'Self-Improving Discovery Agents in AI-Driven Neurodegenerative Disease'
David Scott Lewis, Enrique Zueco, Haley Yi
AIXC Bio, Zaragoza, Spain
reports@aixcbio.com

Key Takeaway: Self-improving AI agents enforce rigorous, evidence-driven refinement to accelerate reproducible Alzheimer's drug discovery.

INTRODUCTION
The sheer volume of multi-omic data, clinical trial readouts, and published literature strains human capacity for systematic integration.
LLMs Hallucinating Citations
Incompetent at retrieving all multi-omic data
Misaligned Hypotheses
What's irrelevant in AD research?

METHODS
Autonomous Discovery Agent system enforces Closed Loop:
Hypothesis Generation -> External Knowledge Retrieval Systems (PubMed, Semantic Scholar, Domain-Specific AD Databases)
LLM-Powered Discovery Agent -> Retrieve & Self-Critique
Iterative Refinement of in silico Models
Improved Proposals -> Feedback-Loop -> Self-Correcting Reasoning -> Hypothesis Generation

RESULTS
Autonomous agent performance across 50 AD research challenges
Reduced Hallucination Rate (Bar chart showing Agents, Baseline, LLMs)
High-Value Combination Drug Targets Identified
Validation via in silico Molecular Docking & Literature Cross-Validation

CONCLUSIONS
A paradigm shift from passive literature mining to active hypothesis generation and self-criticism.
Passive Literature Mining Tool -> Active Hypothesis Generator & Self-Critic
- Evidence-Driven Refinement
- Rigorous, evidence-driven refinement accelerates reproducible drug discovery
- Reproducible AD Drug Discovery
- Future potential of self-correcting agents