ALZHEIMER'S ASSOCIATION
AAIC 26

CO-EVOLVING VIRTUAL CELL MODELS AND PERTURBATION PLANNERS FOR ALZHEIMER'S DISEASE DRUG DISCOVERY
DAVID SCOTT LEWIS, ENRIQUE ZUECO, HALEY YI | AIXC Bio, Zaragoza, Spain | reports@aixcbio.com

KEY TAKEAWAY: Co-evolutionary AI enables the discovery of complex, multi-target combination therapies in silico for AD reversal, surpassing single-target strategies and refining cellular disease simulators.

INTRODUCTION
Current in silico models:
Static, lacking interventions
Static -> Simulates

Unlinked AD complexities:
Amyloid plaques
Tau tangles
Neuroinflam. (glial cells)

-> Optimized multi-target perturbation strategies for AD state reversal.
VCM+Planner

METHODS
Virtual Cell Model (VCM) (Multimodal Latent Diffusion)
Transcriptomic ->
Proteomic ->
Epigenomic ->

Simulate cell state transitions -> Perturbation Planner (RL-Based)
Propose Genetic/Pharmacologic combinations ->
Drive VCM generalization -> Co-Evolution Cycle
Refine proposals -> (Back to VCM)

Validation for synthetic single-cell AD datasets modeling DAM transitions

RESULTS
AD State Transition Reversal
(e.g., DAM to Homeostatic microglia)
Trajectory 22 vs Trajectory 1
DAM -> DAM -> Homeostatic

Pathological Attractor Reversal Score
(outperformed single targets by 38%)
Reversal Score
Synergistic combinations outperformed by 38%
Combinations

Predicted Transcriptomic Recovery Score
Recovery
Combinations

CONCLUSIONS
Powerful New Paradigm for in silico AD Drug Discovery.
Iterative Refinement (Co-Evolution) of Cell Simulator & Planner.
Enables Discovery of Complex Combination Therapies.
Validate before Costly In Vitro or In Vivo Trials.