GiG: Knowledge Graph Deep Learning for Clinical Data

Researchers introduce Graph-in-Graph (GiG), a deep learning framework that uses knowledge graphs to improve clinical prediction from limited patient data. GiG represents each patient as a modular graph where curated biological knowledge defines edges and patient-specific measurements define node features. This preserves gene-gene interactions and integrates multiple knowledge sources. The approach addresses the limitation of traditional AI models that compress graph-encoded biological knowledge into low-dimensional representations, losing structure and performance in small-sample studies. GiG is designed for data-efficient clinical analysis, leveraging structured molecular interaction knowledge.