Gen2PhenSim: a configurable genotype to phenotype simulator

Abstract

Genotype-phenotype studies often lack known ground truth, making it difficult to evaluate whether statistical or machine-learning methods recover the true causal architecture of a trait. In this thesis, an existing genotype-phenotype simulation codebase was extended and refined into Gen2PhenSim, a configurable Python-based simulator that generates binary and quantitative phenotypes from real genotype data under user-defined genetic architectures. The simulator supports additive SNP effects, dominant and recessive encodings, second- and third-order interaction terms, Gaussian noise, and quantitative-trait heritability control. Besides, it has a feature to control for selecting SNPs based on their distance to select independent SNPs. As a proof of concept, simulated phenotypes were analysed with Regenie and GWAS hits were compared with the true simulated causal SNPs.