MedDiscovery: A Multi-Agent Scientific Discovery System for Cross-Domain Therapeutic Hypothesis Generation Multi-Case Validation Across Six Medical Domains

Background: Current AI-assisted drug discovery systems predominantly operate within singledomains, limiting their ability to identify novel therapeutic mechanisms through crossdisciplinary insights. We present MedDiscovery, a multi-agent scientific discovery system thatgenerates testable therapeutic hypotheses by synthesizing knowledge across disparate scientificfields. This work describes the computational methodology and validates the system'sperformance across multiple disease domains.Methods: MedDiscovery employs a tri-core architecture combining BioBERT (knowledgeprocessing), TinyLlama (reasoning/consciousness), and DeepSeek (articulation) withautonomous agent generation, episodic memory, and cross-domain analogical reasoning. Wevalidated the system across six diverse medical conditions: Alzheimer's disease, chronicinflammatory demyelinating polyneuropathy (CIDP), glioblastoma multiforme (GBM), Type 1diabetes (T1D), Tourette syndrome, and idiopathic pulmonary fibrosis (IPF).Results: Across all cases, MedDiscovery generated novel therapeutic hypotheses with meannovelty score of 0.73 (range: 0.45-0.78), synthesizing evidence from 682 sources per case onaverage (total: 4,122 sources). The system achieved 4 Tier-1 (high novelty, strong evidence) and2 Tier-2 (moderate novelty or evidence) classifications. Each hypothesis incorporated crossdomain reasoning from 3-4 scientific fields and included SPIRIT-compliant experimentalprotocols with explicit rejection criteria.Conclusions: MedDiscovery demonstrates that autonomous multi-agent systems can generatescientifically rigorous, experimentally testable therapeutic hypotheses by leveraging crossdomain analogical reasoning. The system's ability to identify non-obvious mechanisticconnections across fields suggests a new paradigm for computational drug discovery thatcomplements rather than replaces human scientific intuition.