ALZHEIMER'S ASSOCIATION
AAIC '26

DYNAMIC CAUSAL DISCOVERY: OPEN-ENDED MAPPING OF NEUROINFLAMMATORY PATHWAYS
David Scott Lewis, Enrique Zueco, Haley Yi
AIXC Bio, Zaragoza, Spain. reports@aixcbio.com

KEY TAKEAWAY: Self-improving AI framework accelerates complex Alzheimer's pathway mapping by recursively evolving scientific principles.

INTRODUCTION
OPEN-ENDED PROBLEM:
Neuroinflammation in AD is complex.

LIMITATIONS OF STATIC AI:
Evolutionary algorithms lack guided exploration, limiting complex hypothesis discovery.

METHODS
REFLECTIVE CAUSAL SCRATCHPAD
Transcriptomic Datasets of Disease-Associated Microglia -> AI Agent -> Bayesian Optimization over an Expanding Principle Space -> Abductive Reasoning for Hidden Variables -> Recursive Graph Modification with Stability Testing -> RECURSIVE BELIEF EVOLUTION (Loop back to AI Agent / Bayesian Optimization based on execution results)

RESULTS
MOMENTUM-DRIVEN EVOLUTION (OUR METHOD) vs STATIC-HYPOTHESIS AI
(Line graph showing Discovery efficiency for Momentum-driven evolution vs Bar chart for Static-hypothesis AI showing 30% SAMPLE COMPLEXITY REDUCTION)

Higher Efficacy
Stability Summary:
Significantly higher momentum-driven stability and overall discovery efficiency.

NON-CANONICAL REGULATORY PATHWAY
LIPID METABOLISM -> NOVEL STABLE REGULATORY NODE -> INFLAMMATORY CYTOKINE RELEASE

CONCLUSIONS
Shift from fixed hypotheses to dynamically evolving principles:
- DRIVES HIGHLY EFFICIENT, AUTONOMOUS DISCOVERY OF AD PATHOLOGY
- BYPASSES LOCAL OPTIMA TO FIND NOVEL PATHWAYS