Graph Pattern Inference for Biomedical Relationship Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current graph data structure approaches for identifying relationships in biomedical data lack specificity and sensitivity, leading to significant false positives and negatives, requiring extensive iterations for result curation.
Innovation Solution
A scaffold query tool that determines a query node in a knowledge graph, identifies target nodes based on connectivity patterns, generates graph-based statistics, and uses machine learning models to assess predicted relationships, improving prediction accuracy by combining hop node categorization and connectivity analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If existing graph analysis approaches are used to identify relationships amongst nodes, then the process can be automated, but the results lack sufficient specificity and sensitivity leading to significant false positives and negatives
Solution Approach 1:
The patent segments the relationship identification process into multiple distinct modules: graph pattern matching module, statistical analysis module, machine learning module, and network propagation module. Each module performs a specific function and their results are integrated to produce the final prediction, thereby maintaining automation while improving prediction accuracy through diversified analysis perspectives
Solution Approach 2:
The patent employs a composite analytical approach by combining multiple different analysis methods (graph patterns, statistics, machine learning, network propagation) into a unified system. This composite strategy leverages the strengths of each individual method while compensating for their weaknesses, achieving both automation and high prediction accuracy
2Productivity
If existing graph analysis approaches are used, then relationships can be predicted, but extensive iterations of curating predicted results are required
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously evaluates prediction results against multiple criteria and automatically refines its outputs. The integrated modules provide mutual validation and correction, reducing the need for manual curation iterations while maintaining high-quality predictions
Solution Approach 2:
The system performs self-curation through automated confidence scoring and result filtering. The machine learning and network propagation modules automatically evaluate and rank predictions, enabling the system to self-optimize without requiring extensive manual intervention, thus improving productivity while minimizing time loss
3Power
If simple connectivity analysis is used, then the process is computationally efficient, but the predictions lack specificity for biological entities
Solution Approach 1:
The patent divides the computational process into segmented stages with increasing complexity: initial graph pattern matching (efficient), followed by statistical filtering, then machine learning scoring, and finally network propagation refinement. This segmentation maintains computational efficiency at early stages while achieving high specificity in later stages
Solution Approach 2:
The patent applies different levels of analytical depth to different parts of the graph structure. Highly connected regions with strong patterns receive more intensive analysis (machine learning and network propagation), while peripheral regions use simpler methods, thereby optimizing computational efficiency while maintaining prediction specificity where it matters most
Data Source
AI summary
A computer-implemented method of querying a graph to assess relationships amongst graph nodes comprises determining a query node on the graph, identifying one or more target nodes on the graph in relation to the query node based on a set of connectivity patterns; generating graph-based statistics for each target node of the one or more target nodes, wherein the graph-based statistics are extracted for subgraphs associated with each target node and the query node; and assessing the graph-based statistics of each target node to determine predicted relationships between the one or more target nodes and the query node.