Window Size or Model Choice? Quantifying the Impact of GWAS Methods on QTN Detection in Maize

Every organism exhibits slight differences in DNA sequences among individuals of the same species, making each one unique. These variations, known as genetic variants, can influence phenotypic traits or susceptibility to diseases. Some genetic variations may provide advantages, such as resistance to pathogens, while others can have detrimental effects, leading to disease. Genome-wide association studies (GWAS) identify associations between single-nucleotide polymorphisms (SNPs) and phenotypic traits, helping to uncover genetic variants that contribute to specific phenotypic expressions. Several statistical models have been developed to enhance the power of GWAS in detecting true genetic associations while minimizing false positives and false negatives. In this study, we compared the power of single-locus and multi-locus statistical models and investigated how varying window sizes around quantitative trait nucleotides (QTNs) affect model performance, using genotypic data from maize (Zea mays L.) and a simulated phenotype. Our results revealed clear differences among models: the General Linear Model (GLM) exhibited severe p-value inflation and excessive false positives, while the Multiple Loci Mixed Model (MLMM) identified the most true QTNs (7/20) while controlling Type I error and false discovery rates. BLINK proved computationally efficient, detecting six true QTNs and outperforming other multi-locus models in runtime. Notably, increasing the window size from 1,000 to 10,000 bp had minimal impact on power, highlighting that method selection—not window size—is critical for accuracy. Thus, MLMM and BLINK offer an optimal balance of statistical rigor and efficiency for maize GWAS, providing actionable insights to enhance trait-marker association studies.