EXECUTION-GROUNDED AI SCIENTIST FOR AUTONOMOUS ALZHEIMER'S DRUG TARGET
David Scott Lewis, Enrique Zueco, Haley Yi
AIXC Bio, Zaragoza, Spain | reports@aixcbio.com

PROOF THROUGH EXECUTION TRANSFORMS AI REASONING INTO VERIFIED THERAPEUTIC TARGETS.

INTRODUCTION
Background
Slow Preclinical Alzheimer's Pipeline -> Human-driven Hypothesis Bottleneck -> Text-Only AI Limitations (LLMs) (Plausible but untestable hypotheses) -> Execution-Grounded AI Scientist Solution
- Bottlenecked by pace
- Valid-looking but flawed hypotheses
- Introduces autonomous testing

METHODS
Iterative Lifecycle
AUTONOMOUS AGENT LIFECYCLE
1. FORMULATE HYPOTHESES (amyloid, tau, neuroinflammation)
2. WRITE SIMULATION CODE
3. EXECUTE CODE (in silico molecular docking, pharmacokinetic simulations, Identification of flaws)
4. RECEIVE EXECUTION FEEDBACK (error identification)
5. AUTONOMOUS REFINEMENT
(Iterative self-improvement loop)
- Escalates from generic amyloid proposals, and progress to verified, brain-penetrant multi-target candidates
- Required human biomimetic analysis

RESULTS
Outcomes
PREDICTED BINDING AFFINITY | BBB PERMEABILITY
(Bar chart and heatmap showing Execution-Grounded AI vs Baseline Text-Only AI)
- Convergence to rigorously self-refined hypotheses
- Three multi-target candidates with verified brain-penetrant properties
- Multi-target candidates, verified significantly higher predicted affinity
- Rigorous code execution & error-driven discovery, leading to verified results

CONCLUSIONS
Implications
PARADIGM SHIFT
Text-based reasoning to Verifiable Experimental Validation
- Produces more rigorous, reproducible targets
ACCELERATES AUTONOMOUS ALZHEIMER'S DRUG DISCOVERY + a clear path forward
Quality control -> Reproducibility -> Growth