Swarm Intelligence for Drug Indication: The CellAwareGNN Multi-Agent Architecture

ABSTRACT

Predicting novel drug indications requires synthesizing vast, heterogeneous networks of gene, disease, and chemical data. We upgrade the CellAwareGNN foundation model into a multi-agent swarm architecture. Specialized agents are deployed to traverse different sub-graphs of a single-cell enhanced knowledge network, allowing for highly parallelized exploration of complex pharmacological relationships. An 'Inference Agent' aggregates these diverse sub-graph traversals to propose highly probable drug repositioning candidates. This swarm-based approach captures heterogeneous biological signals that standard GNNs miss, accelerating the identification of off-target therapies for rare diseases.