliable predictions.
Scalability
Most modern knowledge graph platforms require you to maintain a separate database which severely complicates your overall data architecture and also limits the scalability of your knowledge graph. Make sure that you knowledge graph platform is scalable as your overall Data-lake platform. Ideally, you want your knowledge graph platform to be built on top of your Data-lake platform so you do not have to manage, secure and scale it separately.
Data Privacy and Security
When dealing with Enterprise Knowledge Graphs, Privacy and Security becomes table-stakes. For this purpose, ensure that the knowledge graph platform can be deployed behind your firewall and meets your stringent Privacy and Security requirements before you go down the path of selecting it. At a minimum, the platform should be able to be completely deployed within your Virtual Private Cloud (VPC) to avoid any transfer of any proprietary data outside of your control.
Getting the right Help
Ok, there are actually 8 things, but hopefully, you now have a better idea of some of the important considerations when selecting a knowledge graph platform to help build your next knowledge graph for your Enterprise. If you need help regarding your journey to build your own knowledge graph, drop us a line at Wisecube.ai and we would be glad to help you out.