aiAtlas: High Fidelity Cell Simulations of Genetic Perturbations in Rare Diseases and Cancers

Background: Understanding how genetic perturbations drive cellular phenotypes remains a central challenge in disease modeling. While induced pluripotent stem cells (iPSCs) and tumor-derived cell lines have advanced mechanistic research, they are constrained by scalability, stability, and limited capacity to reproduce complex mutational states. Methods: We developed aiAtlas, a reliable computational system that integrates Large Concept Model (LCM) based causal inference with aiPSC derived simulation libraries. The platform encodes 25 core biological and cellular features, including apoptosis, DNA damage and repair pathways, cell cycle regulation, pluripotency markers, and the hallmarks of cancer, in weighted causal networks. Simulations were iteratively propagated using early-stopping algorithms, with statistical evaluation performed using Mann Whitney U tests, Hodges Lehmann estimators (HLE), and Cliff's delta with Hedges correction. Bonferroni-adjusted significance thresholds (p < 0.002) to compensate for multiple testing were applied. Results: Across 136 simulated aiPSC lines (10 wild type, 126 mutants), all 25 features demonstrated significant divergence between wild type and combined mutant states (median HLE -1.46, IQR -1.89 to -0.36). Subgroup analyses revealed distinct patterns: single mutations drove broad systemic changes; multiple mutations amplified genomic instability; human tumor-derived cell lines exhibited strong divergence but retained some overlap in repair and apoptosis pathways; and gene fusion models produced selective, intermediate phenotypes. Effect sizes were consistently large (Cliff's delta approaching -0.95 with 95% CIs). Conclusions: aiAtlas reliably distinguishes wild type from mutant cell states, capturing both global and subgroup-specific mutational divergence. By combining causal inference with LCM-level abstraction, aiAtlas establishes a scalable, reproducible, and physiologically grounded platform for rare disease modeling, cancer biology, and therapeutic discovery.