CASCADE: An Agentic Regulatory Network Framework for Patient-Data-Validated Downstream Perturbation Prediction

CASCADE leverages ARACNe-inferred gene regulatory networks and exposes them through the Model Context Protocol (MCP), enabling agent-based reasoning over regulatory interactions. The framework's validation approach uses focal-gene copy-number amplification as a dosage-based proxy for the inverse of knockdown, checking not only whether predicted genes are cancer-related but also whether the predicted direction of expression change matches expected biological reality. This addresses a common weakness in prior tools that only perform membership-based validation. The paper is arXiv:2608.05359v1, submitted to cs.AI. The work suggests a shift toward more biologically faithful evaluation of perturbation prediction methods, which could improve target discovery and drug response modeling.