Drug Interaction Prediction via Knowledge Graphs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for detecting drug-drug interactions are time-consuming, costly, and require clinical trials or laboratory tests, making it difficult to predict and prevent potentially harmful drug combinations effectively.

Innovation Solution

A computer-implemented method that analyzes medical and biomedical documents to extract relevant information, combines it with structured data, and uses machine learning to create a graph representing potential drug interactions, allowing for automated prediction of drug-drug interactions without the need for extensive clinical trials or laboratory tests.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If automated text mining and machine learning are used to predict drug interactions, then time and resource consumption are reduced, but the precision and reliability of interaction detection may be compromised

Engineering Contradiction:
Improvetime consumptionVSAvoidinteraction detection precision
Core Design Contradiction:
Loss of timeVSMeasurement precision

Solution Approach 1:

The system performs preliminary actions by pre-processing and indexing biomedical literature, constructing knowledge graphs in advance, and pre-training machine learning models on historical drug interaction data. This allows the system to quickly retrieve and analyze relevant information when predicting new drug interactions, significantly reducing prediction time while maintaining precision through pre-established knowledge structures

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent introduces an intermediary knowledge graph that connects drugs, diseases, genes, and proteins through structured relationships. This knowledge graph serves as a mediator between raw biomedical literature and machine learning models, transforming unstructured text into structured, queryable knowledge that improves both processing speed and detection precision by providing contextual relationships

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If comprehensive biomedical literature is analyzed to improve prediction accuracy, then the quality of interaction identification improves, but the complexity and resource requirements of the system increase

Engineering Contradiction:
Improveprediction reliabilityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system segments the complex biomedical knowledge domain into distinct entity types (drugs, diseases, genes, proteins) and relationship types (treats, causes, interacts with). This segmentation allows the knowledge graph to organize vast amounts of biomedical literature into manageable, structured components, improving prediction reliability while controlling system complexity through modular organization

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates a universal knowledge graph framework that can handle multiple types of biomedical entities and relationships simultaneously. This multi-functional system can predict drug-drug interactions, drug-disease associations, and other biomedical relationships using the same infrastructure, reducing overall system complexity while maintaining high prediction reliability across different application scenarios

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Measurement precision

If traditional clinical trials and laboratory tests are used to detect drug interactions, then the accuracy of interaction detection is high, but the cost and time required increase significantly

Engineering Contradiction:
Improveinteraction detection accuracyVSAvoidresource consumption
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The system creates a virtual copy of the biomedical knowledge ecosystem through structured knowledge graphs and simulated drug interaction scenarios. Instead of physically testing all drug combinations through clinical trials, the system copies relevant biological pathways and interaction patterns from existing literature into a computational model, enabling accurate prediction without consuming physical resources for extensive testing

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3859745A1System and method for identifying drug-drug interactions
Publication Date: 2021.08.04 NAT CENT FOR SCI RES DEMOKRITOS
  • EP3859745A1 patent drawingFigure 1
  • EP3859745A1 patent drawingFigure 2
  • EP3859745A1 patent drawingFigure 3

AI summary

Methods and systems for predicting possible interactions between two or more drugs are disclosed. A plurality of documents is analyzed (212, 214) to extract medical relevant information. The extracted information comprises at least one first entity associated with at least one second entity. In addition, supplementary information is received (220) from a structured data source (220-1, 220-2). A graph is then created to represent the information extracted from the plurality of documents and the supplementary information from the structured data source (220-1, 220-2) in a unified scheme. The created graph comprises a plurality of nodes and a plurality of edges connecting the plurality of nodes. After the creation of the graph, the graph is analyzed (234) using machine learning techniques to identify at least one new edge. The identification of the at least one new edge is based on paths already included in the initially created graph. Based on the at least one new edge, probable drug-to-drug interactions are determined (236) and output.