A new advancement in computational oncology has emerged with the introduction of RegNetAgents, an AI-oriented multi-agent framework designed to identify regulatory candidates across complex genomic networks. Operating as a downstream analytical layer rather than a network inference method, the system focuses on structured, query-driven identification of cancer drivers.
Architecture and Workflow
RegNetAgents is implemented as a multi-agent LangGraph Directed Acyclic Graph (DAG) workflow. It is accessible through a unified Python API and a Model Context Protocol (MCP) client, facilitating integration into modern AI research environments. The framework enables the unified analysis of bulk tumor networks derived from TCGA and large-scale single-cell regulatory networks from the GREmLN project.
The framework processes a focal gene through several stages:
- Dual-network classification across heterogeneous data sources.
- Cancer gene filtering utilizing OncoKB annotations.
- Mode-of-action (MoA) assignment for tumor-derived regulatory relationships.
Validation in BRCA and COAD
The system's performance was validated across eleven breast cancer (BRCA) and twelve colorectal cancer (COAD) focal genes. RegNetAgents identified candidate regulators significantly enriched for OncoKB-annotated cancer genes. TCGA-derived candidates showed strong enrichment scores (Stouffer Z = 6.69 for BRCA and 6.95 for COAD), while single-cell GREmLN-derived candidates also demonstrated significant results (Z = 5.51 for BRCA and 7.06 for COAD; all p < 0.0001).
Importantly, the framework showed no enrichment in housekeeping or non-driver control gene sets, confirming the specificity of its analytical signal. An extended module further supports biological hypothesis generation by evaluating oncogenic potential, druggability, clinical relevance, and network vulnerability.
