Epigenetics

Testing SNP association methods with à¢â‚¬Ëœopen black box' simulated gene interaction network

---

Abstract

How reliable are QTL detection and GWAS methods? Could hundreds of interacting QTL masquerade as thousands under standard analyses? Are simulation studies overly optimistic, using simple additive models for both data generation and analysis? Answering these critical questions requires test datasets where all epistatic and pleiotropic interactions are known-"open black box" (OBB) datasets. Such OBB datasets do not currently exist. Selective Growth Adapted NK models offer a powerful solution. NK models (Kauffman & Levin, 1987) connect genotypes to fitness through tunable epistasis: each locus interacts with K others (neighboring or random), with total fitness as the sum of locus contributions. The parameter K controls fitness landscape complexity-low K yields smooth landscapes, high K creates extreme ruggedness. However, at biologically realistic epistasis levels (K≥4-8), NK landscapes become chaotic, making adaptive evolution impossible. Selective Growth Adapted NK models (Kinghorn & Tanner, 2017) overcome this limitation. They sequentially build beneficial phenotypic contributor patterns-groups of epistatically interacting loci-around a prototypic genotype. Random interaction patterns are only retained if they increase fitness of a lead candidate, mimicking natural selection. This process creates complex, evolvable gene networks where all interactions are explicitly known, providing perfect OBB testbeds. We applied these models (K≤8, gamma-distributed fitness contributions: shape=0.4, scale=1/1.66) to generate synthetic datasets: 1,000 SNPs total, with 100 contributing to phenotype via epistasis and 900 neutral. Standard candidate SNP association analyses were run across sample sizes of 1,000, 10,000, and 100,000 individuals. Detection rates scaled with sample size-1% (±0.0%) for n=1,000, 25% (±5.1%) for n=10,000, 64% (±5.9%) for n=100,000-but 46% false negatives persisted even at n=100,000. This reveals profound limitations of current additive models. Standard GWAS/QTL methods substantially underestimate causal variants in epistatic networks. These OBB datasets provide a rigorous framework to develop next-generation methods capturing non-additive biology, essential for linking genotypes to complex phenotypes effectively.

Keywords: 2026

How to Cite:

Kinghorn, B., (2026) “Testing SNP association methods with à¢â‚¬Ëœopen black box' simulated gene interaction network”, World Congress on Genetics Applied to Livestock Production Digital Archive 2026(1): 2343480. doi: https://doi.org/10.31274/wcgalp.24484

Rights: 1