Structural Learning in GNNS for Medical Decision-Making Using Task-Related Graph Improvements
The method enhances GNN performance by refining graph structures using trainable control parameters and variational optimization, improving predictive accuracy and interpretability for tasks like spatial transcriptomics and medical decision-making.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- NEC LABORATORIES AMERICA INC
- Filing Date
- 2024-05-16
- Publication Date
- 2026-06-16
AI Technical Summary
Graph neural networks (GNNs) face performance degradation due to irrelevant edges between non-interacting components or missing edges between interacting components, affecting tasks like spatial gene expression prediction and protein structure analysis.
A method for graph analysis that involves identifying trainable control parameters, generating sample graph improvements, and selecting parameters for graph neural networks to refine the graph structure, using variational optimization and smoothing-based optimization to enhance performance.
Improves the predictive accuracy and interpretability of GNNs by refining graph structures, enabling better prediction performance and providing valuable biological insights, applicable to tasks such as spatial transcriptomics and medical decision-making.
Smart Images
Figure 2026519466000001_ABST