Understanding Knowledge Graphs
With the exponential growth of data in recent times, data storage and management is becoming a rising concern. Knowledge graphs come in handy while dealing with such humongous volumes of data. A knowledge graph is a database which stores raw, semi−processed or processed data in a graphical format. By forging relationships between multiple data points, it enables effective organization, analysis, and interpretation of information.
A knowledge graph is a model of information created by an expert with the help of intelligent Machine Learning (ML) algorithms. It helps structure and organize information in a manner that shrewd multilateral relations can be established all through the database. Fully structured as an extra virtual information layer, knowledge graph lies above the current database or informational index, connecting every bit of information at scale.
Knowledge graphs form the backbone of most analytical systems. Intelligent chatbots, recommendation systems and search engines leverage concepts from graphs to interpret results. The data in knowledge graphs captures relationship (edge) between two data points (nodes) which are collected from various heterogeneous sources. Once a graph is developed by establishing a relationship between all the data points, high level analytical algorithms are used to derive insights. Due to their interactive and exploratory nature, knowledge graphs are preferred over traditional methods of data query.
Knowledge Graphs in Healthcare
The healthcare and life sciences industry have overwhelming volumes of unstructured, heterogenous data which can be effectively analyzed using knowledge graphs.
Knowledge graphs in healthcare enhances understanding of patients, diseases, and medicines, enabling best possible diagnosis, which earlier models failed to deliver. While life sciences industry has witnessed growth and innovation, it is still plagued with manual processes that impact the overall development process. For instance, a doctor must manually go through a patient’s medical history to recommend medicine, write summary notes, correlate lab measurements, and compare it with similar cases etc. All these activities demand time and effort that make the system slow.
EHR addressed this challenge to some extent as data was converted into electronic readable format. While it improved speed and accuracy, there was still a large amount of unstructured data which was difficult to process and interpret. This is precisely where machine learning and knowledge graphs come into play in healthcare. They can accurately process both structured data as well as unstructured data at a faster pace. This enables doctors to make timely, effective decisions and reduces overall cost for all the stakeholders involved.
Advantages of knowledge graphs
Object Oriented Thinking
Knowledge graph offers complete visibility into data. While using traditional methods, information about complete architecture of tables is necessary to reach from one data point to another but with knowledge graphs it’s possible to jump from one node to another without considering the relationships between any two nodes.
Graph is basically an index data structure with no pre−characterized constrain such as a relational model with information in a tabular format. Graph stores information in its regular relationship, the ideal method to visualize data, as related information is constantly connected.
It’s much easier to add new data to a knowledge graph as compared to a big data management system. For instance, in a Hadoop HDFS framework, adding new data demands a detailed scan of the existing data and deep understanding of the architecture, which is not necessary in the case of a knowledge graph.
Applications of healthcare knowledge graphs
Knowledge Graph in Electronic Health Records (EHR)
When a patient is admitted to a hospital, the event is termed as ‘admissions’ and the entry is recorded in the hospital database. It is defined by parameters such as the date and time of admission. The duration of this admission depends upon the criticality of the disease against which the patient is admitted. The data of admission also depends upon variables such as type of diagnosis, medications recommended during the ongoing admission process and pre-diagnosis already given to the patients.
A pre-designed knowledge graph in the database consists of details about diseases, its symptoms, diagnosis and medications. For example, migraine is directly associated with medical symptom of headache and drug aspirin. It’s easy to form a visualization for nodes like symptoms, diseases or admissions. The relationship established between the nodes could be the relation between the current admission and the pre-diagnosis steps from existing knowledge graph. Thus, a knowledge graph is formed by combining new data with old data.
Modeling polypharmacy side effects
The concurrent use of multiple medications by a patient, also called polypharmacy, is commonly used to treat complex diseases. However, polypharmacy is associated with a high risk of adverse effects. Polypharmacy reactions develop due to drug−drug interaction, a scenario where behavior of one medication may change, positively or negatively, when taken with another drug. Understanding of drug cascading is often constrained as unpredictable connections are uncommon, and are typically not observed in general clinical testing.
Finding polypharmacy reactions with the help of knowledge graphs opens new doors with huge ramifications for patient mortality and improved health outcomes.