71. SELECTION OF DONORS FOR CELLULAR MODELS OF POLYGENIC DISORDERS: A POWER ANALYSIS FRAMEWORK

Background Most risk for psychiatric disorders comes from the polygenic contributions of an individual’s genetic background, however, most cellular model studies target rare, major genes that are only responsible for a vast minority of cases. There is an increasing interest in conducting cellular model studies that capture polygenic effects, but no formal guidance for designing and powering such studies exists. How should individuals be selected for cell line derivation? Researchers may select based on case-control status, polygenic score (PGS) extremes, family history, or combinations of these genetically informative indices. In this work, we consider the implications, efficiency, and power of various selection approaches, finding that optimal selection depends on disease architecture, its relationship to cellular phenotypes, and the available recruitment pool. In doing so, we provide a formal framework to evaluate design and recruitment strategies. Methods We developed an analytical power framework for two-group comparisons using the liability threshold model. An intermediate cellular phenotype (PCell) was modelled as (i) a linear, (ii) a binary thresholded, or (iii) non-linear function of genetic liability. We derived closed-form power expressions for case-control, PGS-extreme, family history-based, and hybrid selection as a function of PGS accuracy (R²l), heritability (h²l), prevalence, and recruitment pool size (Nbiobank). Analytical results were validated by simulation. Results Power depends critically on the genetic value difference (ΔG) between groups. For rare, highly heritable disorders like schizophrenia (SCZ; population prevalence ∼1%, h² ∼0.70), case-control selection captures a large ΔG, whereas PGS-based selection at current accuracy (R 2 1 ∼4–5%) would require taking top and bottom 20 individuals from a pool of 1.0-1.5 million individuals for equivalent ΔG. For common, moderately heritable disorders like major depressive disorder (MDD; prevalence ∼15%, h² ∼0.35), case-control selection has substantially lower power, and PGS-extreme selection (R 2 1 ∼4%) achieves equivalent ΔG with pools of only ∼150 individuals, and superior power with any larger pool. Hybrid strategies offered disorder-specific gains: for SCZ, retaining all cases while selecting PGS-based low-liability controls increased ΔG 1.4-fold over standard case-control; for MDD, double conditioning — selecting high-PGS cases against low-PGS controls — increased ΔG 2.8-fold (both at N_bb = 100,000, n = 20 per group). Under threshold and non-linear genotype-phenotype maps, the advantage of PGS-extreme selection may be smaller. Simulations confirmed the analytical approximations. Discussion There is an increasing interest in designing cellular experiments to detect molecular phenotypes associated with aggregate genetic liability, however, to-date there have been few to any formal frameworks for powering these experiments. We provide a simple power framework that provides insights into the design of experiments investigating polygenic effects on cellular phenotypes.